rdelaage
/
postgresql
mirror of https://git.postgresql.org/git/postgresql.git
1
0
Fork 0
Code Issues Packages Projects Releases Wiki Activity
postgresql/src/test/isolation/isolationtester.c

Ignoring revisions in .git-blame-ignore-revs. Click here to bypass and see the normal blame view.

1126 lines
29 KiB
C
Raw Normal View History

Implement genuine serializable isolation level. Until now, our Serializable mode has in fact been what's called Snapshot Isolation, which allows some anomalies that could not occur in any serialized ordering of the transactions. This patch fixes that using a method called Serializable Snapshot Isolation, based on research papers by Michael J. Cahill (see README-SSI for full references). In Serializable Snapshot Isolation, transactions run like they do in Snapshot Isolation, but a predicate lock manager observes the reads and writes performed and aborts transactions if it detects that an anomaly might occur. This method produces some false positives, ie. it sometimes aborts transactions even though there is no anomaly. To track reads we implement predicate locking, see storage/lmgr/predicate.c. Whenever a tuple is read, a predicate lock is acquired on the tuple. Shared memory is finite, so when a transaction takes many tuple-level locks on a page, the locks are promoted to a single page-level lock, and further to a single relation level lock if necessary. To lock key values with no matching tuple, a sequential scan always takes a relation-level lock, and an index scan acquires a page-level lock that covers the search key, whether or not there are any matching keys at the moment. A predicate lock doesn't conflict with any regular locks or with another predicate locks in the normal sense. They're only used by the predicate lock manager to detect the danger of anomalies. Only serializable transactions participate in predicate locking, so there should be no extra overhead for for other transactions. Predicate locks can't be released at commit, but must be remembered until all the transactions that overlapped with it have completed. That means that we need to remember an unbounded amount of predicate locks, so we apply a lossy but conservative method of tracking locks for committed transactions. If we run short of shared memory, we overflow to a new "pg_serial" SLRU pool. We don't currently allow Serializable transactions in Hot Standby mode. That would be hard, because even read-only transactions can cause anomalies that wouldn't otherwise occur. Serializable isolation mode now means the new fully serializable level. Repeatable Read gives you the old Snapshot Isolation level that we have always had. Kevin Grittner and Dan Ports, reviewed by Jeff Davis, Heikki Linnakangas and Anssi Kääriäinen
2011-02-07 22:46:51 +01:00
/*
* src/test/isolation/isolationtester.c
*
* isolationtester.c
* Runs an isolation test specified by a spec file.
*/
Fix some portability issues in isolation regression test driver. Remove random system #includes in favor of using postgres_fe.h. (The alternative to that is letting this module grow its own configuration testing ability...) Also fix the "make clean" target to actually clean things up. Per local testing.
2011-05-09 01:44:12 +02:00
#include "postgres_fe.h"
Add support for blocked commands in isolationtester This enables us to test that blocking commands (such as foreign keys checks that conflict with some other lock) act as intended. The set of tests that this adds is pretty minimal, but can easily be extended by adding new specs. The intention is that this will serve as a basis for ensuring that further tweaks of locking implementation preserve (or improve) existing behavior. Author: Noah Misch
2011-07-12 22:35:09 +02:00
#include <sys/time.h>
#ifdef HAVE_SYS_SELECT_H
#include <sys/select.h>
#endif
Modify the isolation tester so that multiple sessions can wait. This allows testing of deadlock scenarios. Scenarios that would previously have been considered invalid are now simply taken as a scenario in which more than one backend will wait.
2016-02-11 14:36:30 +01:00
#include "datatype/timestamp.h"
Make the order of the header file includes consistent in non-backend modules. Similar to commit 7e735035f2, this commit makes the order of header file inclusion consistent for non-backend modules. In passing, fix the case where we were using angle brackets (<>) for the local module includes instead of quotes (""). Author: Vignesh C Reviewed-by: Amit Kapila Discussion: https://postgr.es/m/CALDaNm2Sznv8RR6Ex-iJO6xAdsxgWhCoETkaYX=+9DW3q0QCfA@mail.gmail.com
2019-10-23 06:08:53 +02:00
#include "isolationtester.h"
Implement genuine serializable isolation level. Until now, our Serializable mode has in fact been what's called Snapshot Isolation, which allows some anomalies that could not occur in any serialized ordering of the transactions. This patch fixes that using a method called Serializable Snapshot Isolation, based on research papers by Michael J. Cahill (see README-SSI for full references). In Serializable Snapshot Isolation, transactions run like they do in Snapshot Isolation, but a predicate lock manager observes the reads and writes performed and aborts transactions if it detects that an anomaly might occur. This method produces some false positives, ie. it sometimes aborts transactions even though there is no anomaly. To track reads we implement predicate locking, see storage/lmgr/predicate.c. Whenever a tuple is read, a predicate lock is acquired on the tuple. Shared memory is finite, so when a transaction takes many tuple-level locks on a page, the locks are promoted to a single page-level lock, and further to a single relation level lock if necessary. To lock key values with no matching tuple, a sequential scan always takes a relation-level lock, and an index scan acquires a page-level lock that covers the search key, whether or not there are any matching keys at the moment. A predicate lock doesn't conflict with any regular locks or with another predicate locks in the normal sense. They're only used by the predicate lock manager to detect the danger of anomalies. Only serializable transactions participate in predicate locking, so there should be no extra overhead for for other transactions. Predicate locks can't be released at commit, but must be remembered until all the transactions that overlapped with it have completed. That means that we need to remember an unbounded amount of predicate locks, so we apply a lossy but conservative method of tracking locks for committed transactions. If we run short of shared memory, we overflow to a new "pg_serial" SLRU pool. We don't currently allow Serializable transactions in Hot Standby mode. That would be hard, because even read-only transactions can cause anomalies that wouldn't otherwise occur. Serializable isolation mode now means the new fully serializable level. Repeatable Read gives you the old Snapshot Isolation level that we have always had. Kevin Grittner and Dan Ports, reviewed by Jeff Davis, Heikki Linnakangas and Anssi Kääriäinen
2011-02-07 22:46:51 +01:00
#include "libpq-fe.h"
Centralize getopt-related declarations in a new header file pg_getopt.h. We used to have externs for getopt() and its API variables scattered all over the place. Now that we find we're going to need to tweak the variable declarations for Cygwin, it seems like a good idea to have just one place to tweak. In this commit, the variables are declared "#ifndef HAVE_GETOPT_H". That may or may not work everywhere, but we'll soon find out. Andres Freund
2014-02-15 20:31:30 +01:00
#include "pg_getopt.h"
Make the order of the header file includes consistent in non-backend modules. Similar to commit 7e735035f2, this commit makes the order of header file inclusion consistent for non-backend modules. In passing, fix the case where we were using angle brackets (<>) for the local module includes instead of quotes (""). Author: Vignesh C Reviewed-by: Amit Kapila Discussion: https://postgr.es/m/CALDaNm2Sznv8RR6Ex-iJO6xAdsxgWhCoETkaYX=+9DW3q0QCfA@mail.gmail.com
2019-10-23 06:08:53 +02:00
#include "pqexpbuffer.h"
Implement genuine serializable isolation level. Until now, our Serializable mode has in fact been what's called Snapshot Isolation, which allows some anomalies that could not occur in any serialized ordering of the transactions. This patch fixes that using a method called Serializable Snapshot Isolation, based on research papers by Michael J. Cahill (see README-SSI for full references). In Serializable Snapshot Isolation, transactions run like they do in Snapshot Isolation, but a predicate lock manager observes the reads and writes performed and aborts transactions if it detects that an anomaly might occur. This method produces some false positives, ie. it sometimes aborts transactions even though there is no anomaly. To track reads we implement predicate locking, see storage/lmgr/predicate.c. Whenever a tuple is read, a predicate lock is acquired on the tuple. Shared memory is finite, so when a transaction takes many tuple-level locks on a page, the locks are promoted to a single page-level lock, and further to a single relation level lock if necessary. To lock key values with no matching tuple, a sequential scan always takes a relation-level lock, and an index scan acquires a page-level lock that covers the search key, whether or not there are any matching keys at the moment. A predicate lock doesn't conflict with any regular locks or with another predicate locks in the normal sense. They're only used by the predicate lock manager to detect the danger of anomalies. Only serializable transactions participate in predicate locking, so there should be no extra overhead for for other transactions. Predicate locks can't be released at commit, but must be remembered until all the transactions that overlapped with it have completed. That means that we need to remember an unbounded amount of predicate locks, so we apply a lossy but conservative method of tracking locks for committed transactions. If we run short of shared memory, we overflow to a new "pg_serial" SLRU pool. We don't currently allow Serializable transactions in Hot Standby mode. That would be hard, because even read-only transactions can cause anomalies that wouldn't otherwise occur. Serializable isolation mode now means the new fully serializable level. Repeatable Read gives you the old Snapshot Isolation level that we have always had. Kevin Grittner and Dan Ports, reviewed by Jeff Davis, Heikki Linnakangas and Anssi Kääriäinen
2011-02-07 22:46:51 +01:00
Add support for blocked commands in isolationtester This enables us to test that blocking commands (such as foreign keys checks that conflict with some other lock) act as intended. The set of tests that this adds is pretty minimal, but can easily be extended by adding new specs. The intention is that this will serve as a basis for ensuring that further tweaks of locking implementation preserve (or improve) existing behavior. Author: Noah Misch
2011-07-12 22:35:09 +02:00
#define PREP_WAITING "isolationtester_waiting"
/*
* conns[0] is the global setup, teardown, and watchdog connection. Additional
Use annotations to reduce instability of isolation-test results. We've long contended with isolation test results that aren't entirely stable. Some test scripts insert long delays to try to force stable results, which is not terribly desirable; but other erratic failure modes remain, causing unrepeatable buildfarm failures. I've spent a fair amount of time trying to solve this by improving the server-side support code, without much success: that way is fundamentally unable to cope with diffs that stem from chance ordering of arrival of messages from different server processes. We can improve matters on the client side, however, by annotating the test scripts themselves to show the desired reporting order of events that might occur in different orders. This patch adds three types of annotations to deal with (a) test steps that might or might not complete their waits before the isolationtester can see them waiting; (b) test steps in different sessions that can legitimately complete in either order; and (c) NOTIFY messages that might arrive before or after the completion of a step in another session. We might need more annotation types later, but this seems to be enough to deal with the instabilities we've seen in the buildfarm. It also lets us get rid of all the long delays that were previously used, cutting more than a minute off the runtime of the isolation tests. Back-patch to all supported branches, because the buildfarm instabilities affect all the branches, and because it seems desirable to keep isolationtester's capabilities the same across all branches to simplify possible future back-patching of tests. Discussion: https://postgr.es/m/327948.1623725828@sss.pgh.pa.us
2021-06-23 03:43:12 +02:00
* connections represent spec-defined sessions.
Add support for blocked commands in isolationtester This enables us to test that blocking commands (such as foreign keys checks that conflict with some other lock) act as intended. The set of tests that this adds is pretty minimal, but can easily be extended by adding new specs. The intention is that this will serve as a basis for ensuring that further tweaks of locking implementation preserve (or improve) existing behavior. Author: Noah Misch
2011-07-12 22:35:09 +02:00
*/
Use annotations to reduce instability of isolation-test results. We've long contended with isolation test results that aren't entirely stable. Some test scripts insert long delays to try to force stable results, which is not terribly desirable; but other erratic failure modes remain, causing unrepeatable buildfarm failures. I've spent a fair amount of time trying to solve this by improving the server-side support code, without much success: that way is fundamentally unable to cope with diffs that stem from chance ordering of arrival of messages from different server processes. We can improve matters on the client side, however, by annotating the test scripts themselves to show the desired reporting order of events that might occur in different orders. This patch adds three types of annotations to deal with (a) test steps that might or might not complete their waits before the isolationtester can see them waiting; (b) test steps in different sessions that can legitimately complete in either order; and (c) NOTIFY messages that might arrive before or after the completion of a step in another session. We might need more annotation types later, but this seems to be enough to deal with the instabilities we've seen in the buildfarm. It also lets us get rid of all the long delays that were previously used, cutting more than a minute off the runtime of the isolation tests. Back-patch to all supported branches, because the buildfarm instabilities affect all the branches, and because it seems desirable to keep isolationtester's capabilities the same across all branches to simplify possible future back-patching of tests. Discussion: https://postgr.es/m/327948.1623725828@sss.pgh.pa.us
2021-06-23 03:43:12 +02:00
typedef struct IsoConnInfo
{
/* The libpq connection object for this connection. */
PGconn *conn;
/* The backend PID, in numeric and string formats. */
int backend_pid;
const char *backend_pid_str;
/* Name of the associated session. */
const char *sessionname;
/* Active step on this connection, or NULL if idle. */
PermutationStep *active_step;
/* Number of NOTICE messages received from connection. */
int total_notices;
} IsoConnInfo;
static IsoConnInfo *conns = NULL;
Implement genuine serializable isolation level. Until now, our Serializable mode has in fact been what's called Snapshot Isolation, which allows some anomalies that could not occur in any serialized ordering of the transactions. This patch fixes that using a method called Serializable Snapshot Isolation, based on research papers by Michael J. Cahill (see README-SSI for full references). In Serializable Snapshot Isolation, transactions run like they do in Snapshot Isolation, but a predicate lock manager observes the reads and writes performed and aborts transactions if it detects that an anomaly might occur. This method produces some false positives, ie. it sometimes aborts transactions even though there is no anomaly. To track reads we implement predicate locking, see storage/lmgr/predicate.c. Whenever a tuple is read, a predicate lock is acquired on the tuple. Shared memory is finite, so when a transaction takes many tuple-level locks on a page, the locks are promoted to a single page-level lock, and further to a single relation level lock if necessary. To lock key values with no matching tuple, a sequential scan always takes a relation-level lock, and an index scan acquires a page-level lock that covers the search key, whether or not there are any matching keys at the moment. A predicate lock doesn't conflict with any regular locks or with another predicate locks in the normal sense. They're only used by the predicate lock manager to detect the danger of anomalies. Only serializable transactions participate in predicate locking, so there should be no extra overhead for for other transactions. Predicate locks can't be released at commit, but must be remembered until all the transactions that overlapped with it have completed. That means that we need to remember an unbounded amount of predicate locks, so we apply a lossy but conservative method of tracking locks for committed transactions. If we run short of shared memory, we overflow to a new "pg_serial" SLRU pool. We don't currently allow Serializable transactions in Hot Standby mode. That would be hard, because even read-only transactions can cause anomalies that wouldn't otherwise occur. Serializable isolation mode now means the new fully serializable level. Repeatable Read gives you the old Snapshot Isolation level that we have always had. Kevin Grittner and Dan Ports, reviewed by Jeff Davis, Heikki Linnakangas and Anssi Kääriäinen
2011-02-07 22:46:51 +01:00
static int nconns = 0;
Use annotations to reduce instability of isolation-test results. We've long contended with isolation test results that aren't entirely stable. Some test scripts insert long delays to try to force stable results, which is not terribly desirable; but other erratic failure modes remain, causing unrepeatable buildfarm failures. I've spent a fair amount of time trying to solve this by improving the server-side support code, without much success: that way is fundamentally unable to cope with diffs that stem from chance ordering of arrival of messages from different server processes. We can improve matters on the client side, however, by annotating the test scripts themselves to show the desired reporting order of events that might occur in different orders. This patch adds three types of annotations to deal with (a) test steps that might or might not complete their waits before the isolationtester can see them waiting; (b) test steps in different sessions that can legitimately complete in either order; and (c) NOTIFY messages that might arrive before or after the completion of a step in another session. We might need more annotation types later, but this seems to be enough to deal with the instabilities we've seen in the buildfarm. It also lets us get rid of all the long delays that were previously used, cutting more than a minute off the runtime of the isolation tests. Back-patch to all supported branches, because the buildfarm instabilities affect all the branches, and because it seems desirable to keep isolationtester's capabilities the same across all branches to simplify possible future back-patching of tests. Discussion: https://postgr.es/m/327948.1623725828@sss.pgh.pa.us
2021-06-23 03:43:12 +02:00
/* Flag indicating some new NOTICE has arrived */
static bool any_new_notice = false;
Improve isolationtester's timeout management. isolationtester.c had a hard-wired limit of 3 minutes per test step. It now emerges that this isn't quite enough for some of the slowest buildfarm animals. This isn't the first time we've had to raise this limit (cf. 1db439ad4), so let's make it configurable. This patch raises the default to 5 minutes, and introduces an environment variable PGISOLATIONTIMEOUT that can be set if more time is needed, following the precedent of PGCTLTIMEOUT. Also, modify isolationtester so that when the timeout is hit, it explicitly reports having sent a cancel. This makes the regression failure log considerably more intelligible. (In the worst case, a timed-out test might actually be reported as "passing" without this extra output, so arguably this is a bug fix in itself.) In passing, update the README file, which had apparently not gotten touched when we added "make check" support here. Back-patch to 9.6; older versions don't have comparable timeout logic. Discussion: https://postgr.es/m/22964.1575842935@sss.pgh.pa.us
2019-12-09 20:31:57 +01:00
/* Maximum time to wait before giving up on a step (in usec) */
static int64 max_step_wait = 300 * USECS_PER_SEC;
Use annotations to reduce instability of isolation-test results. We've long contended with isolation test results that aren't entirely stable. Some test scripts insert long delays to try to force stable results, which is not terribly desirable; but other erratic failure modes remain, causing unrepeatable buildfarm failures. I've spent a fair amount of time trying to solve this by improving the server-side support code, without much success: that way is fundamentally unable to cope with diffs that stem from chance ordering of arrival of messages from different server processes. We can improve matters on the client side, however, by annotating the test scripts themselves to show the desired reporting order of events that might occur in different orders. This patch adds three types of annotations to deal with (a) test steps that might or might not complete their waits before the isolationtester can see them waiting; (b) test steps in different sessions that can legitimately complete in either order; and (c) NOTIFY messages that might arrive before or after the completion of a step in another session. We might need more annotation types later, but this seems to be enough to deal with the instabilities we've seen in the buildfarm. It also lets us get rid of all the long delays that were previously used, cutting more than a minute off the runtime of the isolation tests. Back-patch to all supported branches, because the buildfarm instabilities affect all the branches, and because it seems desirable to keep isolationtester's capabilities the same across all branches to simplify possible future back-patching of tests. Discussion: https://postgr.es/m/327948.1623725828@sss.pgh.pa.us
2021-06-23 03:43:12 +02:00
static void check_testspec(TestSpec *testspec);
Implement a dry-run mode for isolationtester This mode prints out the permutations that would be run by the given spec file, in the same format used by the permutation lines in spec files. This helps in building new spec files. Author: Alexander Shulgin, with some tweaks by me
2011-11-03 18:06:08 +01:00
static void run_testspec(TestSpec *testspec);
Implement genuine serializable isolation level. Until now, our Serializable mode has in fact been what's called Snapshot Isolation, which allows some anomalies that could not occur in any serialized ordering of the transactions. This patch fixes that using a method called Serializable Snapshot Isolation, based on research papers by Michael J. Cahill (see README-SSI for full references). In Serializable Snapshot Isolation, transactions run like they do in Snapshot Isolation, but a predicate lock manager observes the reads and writes performed and aborts transactions if it detects that an anomaly might occur. This method produces some false positives, ie. it sometimes aborts transactions even though there is no anomaly. To track reads we implement predicate locking, see storage/lmgr/predicate.c. Whenever a tuple is read, a predicate lock is acquired on the tuple. Shared memory is finite, so when a transaction takes many tuple-level locks on a page, the locks are promoted to a single page-level lock, and further to a single relation level lock if necessary. To lock key values with no matching tuple, a sequential scan always takes a relation-level lock, and an index scan acquires a page-level lock that covers the search key, whether or not there are any matching keys at the moment. A predicate lock doesn't conflict with any regular locks or with another predicate locks in the normal sense. They're only used by the predicate lock manager to detect the danger of anomalies. Only serializable transactions participate in predicate locking, so there should be no extra overhead for for other transactions. Predicate locks can't be released at commit, but must be remembered until all the transactions that overlapped with it have completed. That means that we need to remember an unbounded amount of predicate locks, so we apply a lossy but conservative method of tracking locks for committed transactions. If we run short of shared memory, we overflow to a new "pg_serial" SLRU pool. We don't currently allow Serializable transactions in Hot Standby mode. That would be hard, because even read-only transactions can cause anomalies that wouldn't otherwise occur. Serializable isolation mode now means the new fully serializable level. Repeatable Read gives you the old Snapshot Isolation level that we have always had. Kevin Grittner and Dan Ports, reviewed by Jeff Davis, Heikki Linnakangas and Anssi Kääriäinen
2011-02-07 22:46:51 +01:00
static void run_all_permutations(TestSpec *testspec);
Remove memory leaks in isolationtester. specscanner.l leaked a kilobyte of memory per token of the spec file. Apparently somebody thought that the introductory code block would be executed once; but it's once per yylex() call. A couple of functions in isolationtester.c leaked small amounts of memory due to not bothering to free one-time allocations. Might as well improve these so that valgrind gives this program a clean bill of health. Also get rid of an ugly static variable. Coverity complained about one of the one-time leaks, which led me to try valgrind'ing isolationtester, which led to discovery of the larger leak.
2021-06-27 18:45:04 +02:00
static void run_all_permutations_recurse(TestSpec *testspec, int *piles,
int nsteps, PermutationStep **steps);
Implement genuine serializable isolation level. Until now, our Serializable mode has in fact been what's called Snapshot Isolation, which allows some anomalies that could not occur in any serialized ordering of the transactions. This patch fixes that using a method called Serializable Snapshot Isolation, based on research papers by Michael J. Cahill (see README-SSI for full references). In Serializable Snapshot Isolation, transactions run like they do in Snapshot Isolation, but a predicate lock manager observes the reads and writes performed and aborts transactions if it detects that an anomaly might occur. This method produces some false positives, ie. it sometimes aborts transactions even though there is no anomaly. To track reads we implement predicate locking, see storage/lmgr/predicate.c. Whenever a tuple is read, a predicate lock is acquired on the tuple. Shared memory is finite, so when a transaction takes many tuple-level locks on a page, the locks are promoted to a single page-level lock, and further to a single relation level lock if necessary. To lock key values with no matching tuple, a sequential scan always takes a relation-level lock, and an index scan acquires a page-level lock that covers the search key, whether or not there are any matching keys at the moment. A predicate lock doesn't conflict with any regular locks or with another predicate locks in the normal sense. They're only used by the predicate lock manager to detect the danger of anomalies. Only serializable transactions participate in predicate locking, so there should be no extra overhead for for other transactions. Predicate locks can't be released at commit, but must be remembered until all the transactions that overlapped with it have completed. That means that we need to remember an unbounded amount of predicate locks, so we apply a lossy but conservative method of tracking locks for committed transactions. If we run short of shared memory, we overflow to a new "pg_serial" SLRU pool. We don't currently allow Serializable transactions in Hot Standby mode. That would be hard, because even read-only transactions can cause anomalies that wouldn't otherwise occur. Serializable isolation mode now means the new fully serializable level. Repeatable Read gives you the old Snapshot Isolation level that we have always had. Kevin Grittner and Dan Ports, reviewed by Jeff Davis, Heikki Linnakangas and Anssi Kääriäinen
2011-02-07 22:46:51 +01:00
static void run_named_permutations(TestSpec *testspec);
Use annotations to reduce instability of isolation-test results. We've long contended with isolation test results that aren't entirely stable. Some test scripts insert long delays to try to force stable results, which is not terribly desirable; but other erratic failure modes remain, causing unrepeatable buildfarm failures. I've spent a fair amount of time trying to solve this by improving the server-side support code, without much success: that way is fundamentally unable to cope with diffs that stem from chance ordering of arrival of messages from different server processes. We can improve matters on the client side, however, by annotating the test scripts themselves to show the desired reporting order of events that might occur in different orders. This patch adds three types of annotations to deal with (a) test steps that might or might not complete their waits before the isolationtester can see them waiting; (b) test steps in different sessions that can legitimately complete in either order; and (c) NOTIFY messages that might arrive before or after the completion of a step in another session. We might need more annotation types later, but this seems to be enough to deal with the instabilities we've seen in the buildfarm. It also lets us get rid of all the long delays that were previously used, cutting more than a minute off the runtime of the isolation tests. Back-patch to all supported branches, because the buildfarm instabilities affect all the branches, and because it seems desirable to keep isolationtester's capabilities the same across all branches to simplify possible future back-patching of tests. Discussion: https://postgr.es/m/327948.1623725828@sss.pgh.pa.us
2021-06-23 03:43:12 +02:00
static void run_permutation(TestSpec *testspec, int nsteps,
PermutationStep **steps);
Implement genuine serializable isolation level. Until now, our Serializable mode has in fact been what's called Snapshot Isolation, which allows some anomalies that could not occur in any serialized ordering of the transactions. This patch fixes that using a method called Serializable Snapshot Isolation, based on research papers by Michael J. Cahill (see README-SSI for full references). In Serializable Snapshot Isolation, transactions run like they do in Snapshot Isolation, but a predicate lock manager observes the reads and writes performed and aborts transactions if it detects that an anomaly might occur. This method produces some false positives, ie. it sometimes aborts transactions even though there is no anomaly. To track reads we implement predicate locking, see storage/lmgr/predicate.c. Whenever a tuple is read, a predicate lock is acquired on the tuple. Shared memory is finite, so when a transaction takes many tuple-level locks on a page, the locks are promoted to a single page-level lock, and further to a single relation level lock if necessary. To lock key values with no matching tuple, a sequential scan always takes a relation-level lock, and an index scan acquires a page-level lock that covers the search key, whether or not there are any matching keys at the moment. A predicate lock doesn't conflict with any regular locks or with another predicate locks in the normal sense. They're only used by the predicate lock manager to detect the danger of anomalies. Only serializable transactions participate in predicate locking, so there should be no extra overhead for for other transactions. Predicate locks can't be released at commit, but must be remembered until all the transactions that overlapped with it have completed. That means that we need to remember an unbounded amount of predicate locks, so we apply a lossy but conservative method of tracking locks for committed transactions. If we run short of shared memory, we overflow to a new "pg_serial" SLRU pool. We don't currently allow Serializable transactions in Hot Standby mode. That would be hard, because even read-only transactions can cause anomalies that wouldn't otherwise occur. Serializable isolation mode now means the new fully serializable level. Repeatable Read gives you the old Snapshot Isolation level that we have always had. Kevin Grittner and Dan Ports, reviewed by Jeff Davis, Heikki Linnakangas and Anssi Kääriäinen
2011-02-07 22:46:51 +01:00
Use annotations to reduce instability of isolation-test results. We've long contended with isolation test results that aren't entirely stable. Some test scripts insert long delays to try to force stable results, which is not terribly desirable; but other erratic failure modes remain, causing unrepeatable buildfarm failures. I've spent a fair amount of time trying to solve this by improving the server-side support code, without much success: that way is fundamentally unable to cope with diffs that stem from chance ordering of arrival of messages from different server processes. We can improve matters on the client side, however, by annotating the test scripts themselves to show the desired reporting order of events that might occur in different orders. This patch adds three types of annotations to deal with (a) test steps that might or might not complete their waits before the isolationtester can see them waiting; (b) test steps in different sessions that can legitimately complete in either order; and (c) NOTIFY messages that might arrive before or after the completion of a step in another session. We might need more annotation types later, but this seems to be enough to deal with the instabilities we've seen in the buildfarm. It also lets us get rid of all the long delays that were previously used, cutting more than a minute off the runtime of the isolation tests. Back-patch to all supported branches, because the buildfarm instabilities affect all the branches, and because it seems desirable to keep isolationtester's capabilities the same across all branches to simplify possible future back-patching of tests. Discussion: https://postgr.es/m/327948.1623725828@sss.pgh.pa.us
2021-06-23 03:43:12 +02:00
/* Flag bits for try_complete_step(s) */
#define STEP_NONBLOCK 0x1 /* return as soon as cmd waits for a lock */
Add support for blocked commands in isolationtester This enables us to test that blocking commands (such as foreign keys checks that conflict with some other lock) act as intended. The set of tests that this adds is pretty minimal, but can easily be extended by adding new specs. The intention is that this will serve as a basis for ensuring that further tweaks of locking implementation preserve (or improve) existing behavior. Author: Noah Misch
2011-07-12 22:35:09 +02:00
#define STEP_RETRY 0x2 /* this is a retry of a previously-waiting cmd */
Use annotations to reduce instability of isolation-test results. We've long contended with isolation test results that aren't entirely stable. Some test scripts insert long delays to try to force stable results, which is not terribly desirable; but other erratic failure modes remain, causing unrepeatable buildfarm failures. I've spent a fair amount of time trying to solve this by improving the server-side support code, without much success: that way is fundamentally unable to cope with diffs that stem from chance ordering of arrival of messages from different server processes. We can improve matters on the client side, however, by annotating the test scripts themselves to show the desired reporting order of events that might occur in different orders. This patch adds three types of annotations to deal with (a) test steps that might or might not complete their waits before the isolationtester can see them waiting; (b) test steps in different sessions that can legitimately complete in either order; and (c) NOTIFY messages that might arrive before or after the completion of a step in another session. We might need more annotation types later, but this seems to be enough to deal with the instabilities we've seen in the buildfarm. It also lets us get rid of all the long delays that were previously used, cutting more than a minute off the runtime of the isolation tests. Back-patch to all supported branches, because the buildfarm instabilities affect all the branches, and because it seems desirable to keep isolationtester's capabilities the same across all branches to simplify possible future back-patching of tests. Discussion: https://postgr.es/m/327948.1623725828@sss.pgh.pa.us
2021-06-23 03:43:12 +02:00
static int try_complete_steps(TestSpec *testspec, PermutationStep **waiting,
int nwaiting, int flags);
static bool try_complete_step(TestSpec *testspec, PermutationStep *pstep,
int flags);
Add support for blocked commands in isolationtester This enables us to test that blocking commands (such as foreign keys checks that conflict with some other lock) act as intended. The set of tests that this adds is pretty minimal, but can easily be extended by adding new specs. The intention is that this will serve as a basis for ensuring that further tweaks of locking implementation preserve (or improve) existing behavior. Author: Noah Misch
2011-07-12 22:35:09 +02:00
Implement genuine serializable isolation level. Until now, our Serializable mode has in fact been what's called Snapshot Isolation, which allows some anomalies that could not occur in any serialized ordering of the transactions. This patch fixes that using a method called Serializable Snapshot Isolation, based on research papers by Michael J. Cahill (see README-SSI for full references). In Serializable Snapshot Isolation, transactions run like they do in Snapshot Isolation, but a predicate lock manager observes the reads and writes performed and aborts transactions if it detects that an anomaly might occur. This method produces some false positives, ie. it sometimes aborts transactions even though there is no anomaly. To track reads we implement predicate locking, see storage/lmgr/predicate.c. Whenever a tuple is read, a predicate lock is acquired on the tuple. Shared memory is finite, so when a transaction takes many tuple-level locks on a page, the locks are promoted to a single page-level lock, and further to a single relation level lock if necessary. To lock key values with no matching tuple, a sequential scan always takes a relation-level lock, and an index scan acquires a page-level lock that covers the search key, whether or not there are any matching keys at the moment. A predicate lock doesn't conflict with any regular locks or with another predicate locks in the normal sense. They're only used by the predicate lock manager to detect the danger of anomalies. Only serializable transactions participate in predicate locking, so there should be no extra overhead for for other transactions. Predicate locks can't be released at commit, but must be remembered until all the transactions that overlapped with it have completed. That means that we need to remember an unbounded amount of predicate locks, so we apply a lossy but conservative method of tracking locks for committed transactions. If we run short of shared memory, we overflow to a new "pg_serial" SLRU pool. We don't currently allow Serializable transactions in Hot Standby mode. That would be hard, because even read-only transactions can cause anomalies that wouldn't otherwise occur. Serializable isolation mode now means the new fully serializable level. Repeatable Read gives you the old Snapshot Isolation level that we have always had. Kevin Grittner and Dan Ports, reviewed by Jeff Davis, Heikki Linnakangas and Anssi Kääriäinen
2011-02-07 22:46:51 +01:00
static int step_qsort_cmp(const void *a, const void *b);
static int step_bsearch_cmp(const void *a, const void *b);
Use annotations to reduce instability of isolation-test results. We've long contended with isolation test results that aren't entirely stable. Some test scripts insert long delays to try to force stable results, which is not terribly desirable; but other erratic failure modes remain, causing unrepeatable buildfarm failures. I've spent a fair amount of time trying to solve this by improving the server-side support code, without much success: that way is fundamentally unable to cope with diffs that stem from chance ordering of arrival of messages from different server processes. We can improve matters on the client side, however, by annotating the test scripts themselves to show the desired reporting order of events that might occur in different orders. This patch adds three types of annotations to deal with (a) test steps that might or might not complete their waits before the isolationtester can see them waiting; (b) test steps in different sessions that can legitimately complete in either order; and (c) NOTIFY messages that might arrive before or after the completion of a step in another session. We might need more annotation types later, but this seems to be enough to deal with the instabilities we've seen in the buildfarm. It also lets us get rid of all the long delays that were previously used, cutting more than a minute off the runtime of the isolation tests. Back-patch to all supported branches, because the buildfarm instabilities affect all the branches, and because it seems desirable to keep isolationtester's capabilities the same across all branches to simplify possible future back-patching of tests. Discussion: https://postgr.es/m/327948.1623725828@sss.pgh.pa.us
2021-06-23 03:43:12 +02:00
static bool step_has_blocker(PermutationStep *pstep);
Implement genuine serializable isolation level. Until now, our Serializable mode has in fact been what's called Snapshot Isolation, which allows some anomalies that could not occur in any serialized ordering of the transactions. This patch fixes that using a method called Serializable Snapshot Isolation, based on research papers by Michael J. Cahill (see README-SSI for full references). In Serializable Snapshot Isolation, transactions run like they do in Snapshot Isolation, but a predicate lock manager observes the reads and writes performed and aborts transactions if it detects that an anomaly might occur. This method produces some false positives, ie. it sometimes aborts transactions even though there is no anomaly. To track reads we implement predicate locking, see storage/lmgr/predicate.c. Whenever a tuple is read, a predicate lock is acquired on the tuple. Shared memory is finite, so when a transaction takes many tuple-level locks on a page, the locks are promoted to a single page-level lock, and further to a single relation level lock if necessary. To lock key values with no matching tuple, a sequential scan always takes a relation-level lock, and an index scan acquires a page-level lock that covers the search key, whether or not there are any matching keys at the moment. A predicate lock doesn't conflict with any regular locks or with another predicate locks in the normal sense. They're only used by the predicate lock manager to detect the danger of anomalies. Only serializable transactions participate in predicate locking, so there should be no extra overhead for for other transactions. Predicate locks can't be released at commit, but must be remembered until all the transactions that overlapped with it have completed. That means that we need to remember an unbounded amount of predicate locks, so we apply a lossy but conservative method of tracking locks for committed transactions. If we run short of shared memory, we overflow to a new "pg_serial" SLRU pool. We don't currently allow Serializable transactions in Hot Standby mode. That would be hard, because even read-only transactions can cause anomalies that wouldn't otherwise occur. Serializable isolation mode now means the new fully serializable level. Repeatable Read gives you the old Snapshot Isolation level that we have always had. Kevin Grittner and Dan Ports, reviewed by Jeff Davis, Heikki Linnakangas and Anssi Kääriäinen
2011-02-07 22:46:51 +01:00
static void printResultSet(PGresult *res);
Indicate session name in isolationtester notices When a session under isolationtester produces printable notices (NOTICE, WARNING) we were just printing them unadorned, which can be confusing when debugging. Prefix them with the session name, which makes things clearer. Author: Álvaro Herrera Reviewed-by: Hari Babu Kommi Discussion: https://postgr.es/m/20181024213451.75nh3f3dctmcdbfq@alvherre.pgsql
2018-11-09 17:08:00 +01:00
static void isotesterNoticeProcessor(void *arg, const char *message);
static void blackholeNoticeProcessor(void *arg, const char *message);
Implement genuine serializable isolation level. Until now, our Serializable mode has in fact been what's called Snapshot Isolation, which allows some anomalies that could not occur in any serialized ordering of the transactions. This patch fixes that using a method called Serializable Snapshot Isolation, based on research papers by Michael J. Cahill (see README-SSI for full references). In Serializable Snapshot Isolation, transactions run like they do in Snapshot Isolation, but a predicate lock manager observes the reads and writes performed and aborts transactions if it detects that an anomaly might occur. This method produces some false positives, ie. it sometimes aborts transactions even though there is no anomaly. To track reads we implement predicate locking, see storage/lmgr/predicate.c. Whenever a tuple is read, a predicate lock is acquired on the tuple. Shared memory is finite, so when a transaction takes many tuple-level locks on a page, the locks are promoted to a single page-level lock, and further to a single relation level lock if necessary. To lock key values with no matching tuple, a sequential scan always takes a relation-level lock, and an index scan acquires a page-level lock that covers the search key, whether or not there are any matching keys at the moment. A predicate lock doesn't conflict with any regular locks or with another predicate locks in the normal sense. They're only used by the predicate lock manager to detect the danger of anomalies. Only serializable transactions participate in predicate locking, so there should be no extra overhead for for other transactions. Predicate locks can't be released at commit, but must be remembered until all the transactions that overlapped with it have completed. That means that we need to remember an unbounded amount of predicate locks, so we apply a lossy but conservative method of tracking locks for committed transactions. If we run short of shared memory, we overflow to a new "pg_serial" SLRU pool. We don't currently allow Serializable transactions in Hot Standby mode. That would be hard, because even read-only transactions can cause anomalies that wouldn't otherwise occur. Serializable isolation mode now means the new fully serializable level. Repeatable Read gives you the old Snapshot Isolation level that we have always had. Kevin Grittner and Dan Ports, reviewed by Jeff Davis, Heikki Linnakangas and Anssi Kääriäinen
2011-02-07 22:46:51 +01:00
static void
isolationtester: Use atexit() Replace exit_nicely() calls with standard exit() and register the cleanup actions using atexit(). Reviewed-by: Alvaro Herrera <alvherre@2ndquadrant.com> Discussion: https://www.postgresql.org/message-id/flat/ec4135ba-84e9-28bf-b584-0e78d47448d5@2ndquadrant.com/
2019-01-05 15:05:49 +01:00
disconnect_atexit(void)
Implement genuine serializable isolation level. Until now, our Serializable mode has in fact been what's called Snapshot Isolation, which allows some anomalies that could not occur in any serialized ordering of the transactions. This patch fixes that using a method called Serializable Snapshot Isolation, based on research papers by Michael J. Cahill (see README-SSI for full references). In Serializable Snapshot Isolation, transactions run like they do in Snapshot Isolation, but a predicate lock manager observes the reads and writes performed and aborts transactions if it detects that an anomaly might occur. This method produces some false positives, ie. it sometimes aborts transactions even though there is no anomaly. To track reads we implement predicate locking, see storage/lmgr/predicate.c. Whenever a tuple is read, a predicate lock is acquired on the tuple. Shared memory is finite, so when a transaction takes many tuple-level locks on a page, the locks are promoted to a single page-level lock, and further to a single relation level lock if necessary. To lock key values with no matching tuple, a sequential scan always takes a relation-level lock, and an index scan acquires a page-level lock that covers the search key, whether or not there are any matching keys at the moment. A predicate lock doesn't conflict with any regular locks or with another predicate locks in the normal sense. They're only used by the predicate lock manager to detect the danger of anomalies. Only serializable transactions participate in predicate locking, so there should be no extra overhead for for other transactions. Predicate locks can't be released at commit, but must be remembered until all the transactions that overlapped with it have completed. That means that we need to remember an unbounded amount of predicate locks, so we apply a lossy but conservative method of tracking locks for committed transactions. If we run short of shared memory, we overflow to a new "pg_serial" SLRU pool. We don't currently allow Serializable transactions in Hot Standby mode. That would be hard, because even read-only transactions can cause anomalies that wouldn't otherwise occur. Serializable isolation mode now means the new fully serializable level. Repeatable Read gives you the old Snapshot Isolation level that we have always had. Kevin Grittner and Dan Ports, reviewed by Jeff Davis, Heikki Linnakangas and Anssi Kääriäinen
2011-02-07 22:46:51 +01:00
{
int i;
pgindent run before PG 9.1 beta 1.
2011-04-10 17:42:00 +02:00
Implement genuine serializable isolation level. Until now, our Serializable mode has in fact been what's called Snapshot Isolation, which allows some anomalies that could not occur in any serialized ordering of the transactions. This patch fixes that using a method called Serializable Snapshot Isolation, based on research papers by Michael J. Cahill (see README-SSI for full references). In Serializable Snapshot Isolation, transactions run like they do in Snapshot Isolation, but a predicate lock manager observes the reads and writes performed and aborts transactions if it detects that an anomaly might occur. This method produces some false positives, ie. it sometimes aborts transactions even though there is no anomaly. To track reads we implement predicate locking, see storage/lmgr/predicate.c. Whenever a tuple is read, a predicate lock is acquired on the tuple. Shared memory is finite, so when a transaction takes many tuple-level locks on a page, the locks are promoted to a single page-level lock, and further to a single relation level lock if necessary. To lock key values with no matching tuple, a sequential scan always takes a relation-level lock, and an index scan acquires a page-level lock that covers the search key, whether or not there are any matching keys at the moment. A predicate lock doesn't conflict with any regular locks or with another predicate locks in the normal sense. They're only used by the predicate lock manager to detect the danger of anomalies. Only serializable transactions participate in predicate locking, so there should be no extra overhead for for other transactions. Predicate locks can't be released at commit, but must be remembered until all the transactions that overlapped with it have completed. That means that we need to remember an unbounded amount of predicate locks, so we apply a lossy but conservative method of tracking locks for committed transactions. If we run short of shared memory, we overflow to a new "pg_serial" SLRU pool. We don't currently allow Serializable transactions in Hot Standby mode. That would be hard, because even read-only transactions can cause anomalies that wouldn't otherwise occur. Serializable isolation mode now means the new fully serializable level. Repeatable Read gives you the old Snapshot Isolation level that we have always had. Kevin Grittner and Dan Ports, reviewed by Jeff Davis, Heikki Linnakangas and Anssi Kääriäinen
2011-02-07 22:46:51 +01:00
for (i = 0; i < nconns; i++)
Use annotations to reduce instability of isolation-test results. We've long contended with isolation test results that aren't entirely stable. Some test scripts insert long delays to try to force stable results, which is not terribly desirable; but other erratic failure modes remain, causing unrepeatable buildfarm failures. I've spent a fair amount of time trying to solve this by improving the server-side support code, without much success: that way is fundamentally unable to cope with diffs that stem from chance ordering of arrival of messages from different server processes. We can improve matters on the client side, however, by annotating the test scripts themselves to show the desired reporting order of events that might occur in different orders. This patch adds three types of annotations to deal with (a) test steps that might or might not complete their waits before the isolationtester can see them waiting; (b) test steps in different sessions that can legitimately complete in either order; and (c) NOTIFY messages that might arrive before or after the completion of a step in another session. We might need more annotation types later, but this seems to be enough to deal with the instabilities we've seen in the buildfarm. It also lets us get rid of all the long delays that were previously used, cutting more than a minute off the runtime of the isolation tests. Back-patch to all supported branches, because the buildfarm instabilities affect all the branches, and because it seems desirable to keep isolationtester's capabilities the same across all branches to simplify possible future back-patching of tests. Discussion: https://postgr.es/m/327948.1623725828@sss.pgh.pa.us
2021-06-23 03:43:12 +02:00
if (conns[i].conn)
PQfinish(conns[i].conn);
Implement genuine serializable isolation level. Until now, our Serializable mode has in fact been what's called Snapshot Isolation, which allows some anomalies that could not occur in any serialized ordering of the transactions. This patch fixes that using a method called Serializable Snapshot Isolation, based on research papers by Michael J. Cahill (see README-SSI for full references). In Serializable Snapshot Isolation, transactions run like they do in Snapshot Isolation, but a predicate lock manager observes the reads and writes performed and aborts transactions if it detects that an anomaly might occur. This method produces some false positives, ie. it sometimes aborts transactions even though there is no anomaly. To track reads we implement predicate locking, see storage/lmgr/predicate.c. Whenever a tuple is read, a predicate lock is acquired on the tuple. Shared memory is finite, so when a transaction takes many tuple-level locks on a page, the locks are promoted to a single page-level lock, and further to a single relation level lock if necessary. To lock key values with no matching tuple, a sequential scan always takes a relation-level lock, and an index scan acquires a page-level lock that covers the search key, whether or not there are any matching keys at the moment. A predicate lock doesn't conflict with any regular locks or with another predicate locks in the normal sense. They're only used by the predicate lock manager to detect the danger of anomalies. Only serializable transactions participate in predicate locking, so there should be no extra overhead for for other transactions. Predicate locks can't be released at commit, but must be remembered until all the transactions that overlapped with it have completed. That means that we need to remember an unbounded amount of predicate locks, so we apply a lossy but conservative method of tracking locks for committed transactions. If we run short of shared memory, we overflow to a new "pg_serial" SLRU pool. We don't currently allow Serializable transactions in Hot Standby mode. That would be hard, because even read-only transactions can cause anomalies that wouldn't otherwise occur. Serializable isolation mode now means the new fully serializable level. Repeatable Read gives you the old Snapshot Isolation level that we have always had. Kevin Grittner and Dan Ports, reviewed by Jeff Davis, Heikki Linnakangas and Anssi Kääriäinen
2011-02-07 22:46:51 +01:00
}
int
main(int argc, char **argv)
{
const char *conninfo;
Improve isolationtester's timeout management. isolationtester.c had a hard-wired limit of 3 minutes per test step. It now emerges that this isn't quite enough for some of the slowest buildfarm animals. This isn't the first time we've had to raise this limit (cf. 1db439ad4), so let's make it configurable. This patch raises the default to 5 minutes, and introduces an environment variable PGISOLATIONTIMEOUT that can be set if more time is needed, following the precedent of PGCTLTIMEOUT. Also, modify isolationtester so that when the timeout is hit, it explicitly reports having sent a cancel. This makes the regression failure log considerably more intelligible. (In the worst case, a timed-out test might actually be reported as "passing" without this extra output, so arguably this is a bug fix in itself.) In passing, update the README file, which had apparently not gotten touched when we added "make check" support here. Back-patch to 9.6; older versions don't have comparable timeout logic. Discussion: https://postgr.es/m/22964.1575842935@sss.pgh.pa.us
2019-12-09 20:31:57 +01:00
const char *env_wait;
Implement genuine serializable isolation level. Until now, our Serializable mode has in fact been what's called Snapshot Isolation, which allows some anomalies that could not occur in any serialized ordering of the transactions. This patch fixes that using a method called Serializable Snapshot Isolation, based on research papers by Michael J. Cahill (see README-SSI for full references). In Serializable Snapshot Isolation, transactions run like they do in Snapshot Isolation, but a predicate lock manager observes the reads and writes performed and aborts transactions if it detects that an anomaly might occur. This method produces some false positives, ie. it sometimes aborts transactions even though there is no anomaly. To track reads we implement predicate locking, see storage/lmgr/predicate.c. Whenever a tuple is read, a predicate lock is acquired on the tuple. Shared memory is finite, so when a transaction takes many tuple-level locks on a page, the locks are promoted to a single page-level lock, and further to a single relation level lock if necessary. To lock key values with no matching tuple, a sequential scan always takes a relation-level lock, and an index scan acquires a page-level lock that covers the search key, whether or not there are any matching keys at the moment. A predicate lock doesn't conflict with any regular locks or with another predicate locks in the normal sense. They're only used by the predicate lock manager to detect the danger of anomalies. Only serializable transactions participate in predicate locking, so there should be no extra overhead for for other transactions. Predicate locks can't be released at commit, but must be remembered until all the transactions that overlapped with it have completed. That means that we need to remember an unbounded amount of predicate locks, so we apply a lossy but conservative method of tracking locks for committed transactions. If we run short of shared memory, we overflow to a new "pg_serial" SLRU pool. We don't currently allow Serializable transactions in Hot Standby mode. That would be hard, because even read-only transactions can cause anomalies that wouldn't otherwise occur. Serializable isolation mode now means the new fully serializable level. Repeatable Read gives you the old Snapshot Isolation level that we have always had. Kevin Grittner and Dan Ports, reviewed by Jeff Davis, Heikki Linnakangas and Anssi Kääriäinen
2011-02-07 22:46:51 +01:00
TestSpec *testspec;
Add support for blocked commands in isolationtester This enables us to test that blocking commands (such as foreign keys checks that conflict with some other lock) act as intended. The set of tests that this adds is pretty minimal, but can easily be extended by adding new specs. The intention is that this will serve as a basis for ensuring that further tweaks of locking implementation preserve (or improve) existing behavior. Author: Noah Misch
2011-07-12 22:35:09 +02:00
PGresult *res;
Make isolationtester more robust on locked commands Noah Misch diagnosed the buildfarm problems in the isolation tests partly as failure to differentiate backends properly; the old code was using backend IDs, which is not good enough because a new backend might use an already used ID. Use PIDs instead. Also, the code was purposely careless about other concurrent activity, because it isn't expected; and in fact, it doesn't affect the vast majority of the time. However, it can be observed that autovacuum can block tables for long enough to cause sporadic failures. The new code accounts for that by ignoring locks held by processes not explicitly declared in our spec file. Author: Noah Misch
2011-07-19 20:22:42 +02:00
PQExpBufferData wait_query;
Implement a dry-run mode for isolationtester This mode prints out the permutations that would be run by the given spec file, in the same format used by the permutation lines in spec files. This helps in building new spec files. Author: Alexander Shulgin, with some tweaks by me
2011-11-03 18:06:08 +01:00
int opt;
Use annotations to reduce instability of isolation-test results. We've long contended with isolation test results that aren't entirely stable. Some test scripts insert long delays to try to force stable results, which is not terribly desirable; but other erratic failure modes remain, causing unrepeatable buildfarm failures. I've spent a fair amount of time trying to solve this by improving the server-side support code, without much success: that way is fundamentally unable to cope with diffs that stem from chance ordering of arrival of messages from different server processes. We can improve matters on the client side, however, by annotating the test scripts themselves to show the desired reporting order of events that might occur in different orders. This patch adds three types of annotations to deal with (a) test steps that might or might not complete their waits before the isolationtester can see them waiting; (b) test steps in different sessions that can legitimately complete in either order; and (c) NOTIFY messages that might arrive before or after the completion of a step in another session. We might need more annotation types later, but this seems to be enough to deal with the instabilities we've seen in the buildfarm. It also lets us get rid of all the long delays that were previously used, cutting more than a minute off the runtime of the isolation tests. Back-patch to all supported branches, because the buildfarm instabilities affect all the branches, and because it seems desirable to keep isolationtester's capabilities the same across all branches to simplify possible future back-patching of tests. Discussion: https://postgr.es/m/327948.1623725828@sss.pgh.pa.us
2021-06-23 03:43:12 +02:00
int i;
Implement a dry-run mode for isolationtester This mode prints out the permutations that would be run by the given spec file, in the same format used by the permutation lines in spec files. This helps in building new spec files. Author: Alexander Shulgin, with some tweaks by me
2011-11-03 18:06:08 +01:00
Remove dry-run mode from isolationtester The original purpose of the dry-run mode is to be able to print all the possible permutations from a spec file, but it has become less useful since isolation tests has improved regarding deadlock detection as one step not wanted by the author could block indefinitely now (originally the step blocked would have been detected rather quickly). Per discussion, let's remove it. Author: Michael Paquier Reviewed-by: Asim Praveen, Melanie Plageman Discussion: https://postgr.es/m/20190819080820.GG18166@paquier.xyz
2019-08-24 04:35:43 +02:00
while ((opt = getopt(argc, argv, "V")) != -1)
Implement a dry-run mode for isolationtester This mode prints out the permutations that would be run by the given spec file, in the same format used by the permutation lines in spec files. This helps in building new spec files. Author: Alexander Shulgin, with some tweaks by me
2011-11-03 18:06:08 +01:00
{
switch (opt)
{
Fix pg_isolation_regress to work outside its build directory. This makes it possible to, for example, use the isolation tester to test a contrib module. Andres Freund
2013-11-08 20:40:41 +01:00
case 'V':
puts("isolationtester (PostgreSQL) " PG_VERSION);
exit(0);
Implement a dry-run mode for isolationtester This mode prints out the permutations that would be run by the given spec file, in the same format used by the permutation lines in spec files. This helps in building new spec files. Author: Alexander Shulgin, with some tweaks by me
2011-11-03 18:06:08 +01:00
default:
Detect unused steps in isolation specs and do some cleanup This is useful for developers to find out if an isolation spec is over-engineered or if it needs more work by warning at the end of a test run if a step is not used, generating a failure with extra diffs. While on it, clean up all the specs which include steps not used in any permutations to simplify them. Author: Michael Paquier Reviewed-by: Asim Praveen, Melanie Plageman Discussion: https://postgr.es/m/20190819080820.GG18166@paquier.xyz
2019-08-24 04:45:05 +02:00
fprintf(stderr, "Usage: isolationtester [CONNINFO]\n");
Implement a dry-run mode for isolationtester This mode prints out the permutations that would be run by the given spec file, in the same format used by the permutation lines in spec files. This helps in building new spec files. Author: Alexander Shulgin, with some tweaks by me
2011-11-03 18:06:08 +01:00
return EXIT_FAILURE;
}
}
Implement genuine serializable isolation level. Until now, our Serializable mode has in fact been what's called Snapshot Isolation, which allows some anomalies that could not occur in any serialized ordering of the transactions. This patch fixes that using a method called Serializable Snapshot Isolation, based on research papers by Michael J. Cahill (see README-SSI for full references). In Serializable Snapshot Isolation, transactions run like they do in Snapshot Isolation, but a predicate lock manager observes the reads and writes performed and aborts transactions if it detects that an anomaly might occur. This method produces some false positives, ie. it sometimes aborts transactions even though there is no anomaly. To track reads we implement predicate locking, see storage/lmgr/predicate.c. Whenever a tuple is read, a predicate lock is acquired on the tuple. Shared memory is finite, so when a transaction takes many tuple-level locks on a page, the locks are promoted to a single page-level lock, and further to a single relation level lock if necessary. To lock key values with no matching tuple, a sequential scan always takes a relation-level lock, and an index scan acquires a page-level lock that covers the search key, whether or not there are any matching keys at the moment. A predicate lock doesn't conflict with any regular locks or with another predicate locks in the normal sense. They're only used by the predicate lock manager to detect the danger of anomalies. Only serializable transactions participate in predicate locking, so there should be no extra overhead for for other transactions. Predicate locks can't be released at commit, but must be remembered until all the transactions that overlapped with it have completed. That means that we need to remember an unbounded amount of predicate locks, so we apply a lossy but conservative method of tracking locks for committed transactions. If we run short of shared memory, we overflow to a new "pg_serial" SLRU pool. We don't currently allow Serializable transactions in Hot Standby mode. That would be hard, because even read-only transactions can cause anomalies that wouldn't otherwise occur. Serializable isolation mode now means the new fully serializable level. Repeatable Read gives you the old Snapshot Isolation level that we have always had. Kevin Grittner and Dan Ports, reviewed by Jeff Davis, Heikki Linnakangas and Anssi Kääriäinen
2011-02-07 22:46:51 +01:00
isolationtester: Ensure stderr is unbuffered, too
2013-12-20 02:09:30 +01:00
/*
* Make stdout unbuffered to match stderr; and ensure stderr is unbuffered
* too, which it should already be everywhere except sometimes in Windows.
*/
Make stdout unbuffered This ensures that all stdout output is flushed immediately, to match stderr. This eliminates the need for fflush(stdout) calls sprinkled all over the place. Per Daniel Wood in message 519A79C6.90308@salesforce.com
2013-12-19 21:26:27 +01:00
setbuf(stdout, NULL);
isolationtester: Ensure stderr is unbuffered, too
2013-12-20 02:09:30 +01:00
setbuf(stderr, NULL);
Make stdout unbuffered This ensures that all stdout output is flushed immediately, to match stderr. This eliminates the need for fflush(stdout) calls sprinkled all over the place. Per Daniel Wood in message 519A79C6.90308@salesforce.com
2013-12-19 21:26:27 +01:00
Implement genuine serializable isolation level. Until now, our Serializable mode has in fact been what's called Snapshot Isolation, which allows some anomalies that could not occur in any serialized ordering of the transactions. This patch fixes that using a method called Serializable Snapshot Isolation, based on research papers by Michael J. Cahill (see README-SSI for full references). In Serializable Snapshot Isolation, transactions run like they do in Snapshot Isolation, but a predicate lock manager observes the reads and writes performed and aborts transactions if it detects that an anomaly might occur. This method produces some false positives, ie. it sometimes aborts transactions even though there is no anomaly. To track reads we implement predicate locking, see storage/lmgr/predicate.c. Whenever a tuple is read, a predicate lock is acquired on the tuple. Shared memory is finite, so when a transaction takes many tuple-level locks on a page, the locks are promoted to a single page-level lock, and further to a single relation level lock if necessary. To lock key values with no matching tuple, a sequential scan always takes a relation-level lock, and an index scan acquires a page-level lock that covers the search key, whether or not there are any matching keys at the moment. A predicate lock doesn't conflict with any regular locks or with another predicate locks in the normal sense. They're only used by the predicate lock manager to detect the danger of anomalies. Only serializable transactions participate in predicate locking, so there should be no extra overhead for for other transactions. Predicate locks can't be released at commit, but must be remembered until all the transactions that overlapped with it have completed. That means that we need to remember an unbounded amount of predicate locks, so we apply a lossy but conservative method of tracking locks for committed transactions. If we run short of shared memory, we overflow to a new "pg_serial" SLRU pool. We don't currently allow Serializable transactions in Hot Standby mode. That would be hard, because even read-only transactions can cause anomalies that wouldn't otherwise occur. Serializable isolation mode now means the new fully serializable level. Repeatable Read gives you the old Snapshot Isolation level that we have always had. Kevin Grittner and Dan Ports, reviewed by Jeff Davis, Heikki Linnakangas and Anssi Kääriäinen
2011-02-07 22:46:51 +01:00
/*
Implement a dry-run mode for isolationtester This mode prints out the permutations that would be run by the given spec file, in the same format used by the permutation lines in spec files. This helps in building new spec files. Author: Alexander Shulgin, with some tweaks by me
2011-11-03 18:06:08 +01:00
* If the user supplies a non-option parameter on the command line, use it
* as the conninfo string; otherwise default to setting dbname=postgres
* and using environment variables or defaults for all other connection
* parameters.
Implement genuine serializable isolation level. Until now, our Serializable mode has in fact been what's called Snapshot Isolation, which allows some anomalies that could not occur in any serialized ordering of the transactions. This patch fixes that using a method called Serializable Snapshot Isolation, based on research papers by Michael J. Cahill (see README-SSI for full references). In Serializable Snapshot Isolation, transactions run like they do in Snapshot Isolation, but a predicate lock manager observes the reads and writes performed and aborts transactions if it detects that an anomaly might occur. This method produces some false positives, ie. it sometimes aborts transactions even though there is no anomaly. To track reads we implement predicate locking, see storage/lmgr/predicate.c. Whenever a tuple is read, a predicate lock is acquired on the tuple. Shared memory is finite, so when a transaction takes many tuple-level locks on a page, the locks are promoted to a single page-level lock, and further to a single relation level lock if necessary. To lock key values with no matching tuple, a sequential scan always takes a relation-level lock, and an index scan acquires a page-level lock that covers the search key, whether or not there are any matching keys at the moment. A predicate lock doesn't conflict with any regular locks or with another predicate locks in the normal sense. They're only used by the predicate lock manager to detect the danger of anomalies. Only serializable transactions participate in predicate locking, so there should be no extra overhead for for other transactions. Predicate locks can't be released at commit, but must be remembered until all the transactions that overlapped with it have completed. That means that we need to remember an unbounded amount of predicate locks, so we apply a lossy but conservative method of tracking locks for committed transactions. If we run short of shared memory, we overflow to a new "pg_serial" SLRU pool. We don't currently allow Serializable transactions in Hot Standby mode. That would be hard, because even read-only transactions can cause anomalies that wouldn't otherwise occur. Serializable isolation mode now means the new fully serializable level. Repeatable Read gives you the old Snapshot Isolation level that we have always had. Kevin Grittner and Dan Ports, reviewed by Jeff Davis, Heikki Linnakangas and Anssi Kääriäinen
2011-02-07 22:46:51 +01:00
*/
Implement a dry-run mode for isolationtester This mode prints out the permutations that would be run by the given spec file, in the same format used by the permutation lines in spec files. This helps in building new spec files. Author: Alexander Shulgin, with some tweaks by me
2011-11-03 18:06:08 +01:00
if (argc > optind)
conninfo = argv[optind];
Implement genuine serializable isolation level. Until now, our Serializable mode has in fact been what's called Snapshot Isolation, which allows some anomalies that could not occur in any serialized ordering of the transactions. This patch fixes that using a method called Serializable Snapshot Isolation, based on research papers by Michael J. Cahill (see README-SSI for full references). In Serializable Snapshot Isolation, transactions run like they do in Snapshot Isolation, but a predicate lock manager observes the reads and writes performed and aborts transactions if it detects that an anomaly might occur. This method produces some false positives, ie. it sometimes aborts transactions even though there is no anomaly. To track reads we implement predicate locking, see storage/lmgr/predicate.c. Whenever a tuple is read, a predicate lock is acquired on the tuple. Shared memory is finite, so when a transaction takes many tuple-level locks on a page, the locks are promoted to a single page-level lock, and further to a single relation level lock if necessary. To lock key values with no matching tuple, a sequential scan always takes a relation-level lock, and an index scan acquires a page-level lock that covers the search key, whether or not there are any matching keys at the moment. A predicate lock doesn't conflict with any regular locks or with another predicate locks in the normal sense. They're only used by the predicate lock manager to detect the danger of anomalies. Only serializable transactions participate in predicate locking, so there should be no extra overhead for for other transactions. Predicate locks can't be released at commit, but must be remembered until all the transactions that overlapped with it have completed. That means that we need to remember an unbounded amount of predicate locks, so we apply a lossy but conservative method of tracking locks for committed transactions. If we run short of shared memory, we overflow to a new "pg_serial" SLRU pool. We don't currently allow Serializable transactions in Hot Standby mode. That would be hard, because even read-only transactions can cause anomalies that wouldn't otherwise occur. Serializable isolation mode now means the new fully serializable level. Repeatable Read gives you the old Snapshot Isolation level that we have always had. Kevin Grittner and Dan Ports, reviewed by Jeff Davis, Heikki Linnakangas and Anssi Kääriäinen
2011-02-07 22:46:51 +01:00
else
conninfo = "dbname = postgres";
Improve isolationtester's timeout management. isolationtester.c had a hard-wired limit of 3 minutes per test step. It now emerges that this isn't quite enough for some of the slowest buildfarm animals. This isn't the first time we've had to raise this limit (cf. 1db439ad4), so let's make it configurable. This patch raises the default to 5 minutes, and introduces an environment variable PGISOLATIONTIMEOUT that can be set if more time is needed, following the precedent of PGCTLTIMEOUT. Also, modify isolationtester so that when the timeout is hit, it explicitly reports having sent a cancel. This makes the regression failure log considerably more intelligible. (In the worst case, a timed-out test might actually be reported as "passing" without this extra output, so arguably this is a bug fix in itself.) In passing, update the README file, which had apparently not gotten touched when we added "make check" support here. Back-patch to 9.6; older versions don't have comparable timeout logic. Discussion: https://postgr.es/m/22964.1575842935@sss.pgh.pa.us
2019-12-09 20:31:57 +01:00
/*
* If PGISOLATIONTIMEOUT is set in the environment, adopt its value (given
* in seconds) as the max time to wait for any one step to complete.
*/
env_wait = getenv("PGISOLATIONTIMEOUT");
if (env_wait != NULL)
max_step_wait = ((int64) atoi(env_wait)) * USECS_PER_SEC;
Implement genuine serializable isolation level. Until now, our Serializable mode has in fact been what's called Snapshot Isolation, which allows some anomalies that could not occur in any serialized ordering of the transactions. This patch fixes that using a method called Serializable Snapshot Isolation, based on research papers by Michael J. Cahill (see README-SSI for full references). In Serializable Snapshot Isolation, transactions run like they do in Snapshot Isolation, but a predicate lock manager observes the reads and writes performed and aborts transactions if it detects that an anomaly might occur. This method produces some false positives, ie. it sometimes aborts transactions even though there is no anomaly. To track reads we implement predicate locking, see storage/lmgr/predicate.c. Whenever a tuple is read, a predicate lock is acquired on the tuple. Shared memory is finite, so when a transaction takes many tuple-level locks on a page, the locks are promoted to a single page-level lock, and further to a single relation level lock if necessary. To lock key values with no matching tuple, a sequential scan always takes a relation-level lock, and an index scan acquires a page-level lock that covers the search key, whether or not there are any matching keys at the moment. A predicate lock doesn't conflict with any regular locks or with another predicate locks in the normal sense. They're only used by the predicate lock manager to detect the danger of anomalies. Only serializable transactions participate in predicate locking, so there should be no extra overhead for for other transactions. Predicate locks can't be released at commit, but must be remembered until all the transactions that overlapped with it have completed. That means that we need to remember an unbounded amount of predicate locks, so we apply a lossy but conservative method of tracking locks for committed transactions. If we run short of shared memory, we overflow to a new "pg_serial" SLRU pool. We don't currently allow Serializable transactions in Hot Standby mode. That would be hard, because even read-only transactions can cause anomalies that wouldn't otherwise occur. Serializable isolation mode now means the new fully serializable level. Repeatable Read gives you the old Snapshot Isolation level that we have always had. Kevin Grittner and Dan Ports, reviewed by Jeff Davis, Heikki Linnakangas and Anssi Kääriäinen
2011-02-07 22:46:51 +01:00
/* Read the test spec from stdin */
spec_yyparse();
testspec = &parseresult;
Implement a dry-run mode for isolationtester This mode prints out the permutations that would be run by the given spec file, in the same format used by the permutation lines in spec files. This helps in building new spec files. Author: Alexander Shulgin, with some tweaks by me
2011-11-03 18:06:08 +01:00
Use annotations to reduce instability of isolation-test results. We've long contended with isolation test results that aren't entirely stable. Some test scripts insert long delays to try to force stable results, which is not terribly desirable; but other erratic failure modes remain, causing unrepeatable buildfarm failures. I've spent a fair amount of time trying to solve this by improving the server-side support code, without much success: that way is fundamentally unable to cope with diffs that stem from chance ordering of arrival of messages from different server processes. We can improve matters on the client side, however, by annotating the test scripts themselves to show the desired reporting order of events that might occur in different orders. This patch adds three types of annotations to deal with (a) test steps that might or might not complete their waits before the isolationtester can see them waiting; (b) test steps in different sessions that can legitimately complete in either order; and (c) NOTIFY messages that might arrive before or after the completion of a step in another session. We might need more annotation types later, but this seems to be enough to deal with the instabilities we've seen in the buildfarm. It also lets us get rid of all the long delays that were previously used, cutting more than a minute off the runtime of the isolation tests. Back-patch to all supported branches, because the buildfarm instabilities affect all the branches, and because it seems desirable to keep isolationtester's capabilities the same across all branches to simplify possible future back-patching of tests. Discussion: https://postgr.es/m/327948.1623725828@sss.pgh.pa.us
2021-06-23 03:43:12 +02:00
/* Perform post-parse checking, and fill in linking fields */
check_testspec(testspec);
Reject isolation test specifications with duplicate step names. alter-table-1.spec has such a case, so change one instance of step rx1 to rx3 instead.
2015-08-15 04:09:27 +02:00
Implement genuine serializable isolation level. Until now, our Serializable mode has in fact been what's called Snapshot Isolation, which allows some anomalies that could not occur in any serialized ordering of the transactions. This patch fixes that using a method called Serializable Snapshot Isolation, based on research papers by Michael J. Cahill (see README-SSI for full references). In Serializable Snapshot Isolation, transactions run like they do in Snapshot Isolation, but a predicate lock manager observes the reads and writes performed and aborts transactions if it detects that an anomaly might occur. This method produces some false positives, ie. it sometimes aborts transactions even though there is no anomaly. To track reads we implement predicate locking, see storage/lmgr/predicate.c. Whenever a tuple is read, a predicate lock is acquired on the tuple. Shared memory is finite, so when a transaction takes many tuple-level locks on a page, the locks are promoted to a single page-level lock, and further to a single relation level lock if necessary. To lock key values with no matching tuple, a sequential scan always takes a relation-level lock, and an index scan acquires a page-level lock that covers the search key, whether or not there are any matching keys at the moment. A predicate lock doesn't conflict with any regular locks or with another predicate locks in the normal sense. They're only used by the predicate lock manager to detect the danger of anomalies. Only serializable transactions participate in predicate locking, so there should be no extra overhead for for other transactions. Predicate locks can't be released at commit, but must be remembered until all the transactions that overlapped with it have completed. That means that we need to remember an unbounded amount of predicate locks, so we apply a lossy but conservative method of tracking locks for committed transactions. If we run short of shared memory, we overflow to a new "pg_serial" SLRU pool. We don't currently allow Serializable transactions in Hot Standby mode. That would be hard, because even read-only transactions can cause anomalies that wouldn't otherwise occur. Serializable isolation mode now means the new fully serializable level. Repeatable Read gives you the old Snapshot Isolation level that we have always had. Kevin Grittner and Dan Ports, reviewed by Jeff Davis, Heikki Linnakangas and Anssi Kääriäinen
2011-02-07 22:46:51 +01:00
printf("Parsed test spec with %d sessions\n", testspec->nsessions);
Add support for blocked commands in isolationtester This enables us to test that blocking commands (such as foreign keys checks that conflict with some other lock) act as intended. The set of tests that this adds is pretty minimal, but can easily be extended by adding new specs. The intention is that this will serve as a basis for ensuring that further tweaks of locking implementation preserve (or improve) existing behavior. Author: Noah Misch
2011-07-12 22:35:09 +02:00
/*
* Establish connections to the database, one for each session and an
* extra for lock wait detection and global work.
*/
nconns = 1 + testspec->nsessions;
Use annotations to reduce instability of isolation-test results. We've long contended with isolation test results that aren't entirely stable. Some test scripts insert long delays to try to force stable results, which is not terribly desirable; but other erratic failure modes remain, causing unrepeatable buildfarm failures. I've spent a fair amount of time trying to solve this by improving the server-side support code, without much success: that way is fundamentally unable to cope with diffs that stem from chance ordering of arrival of messages from different server processes. We can improve matters on the client side, however, by annotating the test scripts themselves to show the desired reporting order of events that might occur in different orders. This patch adds three types of annotations to deal with (a) test steps that might or might not complete their waits before the isolationtester can see them waiting; (b) test steps in different sessions that can legitimately complete in either order; and (c) NOTIFY messages that might arrive before or after the completion of a step in another session. We might need more annotation types later, but this seems to be enough to deal with the instabilities we've seen in the buildfarm. It also lets us get rid of all the long delays that were previously used, cutting more than a minute off the runtime of the isolation tests. Back-patch to all supported branches, because the buildfarm instabilities affect all the branches, and because it seems desirable to keep isolationtester's capabilities the same across all branches to simplify possible future back-patching of tests. Discussion: https://postgr.es/m/327948.1623725828@sss.pgh.pa.us
2021-06-23 03:43:12 +02:00
conns = (IsoConnInfo *) pg_malloc0(nconns * sizeof(IsoConnInfo));
isolationtester: Use atexit() Replace exit_nicely() calls with standard exit() and register the cleanup actions using atexit(). Reviewed-by: Alvaro Herrera <alvherre@2ndquadrant.com> Discussion: https://www.postgresql.org/message-id/flat/ec4135ba-84e9-28bf-b584-0e78d47448d5@2ndquadrant.com/
2019-01-05 15:05:49 +01:00
atexit(disconnect_atexit);
Improve test coverage for LISTEN/NOTIFY. We had no actual end-to-end test of NOTIFY message delivery. In the core async.sql regression test, testing this is problematic because psql traditionally prints the PID of the sending backend, making the output unstable. We also have an isolation test script, but it likewise failed to prove that delivery worked, because isolationtester.c had no provisions for detecting/reporting NOTIFY messages. Hence, add such provisions to isolationtester.c, and extend async-notify.spec to include direct tests of basic NOTIFY functionality. I also added tests showing that NOTIFY de-duplicates messages normally, but not across subtransaction boundaries. (That's the historical behavior since we introduced subtransactions, though perhaps we ought to change it.) Patch by me, with suggestions/review by Andres Freund. Discussion: https://postgr.es/m/31304.1564246011@sss.pgh.pa.us
2019-07-28 18:02:27 +02:00
Add support for blocked commands in isolationtester This enables us to test that blocking commands (such as foreign keys checks that conflict with some other lock) act as intended. The set of tests that this adds is pretty minimal, but can easily be extended by adding new specs. The intention is that this will serve as a basis for ensuring that further tweaks of locking implementation preserve (or improve) existing behavior. Author: Noah Misch
2011-07-12 22:35:09 +02:00
for (i = 0; i < nconns; i++)
Implement genuine serializable isolation level. Until now, our Serializable mode has in fact been what's called Snapshot Isolation, which allows some anomalies that could not occur in any serialized ordering of the transactions. This patch fixes that using a method called Serializable Snapshot Isolation, based on research papers by Michael J. Cahill (see README-SSI for full references). In Serializable Snapshot Isolation, transactions run like they do in Snapshot Isolation, but a predicate lock manager observes the reads and writes performed and aborts transactions if it detects that an anomaly might occur. This method produces some false positives, ie. it sometimes aborts transactions even though there is no anomaly. To track reads we implement predicate locking, see storage/lmgr/predicate.c. Whenever a tuple is read, a predicate lock is acquired on the tuple. Shared memory is finite, so when a transaction takes many tuple-level locks on a page, the locks are promoted to a single page-level lock, and further to a single relation level lock if necessary. To lock key values with no matching tuple, a sequential scan always takes a relation-level lock, and an index scan acquires a page-level lock that covers the search key, whether or not there are any matching keys at the moment. A predicate lock doesn't conflict with any regular locks or with another predicate locks in the normal sense. They're only used by the predicate lock manager to detect the danger of anomalies. Only serializable transactions participate in predicate locking, so there should be no extra overhead for for other transactions. Predicate locks can't be released at commit, but must be remembered until all the transactions that overlapped with it have completed. That means that we need to remember an unbounded amount of predicate locks, so we apply a lossy but conservative method of tracking locks for committed transactions. If we run short of shared memory, we overflow to a new "pg_serial" SLRU pool. We don't currently allow Serializable transactions in Hot Standby mode. That would be hard, because even read-only transactions can cause anomalies that wouldn't otherwise occur. Serializable isolation mode now means the new fully serializable level. Repeatable Read gives you the old Snapshot Isolation level that we have always had. Kevin Grittner and Dan Ports, reviewed by Jeff Davis, Heikki Linnakangas and Anssi Kääriäinen
2011-02-07 22:46:51 +01:00
{
Use annotations to reduce instability of isolation-test results. We've long contended with isolation test results that aren't entirely stable. Some test scripts insert long delays to try to force stable results, which is not terribly desirable; but other erratic failure modes remain, causing unrepeatable buildfarm failures. I've spent a fair amount of time trying to solve this by improving the server-side support code, without much success: that way is fundamentally unable to cope with diffs that stem from chance ordering of arrival of messages from different server processes. We can improve matters on the client side, however, by annotating the test scripts themselves to show the desired reporting order of events that might occur in different orders. This patch adds three types of annotations to deal with (a) test steps that might or might not complete their waits before the isolationtester can see them waiting; (b) test steps in different sessions that can legitimately complete in either order; and (c) NOTIFY messages that might arrive before or after the completion of a step in another session. We might need more annotation types later, but this seems to be enough to deal with the instabilities we've seen in the buildfarm. It also lets us get rid of all the long delays that were previously used, cutting more than a minute off the runtime of the isolation tests. Back-patch to all supported branches, because the buildfarm instabilities affect all the branches, and because it seems desirable to keep isolationtester's capabilities the same across all branches to simplify possible future back-patching of tests. Discussion: https://postgr.es/m/327948.1623725828@sss.pgh.pa.us
2021-06-23 03:43:12 +02:00
if (i == 0)
conns[i].sessionname = "control connection";
else
conns[i].sessionname = testspec->sessions[i - 1]->name;
conns[i].conn = PQconnectdb(conninfo);
if (PQstatus(conns[i].conn) != CONNECTION_OK)
Implement genuine serializable isolation level. Until now, our Serializable mode has in fact been what's called Snapshot Isolation, which allows some anomalies that could not occur in any serialized ordering of the transactions. This patch fixes that using a method called Serializable Snapshot Isolation, based on research papers by Michael J. Cahill (see README-SSI for full references). In Serializable Snapshot Isolation, transactions run like they do in Snapshot Isolation, but a predicate lock manager observes the reads and writes performed and aborts transactions if it detects that an anomaly might occur. This method produces some false positives, ie. it sometimes aborts transactions even though there is no anomaly. To track reads we implement predicate locking, see storage/lmgr/predicate.c. Whenever a tuple is read, a predicate lock is acquired on the tuple. Shared memory is finite, so when a transaction takes many tuple-level locks on a page, the locks are promoted to a single page-level lock, and further to a single relation level lock if necessary. To lock key values with no matching tuple, a sequential scan always takes a relation-level lock, and an index scan acquires a page-level lock that covers the search key, whether or not there are any matching keys at the moment. A predicate lock doesn't conflict with any regular locks or with another predicate locks in the normal sense. They're only used by the predicate lock manager to detect the danger of anomalies. Only serializable transactions participate in predicate locking, so there should be no extra overhead for for other transactions. Predicate locks can't be released at commit, but must be remembered until all the transactions that overlapped with it have completed. That means that we need to remember an unbounded amount of predicate locks, so we apply a lossy but conservative method of tracking locks for committed transactions. If we run short of shared memory, we overflow to a new "pg_serial" SLRU pool. We don't currently allow Serializable transactions in Hot Standby mode. That would be hard, because even read-only transactions can cause anomalies that wouldn't otherwise occur. Serializable isolation mode now means the new fully serializable level. Repeatable Read gives you the old Snapshot Isolation level that we have always had. Kevin Grittner and Dan Ports, reviewed by Jeff Davis, Heikki Linnakangas and Anssi Kääriäinen
2011-02-07 22:46:51 +01:00
{
Avoid redundantly prefixing PQerrorMessage for a connection failure. libpq's error messages for connection failures pretty well stand on their own, especially since commits 52a10224e/27a48e5a1. Prefixing them with 'could not connect to database "foo"' or the like is just redundant, and perhaps even misleading if the specific database name isn't relevant to the failure. (When it is, we trust that the backend's error message will include the DB name.) Indeed, psql hasn't used any such prefix in a long time. So, make all our other programs and documentation examples agree with psql's practice. Discussion: https://postgr.es/m/1094524.1611266589@sss.pgh.pa.us
2021-01-22 22:52:31 +01:00
fprintf(stderr, "Connection %d failed: %s",
Use annotations to reduce instability of isolation-test results. We've long contended with isolation test results that aren't entirely stable. Some test scripts insert long delays to try to force stable results, which is not terribly desirable; but other erratic failure modes remain, causing unrepeatable buildfarm failures. I've spent a fair amount of time trying to solve this by improving the server-side support code, without much success: that way is fundamentally unable to cope with diffs that stem from chance ordering of arrival of messages from different server processes. We can improve matters on the client side, however, by annotating the test scripts themselves to show the desired reporting order of events that might occur in different orders. This patch adds three types of annotations to deal with (a) test steps that might or might not complete their waits before the isolationtester can see them waiting; (b) test steps in different sessions that can legitimately complete in either order; and (c) NOTIFY messages that might arrive before or after the completion of a step in another session. We might need more annotation types later, but this seems to be enough to deal with the instabilities we've seen in the buildfarm. It also lets us get rid of all the long delays that were previously used, cutting more than a minute off the runtime of the isolation tests. Back-patch to all supported branches, because the buildfarm instabilities affect all the branches, and because it seems desirable to keep isolationtester's capabilities the same across all branches to simplify possible future back-patching of tests. Discussion: https://postgr.es/m/327948.1623725828@sss.pgh.pa.us
2021-06-23 03:43:12 +02:00
i, PQerrorMessage(conns[i].conn));
isolationtester: Use atexit() Replace exit_nicely() calls with standard exit() and register the cleanup actions using atexit(). Reviewed-by: Alvaro Herrera <alvherre@2ndquadrant.com> Discussion: https://www.postgresql.org/message-id/flat/ec4135ba-84e9-28bf-b584-0e78d47448d5@2ndquadrant.com/
2019-01-05 15:05:49 +01:00
exit(1);
Implement genuine serializable isolation level. Until now, our Serializable mode has in fact been what's called Snapshot Isolation, which allows some anomalies that could not occur in any serialized ordering of the transactions. This patch fixes that using a method called Serializable Snapshot Isolation, based on research papers by Michael J. Cahill (see README-SSI for full references). In Serializable Snapshot Isolation, transactions run like they do in Snapshot Isolation, but a predicate lock manager observes the reads and writes performed and aborts transactions if it detects that an anomaly might occur. This method produces some false positives, ie. it sometimes aborts transactions even though there is no anomaly. To track reads we implement predicate locking, see storage/lmgr/predicate.c. Whenever a tuple is read, a predicate lock is acquired on the tuple. Shared memory is finite, so when a transaction takes many tuple-level locks on a page, the locks are promoted to a single page-level lock, and further to a single relation level lock if necessary. To lock key values with no matching tuple, a sequential scan always takes a relation-level lock, and an index scan acquires a page-level lock that covers the search key, whether or not there are any matching keys at the moment. A predicate lock doesn't conflict with any regular locks or with another predicate locks in the normal sense. They're only used by the predicate lock manager to detect the danger of anomalies. Only serializable transactions participate in predicate locking, so there should be no extra overhead for for other transactions. Predicate locks can't be released at commit, but must be remembered until all the transactions that overlapped with it have completed. That means that we need to remember an unbounded amount of predicate locks, so we apply a lossy but conservative method of tracking locks for committed transactions. If we run short of shared memory, we overflow to a new "pg_serial" SLRU pool. We don't currently allow Serializable transactions in Hot Standby mode. That would be hard, because even read-only transactions can cause anomalies that wouldn't otherwise occur. Serializable isolation mode now means the new fully serializable level. Repeatable Read gives you the old Snapshot Isolation level that we have always had. Kevin Grittner and Dan Ports, reviewed by Jeff Davis, Heikki Linnakangas and Anssi Kääriäinen
2011-02-07 22:46:51 +01:00
}
Indicate session name in isolationtester notices When a session under isolationtester produces printable notices (NOTICE, WARNING) we were just printing them unadorned, which can be confusing when debugging. Prefix them with the session name, which makes things clearer. Author: Álvaro Herrera Reviewed-by: Hari Babu Kommi Discussion: https://postgr.es/m/20181024213451.75nh3f3dctmcdbfq@alvherre.pgsql
2018-11-09 17:08:00 +01:00
/*
* Set up notice processors for the user-defined connections, so that
* messages can get printed prefixed with the session names. The
* control connection gets a "blackhole" processor instead (hides all
* messages).
*/
if (i != 0)
Use annotations to reduce instability of isolation-test results. We've long contended with isolation test results that aren't entirely stable. Some test scripts insert long delays to try to force stable results, which is not terribly desirable; but other erratic failure modes remain, causing unrepeatable buildfarm failures. I've spent a fair amount of time trying to solve this by improving the server-side support code, without much success: that way is fundamentally unable to cope with diffs that stem from chance ordering of arrival of messages from different server processes. We can improve matters on the client side, however, by annotating the test scripts themselves to show the desired reporting order of events that might occur in different orders. This patch adds three types of annotations to deal with (a) test steps that might or might not complete their waits before the isolationtester can see them waiting; (b) test steps in different sessions that can legitimately complete in either order; and (c) NOTIFY messages that might arrive before or after the completion of a step in another session. We might need more annotation types later, but this seems to be enough to deal with the instabilities we've seen in the buildfarm. It also lets us get rid of all the long delays that were previously used, cutting more than a minute off the runtime of the isolation tests. Back-patch to all supported branches, because the buildfarm instabilities affect all the branches, and because it seems desirable to keep isolationtester's capabilities the same across all branches to simplify possible future back-patching of tests. Discussion: https://postgr.es/m/327948.1623725828@sss.pgh.pa.us
2021-06-23 03:43:12 +02:00
PQsetNoticeProcessor(conns[i].conn,
Indicate session name in isolationtester notices When a session under isolationtester produces printable notices (NOTICE, WARNING) we were just printing them unadorned, which can be confusing when debugging. Prefix them with the session name, which makes things clearer. Author: Álvaro Herrera Reviewed-by: Hari Babu Kommi Discussion: https://postgr.es/m/20181024213451.75nh3f3dctmcdbfq@alvherre.pgsql
2018-11-09 17:08:00 +01:00
isotesterNoticeProcessor,
Use annotations to reduce instability of isolation-test results. We've long contended with isolation test results that aren't entirely stable. Some test scripts insert long delays to try to force stable results, which is not terribly desirable; but other erratic failure modes remain, causing unrepeatable buildfarm failures. I've spent a fair amount of time trying to solve this by improving the server-side support code, without much success: that way is fundamentally unable to cope with diffs that stem from chance ordering of arrival of messages from different server processes. We can improve matters on the client side, however, by annotating the test scripts themselves to show the desired reporting order of events that might occur in different orders. This patch adds three types of annotations to deal with (a) test steps that might or might not complete their waits before the isolationtester can see them waiting; (b) test steps in different sessions that can legitimately complete in either order; and (c) NOTIFY messages that might arrive before or after the completion of a step in another session. We might need more annotation types later, but this seems to be enough to deal with the instabilities we've seen in the buildfarm. It also lets us get rid of all the long delays that were previously used, cutting more than a minute off the runtime of the isolation tests. Back-patch to all supported branches, because the buildfarm instabilities affect all the branches, and because it seems desirable to keep isolationtester's capabilities the same across all branches to simplify possible future back-patching of tests. Discussion: https://postgr.es/m/327948.1623725828@sss.pgh.pa.us
2021-06-23 03:43:12 +02:00
(void *) &conns[i]);
Indicate session name in isolationtester notices When a session under isolationtester produces printable notices (NOTICE, WARNING) we were just printing them unadorned, which can be confusing when debugging. Prefix them with the session name, which makes things clearer. Author: Álvaro Herrera Reviewed-by: Hari Babu Kommi Discussion: https://postgr.es/m/20181024213451.75nh3f3dctmcdbfq@alvherre.pgsql
2018-11-09 17:08:00 +01:00
else
Use annotations to reduce instability of isolation-test results. We've long contended with isolation test results that aren't entirely stable. Some test scripts insert long delays to try to force stable results, which is not terribly desirable; but other erratic failure modes remain, causing unrepeatable buildfarm failures. I've spent a fair amount of time trying to solve this by improving the server-side support code, without much success: that way is fundamentally unable to cope with diffs that stem from chance ordering of arrival of messages from different server processes. We can improve matters on the client side, however, by annotating the test scripts themselves to show the desired reporting order of events that might occur in different orders. This patch adds three types of annotations to deal with (a) test steps that might or might not complete their waits before the isolationtester can see them waiting; (b) test steps in different sessions that can legitimately complete in either order; and (c) NOTIFY messages that might arrive before or after the completion of a step in another session. We might need more annotation types later, but this seems to be enough to deal with the instabilities we've seen in the buildfarm. It also lets us get rid of all the long delays that were previously used, cutting more than a minute off the runtime of the isolation tests. Back-patch to all supported branches, because the buildfarm instabilities affect all the branches, and because it seems desirable to keep isolationtester's capabilities the same across all branches to simplify possible future back-patching of tests. Discussion: https://postgr.es/m/327948.1623725828@sss.pgh.pa.us
2021-06-23 03:43:12 +02:00
PQsetNoticeProcessor(conns[i].conn,
Indicate session name in isolationtester notices When a session under isolationtester produces printable notices (NOTICE, WARNING) we were just printing them unadorned, which can be confusing when debugging. Prefix them with the session name, which makes things clearer. Author: Álvaro Herrera Reviewed-by: Hari Babu Kommi Discussion: https://postgr.es/m/20181024213451.75nh3f3dctmcdbfq@alvherre.pgsql
2018-11-09 17:08:00 +01:00
blackholeNoticeProcessor,
NULL);
Improve test coverage for LISTEN/NOTIFY. We had no actual end-to-end test of NOTIFY message delivery. In the core async.sql regression test, testing this is problematic because psql traditionally prints the PID of the sending backend, making the output unstable. We also have an isolation test script, but it likewise failed to prove that delivery worked, because isolationtester.c had no provisions for detecting/reporting NOTIFY messages. Hence, add such provisions to isolationtester.c, and extend async-notify.spec to include direct tests of basic NOTIFY functionality. I also added tests showing that NOTIFY de-duplicates messages normally, but not across subtransaction boundaries. (That's the historical behavior since we introduced subtransactions, though perhaps we ought to change it.) Patch by me, with suggestions/review by Andres Freund. Discussion: https://postgr.es/m/31304.1564246011@sss.pgh.pa.us
2019-07-28 18:02:27 +02:00
/* Save each connection's backend PID for subsequent use. */
Use annotations to reduce instability of isolation-test results. We've long contended with isolation test results that aren't entirely stable. Some test scripts insert long delays to try to force stable results, which is not terribly desirable; but other erratic failure modes remain, causing unrepeatable buildfarm failures. I've spent a fair amount of time trying to solve this by improving the server-side support code, without much success: that way is fundamentally unable to cope with diffs that stem from chance ordering of arrival of messages from different server processes. We can improve matters on the client side, however, by annotating the test scripts themselves to show the desired reporting order of events that might occur in different orders. This patch adds three types of annotations to deal with (a) test steps that might or might not complete their waits before the isolationtester can see them waiting; (b) test steps in different sessions that can legitimately complete in either order; and (c) NOTIFY messages that might arrive before or after the completion of a step in another session. We might need more annotation types later, but this seems to be enough to deal with the instabilities we've seen in the buildfarm. It also lets us get rid of all the long delays that were previously used, cutting more than a minute off the runtime of the isolation tests. Back-patch to all supported branches, because the buildfarm instabilities affect all the branches, and because it seems desirable to keep isolationtester's capabilities the same across all branches to simplify possible future back-patching of tests. Discussion: https://postgr.es/m/327948.1623725828@sss.pgh.pa.us
2021-06-23 03:43:12 +02:00
conns[i].backend_pid = PQbackendPID(conns[i].conn);
conns[i].backend_pid_str = psprintf("%d", conns[i].backend_pid);
Implement genuine serializable isolation level. Until now, our Serializable mode has in fact been what's called Snapshot Isolation, which allows some anomalies that could not occur in any serialized ordering of the transactions. This patch fixes that using a method called Serializable Snapshot Isolation, based on research papers by Michael J. Cahill (see README-SSI for full references). In Serializable Snapshot Isolation, transactions run like they do in Snapshot Isolation, but a predicate lock manager observes the reads and writes performed and aborts transactions if it detects that an anomaly might occur. This method produces some false positives, ie. it sometimes aborts transactions even though there is no anomaly. To track reads we implement predicate locking, see storage/lmgr/predicate.c. Whenever a tuple is read, a predicate lock is acquired on the tuple. Shared memory is finite, so when a transaction takes many tuple-level locks on a page, the locks are promoted to a single page-level lock, and further to a single relation level lock if necessary. To lock key values with no matching tuple, a sequential scan always takes a relation-level lock, and an index scan acquires a page-level lock that covers the search key, whether or not there are any matching keys at the moment. A predicate lock doesn't conflict with any regular locks or with another predicate locks in the normal sense. They're only used by the predicate lock manager to detect the danger of anomalies. Only serializable transactions participate in predicate locking, so there should be no extra overhead for for other transactions. Predicate locks can't be released at commit, but must be remembered until all the transactions that overlapped with it have completed. That means that we need to remember an unbounded amount of predicate locks, so we apply a lossy but conservative method of tracking locks for committed transactions. If we run short of shared memory, we overflow to a new "pg_serial" SLRU pool. We don't currently allow Serializable transactions in Hot Standby mode. That would be hard, because even read-only transactions can cause anomalies that wouldn't otherwise occur. Serializable isolation mode now means the new fully serializable level. Repeatable Read gives you the old Snapshot Isolation level that we have always had. Kevin Grittner and Dan Ports, reviewed by Jeff Davis, Heikki Linnakangas and Anssi Kääriäinen
2011-02-07 22:46:51 +01:00
}
Make isolationtester more robust on locked commands Noah Misch diagnosed the buildfarm problems in the isolation tests partly as failure to differentiate backends properly; the old code was using backend IDs, which is not good enough because a new backend might use an already used ID. Use PIDs instead. Also, the code was purposely careless about other concurrent activity, because it isn't expected; and in fact, it doesn't affect the vast majority of the time. However, it can be observed that autovacuum can block tables for long enough to cause sporadic failures. The new code accounts for that by ignoring locks held by processes not explicitly declared in our spec file. Author: Noah Misch
2011-07-19 20:22:42 +02:00
/*
* Build the query we'll use to detect lock contention among sessions in
* the test specification. Most of the time, we could get away with
* simply checking whether a session is waiting for *any* lock: we don't
* exactly expect concurrent use of test tables. However, autovacuum will
* occasionally take AccessExclusiveLock to truncate a table, and we must
* ignore that transient wait.
*/
initPQExpBuffer(&wait_query);
appendPQExpBufferStr(&wait_query,
Move isolationtester's is-blocked query into C code for speed. Commit 4deb41381 modified isolationtester's query to see whether a session is blocked to also check for waits occurring in GetSafeSnapshot. However, it did that in a way that enormously increased the query's runtime under CLOBBER_CACHE_ALWAYS, causing the buildfarm members that use that to run about four times slower than before, and in some cases fail entirely. To fix, push the entire logic into a dedicated backend function. This should actually reduce the CLOBBER_CACHE_ALWAYS runtime from what it was previously, though I've not checked that. In passing, expose a SQL function to check for safe-snapshot blockage, comparable to pg_blocking_pids. This is more or less free given the infrastructure built to solve the other problem, so we might as well. Thomas Munro Discussion: https://postgr.es/m/20170407165749.pstcakbc637opkax@alap3.anarazel.de
2017-04-10 16:26:54 +02:00
"SELECT pg_catalog.pg_isolation_test_session_is_blocked($1, '{");
Make isolationtester more robust on locked commands Noah Misch diagnosed the buildfarm problems in the isolation tests partly as failure to differentiate backends properly; the old code was using backend IDs, which is not good enough because a new backend might use an already used ID. Use PIDs instead. Also, the code was purposely careless about other concurrent activity, because it isn't expected; and in fact, it doesn't affect the vast majority of the time. However, it can be observed that autovacuum can block tables for long enough to cause sporadic failures. The new code accounts for that by ignoring locks held by processes not explicitly declared in our spec file. Author: Noah Misch
2011-07-19 20:22:42 +02:00
/* The spec syntax requires at least one session; assume that here. */
Use annotations to reduce instability of isolation-test results. We've long contended with isolation test results that aren't entirely stable. Some test scripts insert long delays to try to force stable results, which is not terribly desirable; but other erratic failure modes remain, causing unrepeatable buildfarm failures. I've spent a fair amount of time trying to solve this by improving the server-side support code, without much success: that way is fundamentally unable to cope with diffs that stem from chance ordering of arrival of messages from different server processes. We can improve matters on the client side, however, by annotating the test scripts themselves to show the desired reporting order of events that might occur in different orders. This patch adds three types of annotations to deal with (a) test steps that might or might not complete their waits before the isolationtester can see them waiting; (b) test steps in different sessions that can legitimately complete in either order; and (c) NOTIFY messages that might arrive before or after the completion of a step in another session. We might need more annotation types later, but this seems to be enough to deal with the instabilities we've seen in the buildfarm. It also lets us get rid of all the long delays that were previously used, cutting more than a minute off the runtime of the isolation tests. Back-patch to all supported branches, because the buildfarm instabilities affect all the branches, and because it seems desirable to keep isolationtester's capabilities the same across all branches to simplify possible future back-patching of tests. Discussion: https://postgr.es/m/327948.1623725828@sss.pgh.pa.us
2021-06-23 03:43:12 +02:00
appendPQExpBufferStr(&wait_query, conns[1].backend_pid_str);
Make isolationtester more robust on locked commands Noah Misch diagnosed the buildfarm problems in the isolation tests partly as failure to differentiate backends properly; the old code was using backend IDs, which is not good enough because a new backend might use an already used ID. Use PIDs instead. Also, the code was purposely careless about other concurrent activity, because it isn't expected; and in fact, it doesn't affect the vast majority of the time. However, it can be observed that autovacuum can block tables for long enough to cause sporadic failures. The new code accounts for that by ignoring locks held by processes not explicitly declared in our spec file. Author: Noah Misch
2011-07-19 20:22:42 +02:00
for (i = 2; i < nconns; i++)
Use annotations to reduce instability of isolation-test results. We've long contended with isolation test results that aren't entirely stable. Some test scripts insert long delays to try to force stable results, which is not terribly desirable; but other erratic failure modes remain, causing unrepeatable buildfarm failures. I've spent a fair amount of time trying to solve this by improving the server-side support code, without much success: that way is fundamentally unable to cope with diffs that stem from chance ordering of arrival of messages from different server processes. We can improve matters on the client side, however, by annotating the test scripts themselves to show the desired reporting order of events that might occur in different orders. This patch adds three types of annotations to deal with (a) test steps that might or might not complete their waits before the isolationtester can see them waiting; (b) test steps in different sessions that can legitimately complete in either order; and (c) NOTIFY messages that might arrive before or after the completion of a step in another session. We might need more annotation types later, but this seems to be enough to deal with the instabilities we've seen in the buildfarm. It also lets us get rid of all the long delays that were previously used, cutting more than a minute off the runtime of the isolation tests. Back-patch to all supported branches, because the buildfarm instabilities affect all the branches, and because it seems desirable to keep isolationtester's capabilities the same across all branches to simplify possible future back-patching of tests. Discussion: https://postgr.es/m/327948.1623725828@sss.pgh.pa.us
2021-06-23 03:43:12 +02:00
appendPQExpBuffer(&wait_query, ",%s", conns[i].backend_pid_str);
Move isolationtester's is-blocked query into C code for speed. Commit 4deb41381 modified isolationtester's query to see whether a session is blocked to also check for waits occurring in GetSafeSnapshot. However, it did that in a way that enormously increased the query's runtime under CLOBBER_CACHE_ALWAYS, causing the buildfarm members that use that to run about four times slower than before, and in some cases fail entirely. To fix, push the entire logic into a dedicated backend function. This should actually reduce the CLOBBER_CACHE_ALWAYS runtime from what it was previously, though I've not checked that. In passing, expose a SQL function to check for safe-snapshot blockage, comparable to pg_blocking_pids. This is more or less free given the infrastructure built to solve the other problem, so we might as well. Thomas Munro Discussion: https://postgr.es/m/20170407165749.pstcakbc637opkax@alap3.anarazel.de
2017-04-10 16:26:54 +02:00
appendPQExpBufferStr(&wait_query, "}')");
Add isolation test for SERIALIZABLE READ ONLY DEFERRABLE. This improves code coverage and lays a foundation for testing similar issues in a distributed environment. Author: Thomas Munro <thomas.munro@enterprisedb.com> Reviewed-by: Michael Paquier <michael.paquier@gmail.com>
2017-04-05 17:04:36 +02:00
Use annotations to reduce instability of isolation-test results. We've long contended with isolation test results that aren't entirely stable. Some test scripts insert long delays to try to force stable results, which is not terribly desirable; but other erratic failure modes remain, causing unrepeatable buildfarm failures. I've spent a fair amount of time trying to solve this by improving the server-side support code, without much success: that way is fundamentally unable to cope with diffs that stem from chance ordering of arrival of messages from different server processes. We can improve matters on the client side, however, by annotating the test scripts themselves to show the desired reporting order of events that might occur in different orders. This patch adds three types of annotations to deal with (a) test steps that might or might not complete their waits before the isolationtester can see them waiting; (b) test steps in different sessions that can legitimately complete in either order; and (c) NOTIFY messages that might arrive before or after the completion of a step in another session. We might need more annotation types later, but this seems to be enough to deal with the instabilities we've seen in the buildfarm. It also lets us get rid of all the long delays that were previously used, cutting more than a minute off the runtime of the isolation tests. Back-patch to all supported branches, because the buildfarm instabilities affect all the branches, and because it seems desirable to keep isolationtester's capabilities the same across all branches to simplify possible future back-patching of tests. Discussion: https://postgr.es/m/327948.1623725828@sss.pgh.pa.us
2021-06-23 03:43:12 +02:00
res = PQprepare(conns[0].conn, PREP_WAITING, wait_query.data, 0, NULL);
Add support for blocked commands in isolationtester This enables us to test that blocking commands (such as foreign keys checks that conflict with some other lock) act as intended. The set of tests that this adds is pretty minimal, but can easily be extended by adding new specs. The intention is that this will serve as a basis for ensuring that further tweaks of locking implementation preserve (or improve) existing behavior. Author: Noah Misch
2011-07-12 22:35:09 +02:00
if (PQresultStatus(res) != PGRES_COMMAND_OK)
{
fprintf(stderr, "prepare of lock wait query failed: %s",
Use annotations to reduce instability of isolation-test results. We've long contended with isolation test results that aren't entirely stable. Some test scripts insert long delays to try to force stable results, which is not terribly desirable; but other erratic failure modes remain, causing unrepeatable buildfarm failures. I've spent a fair amount of time trying to solve this by improving the server-side support code, without much success: that way is fundamentally unable to cope with diffs that stem from chance ordering of arrival of messages from different server processes. We can improve matters on the client side, however, by annotating the test scripts themselves to show the desired reporting order of events that might occur in different orders. This patch adds three types of annotations to deal with (a) test steps that might or might not complete their waits before the isolationtester can see them waiting; (b) test steps in different sessions that can legitimately complete in either order; and (c) NOTIFY messages that might arrive before or after the completion of a step in another session. We might need more annotation types later, but this seems to be enough to deal with the instabilities we've seen in the buildfarm. It also lets us get rid of all the long delays that were previously used, cutting more than a minute off the runtime of the isolation tests. Back-patch to all supported branches, because the buildfarm instabilities affect all the branches, and because it seems desirable to keep isolationtester's capabilities the same across all branches to simplify possible future back-patching of tests. Discussion: https://postgr.es/m/327948.1623725828@sss.pgh.pa.us
2021-06-23 03:43:12 +02:00
PQerrorMessage(conns[0].conn));
isolationtester: Use atexit() Replace exit_nicely() calls with standard exit() and register the cleanup actions using atexit(). Reviewed-by: Alvaro Herrera <alvherre@2ndquadrant.com> Discussion: https://www.postgresql.org/message-id/flat/ec4135ba-84e9-28bf-b584-0e78d47448d5@2ndquadrant.com/
2019-01-05 15:05:49 +01:00
exit(1);
Add support for blocked commands in isolationtester This enables us to test that blocking commands (such as foreign keys checks that conflict with some other lock) act as intended. The set of tests that this adds is pretty minimal, but can easily be extended by adding new specs. The intention is that this will serve as a basis for ensuring that further tweaks of locking implementation preserve (or improve) existing behavior. Author: Noah Misch
2011-07-12 22:35:09 +02:00
}
PQclear(res);
Make isolationtester more robust on locked commands Noah Misch diagnosed the buildfarm problems in the isolation tests partly as failure to differentiate backends properly; the old code was using backend IDs, which is not good enough because a new backend might use an already used ID. Use PIDs instead. Also, the code was purposely careless about other concurrent activity, because it isn't expected; and in fact, it doesn't affect the vast majority of the time. However, it can be observed that autovacuum can block tables for long enough to cause sporadic failures. The new code accounts for that by ignoring locks held by processes not explicitly declared in our spec file. Author: Noah Misch
2011-07-19 20:22:42 +02:00
termPQExpBuffer(&wait_query);
Add support for blocked commands in isolationtester This enables us to test that blocking commands (such as foreign keys checks that conflict with some other lock) act as intended. The set of tests that this adds is pretty minimal, but can easily be extended by adding new specs. The intention is that this will serve as a basis for ensuring that further tweaks of locking implementation preserve (or improve) existing behavior. Author: Noah Misch
2011-07-12 22:35:09 +02:00
Implement genuine serializable isolation level. Until now, our Serializable mode has in fact been what's called Snapshot Isolation, which allows some anomalies that could not occur in any serialized ordering of the transactions. This patch fixes that using a method called Serializable Snapshot Isolation, based on research papers by Michael J. Cahill (see README-SSI for full references). In Serializable Snapshot Isolation, transactions run like they do in Snapshot Isolation, but a predicate lock manager observes the reads and writes performed and aborts transactions if it detects that an anomaly might occur. This method produces some false positives, ie. it sometimes aborts transactions even though there is no anomaly. To track reads we implement predicate locking, see storage/lmgr/predicate.c. Whenever a tuple is read, a predicate lock is acquired on the tuple. Shared memory is finite, so when a transaction takes many tuple-level locks on a page, the locks are promoted to a single page-level lock, and further to a single relation level lock if necessary. To lock key values with no matching tuple, a sequential scan always takes a relation-level lock, and an index scan acquires a page-level lock that covers the search key, whether or not there are any matching keys at the moment. A predicate lock doesn't conflict with any regular locks or with another predicate locks in the normal sense. They're only used by the predicate lock manager to detect the danger of anomalies. Only serializable transactions participate in predicate locking, so there should be no extra overhead for for other transactions. Predicate locks can't be released at commit, but must be remembered until all the transactions that overlapped with it have completed. That means that we need to remember an unbounded amount of predicate locks, so we apply a lossy but conservative method of tracking locks for committed transactions. If we run short of shared memory, we overflow to a new "pg_serial" SLRU pool. We don't currently allow Serializable transactions in Hot Standby mode. That would be hard, because even read-only transactions can cause anomalies that wouldn't otherwise occur. Serializable isolation mode now means the new fully serializable level. Repeatable Read gives you the old Snapshot Isolation level that we have always had. Kevin Grittner and Dan Ports, reviewed by Jeff Davis, Heikki Linnakangas and Anssi Kääriäinen
2011-02-07 22:46:51 +01:00
/*
* Run the permutations specified in the spec, or all if none were
* explicitly specified.
*/
Implement a dry-run mode for isolationtester This mode prints out the permutations that would be run by the given spec file, in the same format used by the permutation lines in spec files. This helps in building new spec files. Author: Alexander Shulgin, with some tweaks by me
2011-11-03 18:06:08 +01:00
run_testspec(testspec);
Implement genuine serializable isolation level. Until now, our Serializable mode has in fact been what's called Snapshot Isolation, which allows some anomalies that could not occur in any serialized ordering of the transactions. This patch fixes that using a method called Serializable Snapshot Isolation, based on research papers by Michael J. Cahill (see README-SSI for full references). In Serializable Snapshot Isolation, transactions run like they do in Snapshot Isolation, but a predicate lock manager observes the reads and writes performed and aborts transactions if it detects that an anomaly might occur. This method produces some false positives, ie. it sometimes aborts transactions even though there is no anomaly. To track reads we implement predicate locking, see storage/lmgr/predicate.c. Whenever a tuple is read, a predicate lock is acquired on the tuple. Shared memory is finite, so when a transaction takes many tuple-level locks on a page, the locks are promoted to a single page-level lock, and further to a single relation level lock if necessary. To lock key values with no matching tuple, a sequential scan always takes a relation-level lock, and an index scan acquires a page-level lock that covers the search key, whether or not there are any matching keys at the moment. A predicate lock doesn't conflict with any regular locks or with another predicate locks in the normal sense. They're only used by the predicate lock manager to detect the danger of anomalies. Only serializable transactions participate in predicate locking, so there should be no extra overhead for for other transactions. Predicate locks can't be released at commit, but must be remembered until all the transactions that overlapped with it have completed. That means that we need to remember an unbounded amount of predicate locks, so we apply a lossy but conservative method of tracking locks for committed transactions. If we run short of shared memory, we overflow to a new "pg_serial" SLRU pool. We don't currently allow Serializable transactions in Hot Standby mode. That would be hard, because even read-only transactions can cause anomalies that wouldn't otherwise occur. Serializable isolation mode now means the new fully serializable level. Repeatable Read gives you the old Snapshot Isolation level that we have always had. Kevin Grittner and Dan Ports, reviewed by Jeff Davis, Heikki Linnakangas and Anssi Kääriäinen
2011-02-07 22:46:51 +01:00
return 0;
}
Use annotations to reduce instability of isolation-test results. We've long contended with isolation test results that aren't entirely stable. Some test scripts insert long delays to try to force stable results, which is not terribly desirable; but other erratic failure modes remain, causing unrepeatable buildfarm failures. I've spent a fair amount of time trying to solve this by improving the server-side support code, without much success: that way is fundamentally unable to cope with diffs that stem from chance ordering of arrival of messages from different server processes. We can improve matters on the client side, however, by annotating the test scripts themselves to show the desired reporting order of events that might occur in different orders. This patch adds three types of annotations to deal with (a) test steps that might or might not complete their waits before the isolationtester can see them waiting; (b) test steps in different sessions that can legitimately complete in either order; and (c) NOTIFY messages that might arrive before or after the completion of a step in another session. We might need more annotation types later, but this seems to be enough to deal with the instabilities we've seen in the buildfarm. It also lets us get rid of all the long delays that were previously used, cutting more than a minute off the runtime of the isolation tests. Back-patch to all supported branches, because the buildfarm instabilities affect all the branches, and because it seems desirable to keep isolationtester's capabilities the same across all branches to simplify possible future back-patching of tests. Discussion: https://postgr.es/m/327948.1623725828@sss.pgh.pa.us
2021-06-23 03:43:12 +02:00
/*
* Validity-check the test spec and fill in cross-links between nodes.
*/
static void
check_testspec(TestSpec *testspec)
{
int nallsteps;
Step **allsteps;
int i,
j,
k;
/* Create a sorted lookup table of all steps. */
nallsteps = 0;
for (i = 0; i < testspec->nsessions; i++)
nallsteps += testspec->sessions[i]->nsteps;
allsteps = pg_malloc(nallsteps * sizeof(Step *));
k = 0;
for (i = 0; i < testspec->nsessions; i++)
{
for (j = 0; j < testspec->sessions[i]->nsteps; j++)
allsteps[k++] = testspec->sessions[i]->steps[j];
}
qsort(allsteps, nallsteps, sizeof(Step *), step_qsort_cmp);
/* Verify that all step names are unique. */
for (i = 1; i < nallsteps; i++)
{
if (strcmp(allsteps[i - 1]->name,
allsteps[i]->name) == 0)
{
fprintf(stderr, "duplicate step name: %s\n",
allsteps[i]->name);
exit(1);
}
}
/* Set the session index fields in steps. */
for (i = 0; i < testspec->nsessions; i++)
{
Session *session = testspec->sessions[i];
for (j = 0; j < session->nsteps; j++)
session->steps[j]->session = i;
}
/*
* If we have manually-specified permutations, link PermutationSteps to
* Steps, and fill in blocker links.
*/
for (i = 0; i < testspec->npermutations; i++)
{
Permutation *p = testspec->permutations[i];
for (j = 0; j < p->nsteps; j++)
{
PermutationStep *pstep = p->steps[j];
Step **this = (Step **) bsearch(pstep->name,
allsteps,
nallsteps,
sizeof(Step *),
step_bsearch_cmp);
if (this == NULL)
{
fprintf(stderr, "undefined step \"%s\" specified in permutation\n",
pstep->name);
exit(1);
}
pstep->step = *this;
/* Mark the step used, for check below */
pstep->step->used = true;
}
/*
* Identify any blocker steps. We search only the current
* permutation, since steps not used there couldn't be concurrent.
* Note that it's OK to reference later permutation steps, so this
* can't be combined with the previous loop.
*/
for (j = 0; j < p->nsteps; j++)
{
PermutationStep *pstep = p->steps[j];
for (k = 0; k < pstep->nblockers; k++)
{
PermutationStepBlocker *blocker = pstep->blockers[k];
int n;
if (blocker->blocktype == PSB_ONCE)
continue; /* nothing to link to */
blocker->step = NULL;
for (n = 0; n < p->nsteps; n++)
{
PermutationStep *otherp = p->steps[n];
if (strcmp(otherp->name, blocker->stepname) == 0)
{
blocker->step = otherp->step;
break;
}
}
if (blocker->step == NULL)
{
fprintf(stderr, "undefined blocking step \"%s\" referenced in permutation step \"%s\"\n",
blocker->stepname, pstep->name);
exit(1);
}
/* can't block on completion of step of own session */
if (blocker->step->session == pstep->step->session)
{
fprintf(stderr, "permutation step \"%s\" cannot block on its own session\n",
pstep->name);
exit(1);
}
}
}
}
/*
* If we have manually-specified permutations, verify that all steps have
* been used, warning about anything defined but not used. We can skip
* this when using automatically-generated permutations.
*/
if (testspec->permutations)
{
for (i = 0; i < nallsteps; i++)
{
if (!allsteps[i]->used)
fprintf(stderr, "unused step name: %s\n", allsteps[i]->name);
}
}
Remove memory leaks in isolationtester. specscanner.l leaked a kilobyte of memory per token of the spec file. Apparently somebody thought that the introductory code block would be executed once; but it's once per yylex() call. A couple of functions in isolationtester.c leaked small amounts of memory due to not bothering to free one-time allocations. Might as well improve these so that valgrind gives this program a clean bill of health. Also get rid of an ugly static variable. Coverity complained about one of the one-time leaks, which led me to try valgrind'ing isolationtester, which led to discovery of the larger leak.
2021-06-27 18:45:04 +02:00
free(allsteps);
}
Implement genuine serializable isolation level. Until now, our Serializable mode has in fact been what's called Snapshot Isolation, which allows some anomalies that could not occur in any serialized ordering of the transactions. This patch fixes that using a method called Serializable Snapshot Isolation, based on research papers by Michael J. Cahill (see README-SSI for full references). In Serializable Snapshot Isolation, transactions run like they do in Snapshot Isolation, but a predicate lock manager observes the reads and writes performed and aborts transactions if it detects that an anomaly might occur. This method produces some false positives, ie. it sometimes aborts transactions even though there is no anomaly. To track reads we implement predicate locking, see storage/lmgr/predicate.c. Whenever a tuple is read, a predicate lock is acquired on the tuple. Shared memory is finite, so when a transaction takes many tuple-level locks on a page, the locks are promoted to a single page-level lock, and further to a single relation level lock if necessary. To lock key values with no matching tuple, a sequential scan always takes a relation-level lock, and an index scan acquires a page-level lock that covers the search key, whether or not there are any matching keys at the moment. A predicate lock doesn't conflict with any regular locks or with another predicate locks in the normal sense. They're only used by the predicate lock manager to detect the danger of anomalies. Only serializable transactions participate in predicate locking, so there should be no extra overhead for for other transactions. Predicate locks can't be released at commit, but must be remembered until all the transactions that overlapped with it have completed. That means that we need to remember an unbounded amount of predicate locks, so we apply a lossy but conservative method of tracking locks for committed transactions. If we run short of shared memory, we overflow to a new "pg_serial" SLRU pool. We don't currently allow Serializable transactions in Hot Standby mode. That would be hard, because even read-only transactions can cause anomalies that wouldn't otherwise occur. Serializable isolation mode now means the new fully serializable level. Repeatable Read gives you the old Snapshot Isolation level that we have always had. Kevin Grittner and Dan Ports, reviewed by Jeff Davis, Heikki Linnakangas and Anssi Kääriäinen
2011-02-07 22:46:51 +01:00
Implement a dry-run mode for isolationtester This mode prints out the permutations that would be run by the given spec file, in the same format used by the permutation lines in spec files. This helps in building new spec files. Author: Alexander Shulgin, with some tweaks by me
2011-11-03 18:06:08 +01:00
/*
* Run the permutations specified in the spec, or all if none were
* explicitly specified.
*/
static void
run_testspec(TestSpec *testspec)
{
if (testspec->permutations)
run_named_permutations(testspec);
else
run_all_permutations(testspec);
}
Implement genuine serializable isolation level. Until now, our Serializable mode has in fact been what's called Snapshot Isolation, which allows some anomalies that could not occur in any serialized ordering of the transactions. This patch fixes that using a method called Serializable Snapshot Isolation, based on research papers by Michael J. Cahill (see README-SSI for full references). In Serializable Snapshot Isolation, transactions run like they do in Snapshot Isolation, but a predicate lock manager observes the reads and writes performed and aborts transactions if it detects that an anomaly might occur. This method produces some false positives, ie. it sometimes aborts transactions even though there is no anomaly. To track reads we implement predicate locking, see storage/lmgr/predicate.c. Whenever a tuple is read, a predicate lock is acquired on the tuple. Shared memory is finite, so when a transaction takes many tuple-level locks on a page, the locks are promoted to a single page-level lock, and further to a single relation level lock if necessary. To lock key values with no matching tuple, a sequential scan always takes a relation-level lock, and an index scan acquires a page-level lock that covers the search key, whether or not there are any matching keys at the moment. A predicate lock doesn't conflict with any regular locks or with another predicate locks in the normal sense. They're only used by the predicate lock manager to detect the danger of anomalies. Only serializable transactions participate in predicate locking, so there should be no extra overhead for for other transactions. Predicate locks can't be released at commit, but must be remembered until all the transactions that overlapped with it have completed. That means that we need to remember an unbounded amount of predicate locks, so we apply a lossy but conservative method of tracking locks for committed transactions. If we run short of shared memory, we overflow to a new "pg_serial" SLRU pool. We don't currently allow Serializable transactions in Hot Standby mode. That would be hard, because even read-only transactions can cause anomalies that wouldn't otherwise occur. Serializable isolation mode now means the new fully serializable level. Repeatable Read gives you the old Snapshot Isolation level that we have always had. Kevin Grittner and Dan Ports, reviewed by Jeff Davis, Heikki Linnakangas and Anssi Kääriäinen
2011-02-07 22:46:51 +01:00
/*
* Run all permutations of the steps and sessions.
*/
static void
run_all_permutations(TestSpec *testspec)
{
int nsteps;
int i;
Use annotations to reduce instability of isolation-test results. We've long contended with isolation test results that aren't entirely stable. Some test scripts insert long delays to try to force stable results, which is not terribly desirable; but other erratic failure modes remain, causing unrepeatable buildfarm failures. I've spent a fair amount of time trying to solve this by improving the server-side support code, without much success: that way is fundamentally unable to cope with diffs that stem from chance ordering of arrival of messages from different server processes. We can improve matters on the client side, however, by annotating the test scripts themselves to show the desired reporting order of events that might occur in different orders. This patch adds three types of annotations to deal with (a) test steps that might or might not complete their waits before the isolationtester can see them waiting; (b) test steps in different sessions that can legitimately complete in either order; and (c) NOTIFY messages that might arrive before or after the completion of a step in another session. We might need more annotation types later, but this seems to be enough to deal with the instabilities we've seen in the buildfarm. It also lets us get rid of all the long delays that were previously used, cutting more than a minute off the runtime of the isolation tests. Back-patch to all supported branches, because the buildfarm instabilities affect all the branches, and because it seems desirable to keep isolationtester's capabilities the same across all branches to simplify possible future back-patching of tests. Discussion: https://postgr.es/m/327948.1623725828@sss.pgh.pa.us
2021-06-23 03:43:12 +02:00
PermutationStep *steps;
PermutationStep **stepptrs;
Remove memory leaks in isolationtester. specscanner.l leaked a kilobyte of memory per token of the spec file. Apparently somebody thought that the introductory code block would be executed once; but it's once per yylex() call. A couple of functions in isolationtester.c leaked small amounts of memory due to not bothering to free one-time allocations. Might as well improve these so that valgrind gives this program a clean bill of health. Also get rid of an ugly static variable. Coverity complained about one of the one-time leaks, which led me to try valgrind'ing isolationtester, which led to discovery of the larger leak.
2021-06-27 18:45:04 +02:00
int *piles;
Implement genuine serializable isolation level. Until now, our Serializable mode has in fact been what's called Snapshot Isolation, which allows some anomalies that could not occur in any serialized ordering of the transactions. This patch fixes that using a method called Serializable Snapshot Isolation, based on research papers by Michael J. Cahill (see README-SSI for full references). In Serializable Snapshot Isolation, transactions run like they do in Snapshot Isolation, but a predicate lock manager observes the reads and writes performed and aborts transactions if it detects that an anomaly might occur. This method produces some false positives, ie. it sometimes aborts transactions even though there is no anomaly. To track reads we implement predicate locking, see storage/lmgr/predicate.c. Whenever a tuple is read, a predicate lock is acquired on the tuple. Shared memory is finite, so when a transaction takes many tuple-level locks on a page, the locks are promoted to a single page-level lock, and further to a single relation level lock if necessary. To lock key values with no matching tuple, a sequential scan always takes a relation-level lock, and an index scan acquires a page-level lock that covers the search key, whether or not there are any matching keys at the moment. A predicate lock doesn't conflict with any regular locks or with another predicate locks in the normal sense. They're only used by the predicate lock manager to detect the danger of anomalies. Only serializable transactions participate in predicate locking, so there should be no extra overhead for for other transactions. Predicate locks can't be released at commit, but must be remembered until all the transactions that overlapped with it have completed. That means that we need to remember an unbounded amount of predicate locks, so we apply a lossy but conservative method of tracking locks for committed transactions. If we run short of shared memory, we overflow to a new "pg_serial" SLRU pool. We don't currently allow Serializable transactions in Hot Standby mode. That would be hard, because even read-only transactions can cause anomalies that wouldn't otherwise occur. Serializable isolation mode now means the new fully serializable level. Repeatable Read gives you the old Snapshot Isolation level that we have always had. Kevin Grittner and Dan Ports, reviewed by Jeff Davis, Heikki Linnakangas and Anssi Kääriäinen
2011-02-07 22:46:51 +01:00
/* Count the total number of steps in all sessions */
nsteps = 0;
for (i = 0; i < testspec->nsessions; i++)
nsteps += testspec->sessions[i]->nsteps;
Use annotations to reduce instability of isolation-test results. We've long contended with isolation test results that aren't entirely stable. Some test scripts insert long delays to try to force stable results, which is not terribly desirable; but other erratic failure modes remain, causing unrepeatable buildfarm failures. I've spent a fair amount of time trying to solve this by improving the server-side support code, without much success: that way is fundamentally unable to cope with diffs that stem from chance ordering of arrival of messages from different server processes. We can improve matters on the client side, however, by annotating the test scripts themselves to show the desired reporting order of events that might occur in different orders. This patch adds three types of annotations to deal with (a) test steps that might or might not complete their waits before the isolationtester can see them waiting; (b) test steps in different sessions that can legitimately complete in either order; and (c) NOTIFY messages that might arrive before or after the completion of a step in another session. We might need more annotation types later, but this seems to be enough to deal with the instabilities we've seen in the buildfarm. It also lets us get rid of all the long delays that were previously used, cutting more than a minute off the runtime of the isolation tests. Back-patch to all supported branches, because the buildfarm instabilities affect all the branches, and because it seems desirable to keep isolationtester's capabilities the same across all branches to simplify possible future back-patching of tests. Discussion: https://postgr.es/m/327948.1623725828@sss.pgh.pa.us
2021-06-23 03:43:12 +02:00
/* Create PermutationStep workspace array */
steps = (PermutationStep *) pg_malloc0(sizeof(PermutationStep) * nsteps);
stepptrs = (PermutationStep **) pg_malloc(sizeof(PermutationStep *) * nsteps);
for (i = 0; i < nsteps; i++)
stepptrs[i] = steps + i;
Implement genuine serializable isolation level. Until now, our Serializable mode has in fact been what's called Snapshot Isolation, which allows some anomalies that could not occur in any serialized ordering of the transactions. This patch fixes that using a method called Serializable Snapshot Isolation, based on research papers by Michael J. Cahill (see README-SSI for full references). In Serializable Snapshot Isolation, transactions run like they do in Snapshot Isolation, but a predicate lock manager observes the reads and writes performed and aborts transactions if it detects that an anomaly might occur. This method produces some false positives, ie. it sometimes aborts transactions even though there is no anomaly. To track reads we implement predicate locking, see storage/lmgr/predicate.c. Whenever a tuple is read, a predicate lock is acquired on the tuple. Shared memory is finite, so when a transaction takes many tuple-level locks on a page, the locks are promoted to a single page-level lock, and further to a single relation level lock if necessary. To lock key values with no matching tuple, a sequential scan always takes a relation-level lock, and an index scan acquires a page-level lock that covers the search key, whether or not there are any matching keys at the moment. A predicate lock doesn't conflict with any regular locks or with another predicate locks in the normal sense. They're only used by the predicate lock manager to detect the danger of anomalies. Only serializable transactions participate in predicate locking, so there should be no extra overhead for for other transactions. Predicate locks can't be released at commit, but must be remembered until all the transactions that overlapped with it have completed. That means that we need to remember an unbounded amount of predicate locks, so we apply a lossy but conservative method of tracking locks for committed transactions. If we run short of shared memory, we overflow to a new "pg_serial" SLRU pool. We don't currently allow Serializable transactions in Hot Standby mode. That would be hard, because even read-only transactions can cause anomalies that wouldn't otherwise occur. Serializable isolation mode now means the new fully serializable level. Repeatable Read gives you the old Snapshot Isolation level that we have always had. Kevin Grittner and Dan Ports, reviewed by Jeff Davis, Heikki Linnakangas and Anssi Kääriäinen
2011-02-07 22:46:51 +01:00
/*
* To generate the permutations, we conceptually put the steps of each
Fix typo
2011-09-21 18:26:19 +02:00
* session on a pile. To generate a permutation, we pick steps from the
Implement genuine serializable isolation level. Until now, our Serializable mode has in fact been what's called Snapshot Isolation, which allows some anomalies that could not occur in any serialized ordering of the transactions. This patch fixes that using a method called Serializable Snapshot Isolation, based on research papers by Michael J. Cahill (see README-SSI for full references). In Serializable Snapshot Isolation, transactions run like they do in Snapshot Isolation, but a predicate lock manager observes the reads and writes performed and aborts transactions if it detects that an anomaly might occur. This method produces some false positives, ie. it sometimes aborts transactions even though there is no anomaly. To track reads we implement predicate locking, see storage/lmgr/predicate.c. Whenever a tuple is read, a predicate lock is acquired on the tuple. Shared memory is finite, so when a transaction takes many tuple-level locks on a page, the locks are promoted to a single page-level lock, and further to a single relation level lock if necessary. To lock key values with no matching tuple, a sequential scan always takes a relation-level lock, and an index scan acquires a page-level lock that covers the search key, whether or not there are any matching keys at the moment. A predicate lock doesn't conflict with any regular locks or with another predicate locks in the normal sense. They're only used by the predicate lock manager to detect the danger of anomalies. Only serializable transactions participate in predicate locking, so there should be no extra overhead for for other transactions. Predicate locks can't be released at commit, but must be remembered until all the transactions that overlapped with it have completed. That means that we need to remember an unbounded amount of predicate locks, so we apply a lossy but conservative method of tracking locks for committed transactions. If we run short of shared memory, we overflow to a new "pg_serial" SLRU pool. We don't currently allow Serializable transactions in Hot Standby mode. That would be hard, because even read-only transactions can cause anomalies that wouldn't otherwise occur. Serializable isolation mode now means the new fully serializable level. Repeatable Read gives you the old Snapshot Isolation level that we have always had. Kevin Grittner and Dan Ports, reviewed by Jeff Davis, Heikki Linnakangas and Anssi Kääriäinen
2011-02-07 22:46:51 +01:00
* piles until all piles are empty. By picking steps from piles in
* different order, we get different permutations.
*
* A pile is actually just an integer which tells how many steps we've
* already picked from this pile.
*/
Fix a bunch of places that called malloc and friends with no NULL check. Where possible, use palloc or pg_malloc instead; otherwise, insert explicit NULL checks. Generally speaking, these are places where an actual OOM is quite unlikely, either because they're in client programs that don't allocate all that much, or they're very early in process startup so that we'd likely have had a fork() failure instead. Hence, no back-patch, even though this is nominally a bug fix. Michael Paquier, with some adjustments by me Discussion: <CAB7nPqRu07Ot6iht9i9KRfYLpDaF2ZuUv5y_+72uP23ZAGysRg@mail.gmail.com>
2016-08-31 00:22:43 +02:00
piles = pg_malloc(sizeof(int) * testspec->nsessions);
Implement genuine serializable isolation level. Until now, our Serializable mode has in fact been what's called Snapshot Isolation, which allows some anomalies that could not occur in any serialized ordering of the transactions. This patch fixes that using a method called Serializable Snapshot Isolation, based on research papers by Michael J. Cahill (see README-SSI for full references). In Serializable Snapshot Isolation, transactions run like they do in Snapshot Isolation, but a predicate lock manager observes the reads and writes performed and aborts transactions if it detects that an anomaly might occur. This method produces some false positives, ie. it sometimes aborts transactions even though there is no anomaly. To track reads we implement predicate locking, see storage/lmgr/predicate.c. Whenever a tuple is read, a predicate lock is acquired on the tuple. Shared memory is finite, so when a transaction takes many tuple-level locks on a page, the locks are promoted to a single page-level lock, and further to a single relation level lock if necessary. To lock key values with no matching tuple, a sequential scan always takes a relation-level lock, and an index scan acquires a page-level lock that covers the search key, whether or not there are any matching keys at the moment. A predicate lock doesn't conflict with any regular locks or with another predicate locks in the normal sense. They're only used by the predicate lock manager to detect the danger of anomalies. Only serializable transactions participate in predicate locking, so there should be no extra overhead for for other transactions. Predicate locks can't be released at commit, but must be remembered until all the transactions that overlapped with it have completed. That means that we need to remember an unbounded amount of predicate locks, so we apply a lossy but conservative method of tracking locks for committed transactions. If we run short of shared memory, we overflow to a new "pg_serial" SLRU pool. We don't currently allow Serializable transactions in Hot Standby mode. That would be hard, because even read-only transactions can cause anomalies that wouldn't otherwise occur. Serializable isolation mode now means the new fully serializable level. Repeatable Read gives you the old Snapshot Isolation level that we have always had. Kevin Grittner and Dan Ports, reviewed by Jeff Davis, Heikki Linnakangas and Anssi Kääriäinen
2011-02-07 22:46:51 +01:00
for (i = 0; i < testspec->nsessions; i++)
piles[i] = 0;
Remove memory leaks in isolationtester. specscanner.l leaked a kilobyte of memory per token of the spec file. Apparently somebody thought that the introductory code block would be executed once; but it's once per yylex() call. A couple of functions in isolationtester.c leaked small amounts of memory due to not bothering to free one-time allocations. Might as well improve these so that valgrind gives this program a clean bill of health. Also get rid of an ugly static variable. Coverity complained about one of the one-time leaks, which led me to try valgrind'ing isolationtester, which led to discovery of the larger leak.
2021-06-27 18:45:04 +02:00
run_all_permutations_recurse(testspec, piles, 0, stepptrs);
free(steps);
free(stepptrs);
free(piles);
Implement genuine serializable isolation level. Until now, our Serializable mode has in fact been what's called Snapshot Isolation, which allows some anomalies that could not occur in any serialized ordering of the transactions. This patch fixes that using a method called Serializable Snapshot Isolation, based on research papers by Michael J. Cahill (see README-SSI for full references). In Serializable Snapshot Isolation, transactions run like they do in Snapshot Isolation, but a predicate lock manager observes the reads and writes performed and aborts transactions if it detects that an anomaly might occur. This method produces some false positives, ie. it sometimes aborts transactions even though there is no anomaly. To track reads we implement predicate locking, see storage/lmgr/predicate.c. Whenever a tuple is read, a predicate lock is acquired on the tuple. Shared memory is finite, so when a transaction takes many tuple-level locks on a page, the locks are promoted to a single page-level lock, and further to a single relation level lock if necessary. To lock key values with no matching tuple, a sequential scan always takes a relation-level lock, and an index scan acquires a page-level lock that covers the search key, whether or not there are any matching keys at the moment. A predicate lock doesn't conflict with any regular locks or with another predicate locks in the normal sense. They're only used by the predicate lock manager to detect the danger of anomalies. Only serializable transactions participate in predicate locking, so there should be no extra overhead for for other transactions. Predicate locks can't be released at commit, but must be remembered until all the transactions that overlapped with it have completed. That means that we need to remember an unbounded amount of predicate locks, so we apply a lossy but conservative method of tracking locks for committed transactions. If we run short of shared memory, we overflow to a new "pg_serial" SLRU pool. We don't currently allow Serializable transactions in Hot Standby mode. That would be hard, because even read-only transactions can cause anomalies that wouldn't otherwise occur. Serializable isolation mode now means the new fully serializable level. Repeatable Read gives you the old Snapshot Isolation level that we have always had. Kevin Grittner and Dan Ports, reviewed by Jeff Davis, Heikki Linnakangas and Anssi Kääriäinen
2011-02-07 22:46:51 +01:00
}
static void
Remove memory leaks in isolationtester. specscanner.l leaked a kilobyte of memory per token of the spec file. Apparently somebody thought that the introductory code block would be executed once; but it's once per yylex() call. A couple of functions in isolationtester.c leaked small amounts of memory due to not bothering to free one-time allocations. Might as well improve these so that valgrind gives this program a clean bill of health. Also get rid of an ugly static variable. Coverity complained about one of the one-time leaks, which led me to try valgrind'ing isolationtester, which led to discovery of the larger leak.
2021-06-27 18:45:04 +02:00
run_all_permutations_recurse(TestSpec *testspec, int *piles,
int nsteps, PermutationStep **steps)
Implement genuine serializable isolation level. Until now, our Serializable mode has in fact been what's called Snapshot Isolation, which allows some anomalies that could not occur in any serialized ordering of the transactions. This patch fixes that using a method called Serializable Snapshot Isolation, based on research papers by Michael J. Cahill (see README-SSI for full references). In Serializable Snapshot Isolation, transactions run like they do in Snapshot Isolation, but a predicate lock manager observes the reads and writes performed and aborts transactions if it detects that an anomaly might occur. This method produces some false positives, ie. it sometimes aborts transactions even though there is no anomaly. To track reads we implement predicate locking, see storage/lmgr/predicate.c. Whenever a tuple is read, a predicate lock is acquired on the tuple. Shared memory is finite, so when a transaction takes many tuple-level locks on a page, the locks are promoted to a single page-level lock, and further to a single relation level lock if necessary. To lock key values with no matching tuple, a sequential scan always takes a relation-level lock, and an index scan acquires a page-level lock that covers the search key, whether or not there are any matching keys at the moment. A predicate lock doesn't conflict with any regular locks or with another predicate locks in the normal sense. They're only used by the predicate lock manager to detect the danger of anomalies. Only serializable transactions participate in predicate locking, so there should be no extra overhead for for other transactions. Predicate locks can't be released at commit, but must be remembered until all the transactions that overlapped with it have completed. That means that we need to remember an unbounded amount of predicate locks, so we apply a lossy but conservative method of tracking locks for committed transactions. If we run short of shared memory, we overflow to a new "pg_serial" SLRU pool. We don't currently allow Serializable transactions in Hot Standby mode. That would be hard, because even read-only transactions can cause anomalies that wouldn't otherwise occur. Serializable isolation mode now means the new fully serializable level. Repeatable Read gives you the old Snapshot Isolation level that we have always had. Kevin Grittner and Dan Ports, reviewed by Jeff Davis, Heikki Linnakangas and Anssi Kääriäinen
2011-02-07 22:46:51 +01:00
{
int i;
Use annotations to reduce instability of isolation-test results. We've long contended with isolation test results that aren't entirely stable. Some test scripts insert long delays to try to force stable results, which is not terribly desirable; but other erratic failure modes remain, causing unrepeatable buildfarm failures. I've spent a fair amount of time trying to solve this by improving the server-side support code, without much success: that way is fundamentally unable to cope with diffs that stem from chance ordering of arrival of messages from different server processes. We can improve matters on the client side, however, by annotating the test scripts themselves to show the desired reporting order of events that might occur in different orders. This patch adds three types of annotations to deal with (a) test steps that might or might not complete their waits before the isolationtester can see them waiting; (b) test steps in different sessions that can legitimately complete in either order; and (c) NOTIFY messages that might arrive before or after the completion of a step in another session. We might need more annotation types later, but this seems to be enough to deal with the instabilities we've seen in the buildfarm. It also lets us get rid of all the long delays that were previously used, cutting more than a minute off the runtime of the isolation tests. Back-patch to all supported branches, because the buildfarm instabilities affect all the branches, and because it seems desirable to keep isolationtester's capabilities the same across all branches to simplify possible future back-patching of tests. Discussion: https://postgr.es/m/327948.1623725828@sss.pgh.pa.us
2021-06-23 03:43:12 +02:00
bool found = false;
Implement genuine serializable isolation level. Until now, our Serializable mode has in fact been what's called Snapshot Isolation, which allows some anomalies that could not occur in any serialized ordering of the transactions. This patch fixes that using a method called Serializable Snapshot Isolation, based on research papers by Michael J. Cahill (see README-SSI for full references). In Serializable Snapshot Isolation, transactions run like they do in Snapshot Isolation, but a predicate lock manager observes the reads and writes performed and aborts transactions if it detects that an anomaly might occur. This method produces some false positives, ie. it sometimes aborts transactions even though there is no anomaly. To track reads we implement predicate locking, see storage/lmgr/predicate.c. Whenever a tuple is read, a predicate lock is acquired on the tuple. Shared memory is finite, so when a transaction takes many tuple-level locks on a page, the locks are promoted to a single page-level lock, and further to a single relation level lock if necessary. To lock key values with no matching tuple, a sequential scan always takes a relation-level lock, and an index scan acquires a page-level lock that covers the search key, whether or not there are any matching keys at the moment. A predicate lock doesn't conflict with any regular locks or with another predicate locks in the normal sense. They're only used by the predicate lock manager to detect the danger of anomalies. Only serializable transactions participate in predicate locking, so there should be no extra overhead for for other transactions. Predicate locks can't be released at commit, but must be remembered until all the transactions that overlapped with it have completed. That means that we need to remember an unbounded amount of predicate locks, so we apply a lossy but conservative method of tracking locks for committed transactions. If we run short of shared memory, we overflow to a new "pg_serial" SLRU pool. We don't currently allow Serializable transactions in Hot Standby mode. That would be hard, because even read-only transactions can cause anomalies that wouldn't otherwise occur. Serializable isolation mode now means the new fully serializable level. Repeatable Read gives you the old Snapshot Isolation level that we have always had. Kevin Grittner and Dan Ports, reviewed by Jeff Davis, Heikki Linnakangas and Anssi Kääriäinen
2011-02-07 22:46:51 +01:00
for (i = 0; i < testspec->nsessions; i++)
{
/* If there's any more steps in this pile, pick it and recurse */
if (piles[i] < testspec->sessions[i]->nsteps)
{
Use annotations to reduce instability of isolation-test results. We've long contended with isolation test results that aren't entirely stable. Some test scripts insert long delays to try to force stable results, which is not terribly desirable; but other erratic failure modes remain, causing unrepeatable buildfarm failures. I've spent a fair amount of time trying to solve this by improving the server-side support code, without much success: that way is fundamentally unable to cope with diffs that stem from chance ordering of arrival of messages from different server processes. We can improve matters on the client side, however, by annotating the test scripts themselves to show the desired reporting order of events that might occur in different orders. This patch adds three types of annotations to deal with (a) test steps that might or might not complete their waits before the isolationtester can see them waiting; (b) test steps in different sessions that can legitimately complete in either order; and (c) NOTIFY messages that might arrive before or after the completion of a step in another session. We might need more annotation types later, but this seems to be enough to deal with the instabilities we've seen in the buildfarm. It also lets us get rid of all the long delays that were previously used, cutting more than a minute off the runtime of the isolation tests. Back-patch to all supported branches, because the buildfarm instabilities affect all the branches, and because it seems desirable to keep isolationtester's capabilities the same across all branches to simplify possible future back-patching of tests. Discussion: https://postgr.es/m/327948.1623725828@sss.pgh.pa.us
2021-06-23 03:43:12 +02:00
Step *newstep = testspec->sessions[i]->steps[piles[i]];
/*
* These automatically-generated PermutationSteps never have
* blocker conditions. So we need only fill these fields, relying
* on run_all_permutations() to have zeroed the rest:
*/
steps[nsteps]->name = newstep->name;
steps[nsteps]->step = newstep;
Implement genuine serializable isolation level. Until now, our Serializable mode has in fact been what's called Snapshot Isolation, which allows some anomalies that could not occur in any serialized ordering of the transactions. This patch fixes that using a method called Serializable Snapshot Isolation, based on research papers by Michael J. Cahill (see README-SSI for full references). In Serializable Snapshot Isolation, transactions run like they do in Snapshot Isolation, but a predicate lock manager observes the reads and writes performed and aborts transactions if it detects that an anomaly might occur. This method produces some false positives, ie. it sometimes aborts transactions even though there is no anomaly. To track reads we implement predicate locking, see storage/lmgr/predicate.c. Whenever a tuple is read, a predicate lock is acquired on the tuple. Shared memory is finite, so when a transaction takes many tuple-level locks on a page, the locks are promoted to a single page-level lock, and further to a single relation level lock if necessary. To lock key values with no matching tuple, a sequential scan always takes a relation-level lock, and an index scan acquires a page-level lock that covers the search key, whether or not there are any matching keys at the moment. A predicate lock doesn't conflict with any regular locks or with another predicate locks in the normal sense. They're only used by the predicate lock manager to detect the danger of anomalies. Only serializable transactions participate in predicate locking, so there should be no extra overhead for for other transactions. Predicate locks can't be released at commit, but must be remembered until all the transactions that overlapped with it have completed. That means that we need to remember an unbounded amount of predicate locks, so we apply a lossy but conservative method of tracking locks for committed transactions. If we run short of shared memory, we overflow to a new "pg_serial" SLRU pool. We don't currently allow Serializable transactions in Hot Standby mode. That would be hard, because even read-only transactions can cause anomalies that wouldn't otherwise occur. Serializable isolation mode now means the new fully serializable level. Repeatable Read gives you the old Snapshot Isolation level that we have always had. Kevin Grittner and Dan Ports, reviewed by Jeff Davis, Heikki Linnakangas and Anssi Kääriäinen
2011-02-07 22:46:51 +01:00
piles[i]++;
Remove memory leaks in isolationtester. specscanner.l leaked a kilobyte of memory per token of the spec file. Apparently somebody thought that the introductory code block would be executed once; but it's once per yylex() call. A couple of functions in isolationtester.c leaked small amounts of memory due to not bothering to free one-time allocations. Might as well improve these so that valgrind gives this program a clean bill of health. Also get rid of an ugly static variable. Coverity complained about one of the one-time leaks, which led me to try valgrind'ing isolationtester, which led to discovery of the larger leak.
2021-06-27 18:45:04 +02:00
run_all_permutations_recurse(testspec, piles, nsteps + 1, steps);
Implement genuine serializable isolation level. Until now, our Serializable mode has in fact been what's called Snapshot Isolation, which allows some anomalies that could not occur in any serialized ordering of the transactions. This patch fixes that using a method called Serializable Snapshot Isolation, based on research papers by Michael J. Cahill (see README-SSI for full references). In Serializable Snapshot Isolation, transactions run like they do in Snapshot Isolation, but a predicate lock manager observes the reads and writes performed and aborts transactions if it detects that an anomaly might occur. This method produces some false positives, ie. it sometimes aborts transactions even though there is no anomaly. To track reads we implement predicate locking, see storage/lmgr/predicate.c. Whenever a tuple is read, a predicate lock is acquired on the tuple. Shared memory is finite, so when a transaction takes many tuple-level locks on a page, the locks are promoted to a single page-level lock, and further to a single relation level lock if necessary. To lock key values with no matching tuple, a sequential scan always takes a relation-level lock, and an index scan acquires a page-level lock that covers the search key, whether or not there are any matching keys at the moment. A predicate lock doesn't conflict with any regular locks or with another predicate locks in the normal sense. They're only used by the predicate lock manager to detect the danger of anomalies. Only serializable transactions participate in predicate locking, so there should be no extra overhead for for other transactions. Predicate locks can't be released at commit, but must be remembered until all the transactions that overlapped with it have completed. That means that we need to remember an unbounded amount of predicate locks, so we apply a lossy but conservative method of tracking locks for committed transactions. If we run short of shared memory, we overflow to a new "pg_serial" SLRU pool. We don't currently allow Serializable transactions in Hot Standby mode. That would be hard, because even read-only transactions can cause anomalies that wouldn't otherwise occur. Serializable isolation mode now means the new fully serializable level. Repeatable Read gives you the old Snapshot Isolation level that we have always had. Kevin Grittner and Dan Ports, reviewed by Jeff Davis, Heikki Linnakangas and Anssi Kääriäinen
2011-02-07 22:46:51 +01:00
piles[i]--;
Use annotations to reduce instability of isolation-test results. We've long contended with isolation test results that aren't entirely stable. Some test scripts insert long delays to try to force stable results, which is not terribly desirable; but other erratic failure modes remain, causing unrepeatable buildfarm failures. I've spent a fair amount of time trying to solve this by improving the server-side support code, without much success: that way is fundamentally unable to cope with diffs that stem from chance ordering of arrival of messages from different server processes. We can improve matters on the client side, however, by annotating the test scripts themselves to show the desired reporting order of events that might occur in different orders. This patch adds three types of annotations to deal with (a) test steps that might or might not complete their waits before the isolationtester can see them waiting; (b) test steps in different sessions that can legitimately complete in either order; and (c) NOTIFY messages that might arrive before or after the completion of a step in another session. We might need more annotation types later, but this seems to be enough to deal with the instabilities we've seen in the buildfarm. It also lets us get rid of all the long delays that were previously used, cutting more than a minute off the runtime of the isolation tests. Back-patch to all supported branches, because the buildfarm instabilities affect all the branches, and because it seems desirable to keep isolationtester's capabilities the same across all branches to simplify possible future back-patching of tests. Discussion: https://postgr.es/m/327948.1623725828@sss.pgh.pa.us
2021-06-23 03:43:12 +02:00
found = true;
Implement genuine serializable isolation level. Until now, our Serializable mode has in fact been what's called Snapshot Isolation, which allows some anomalies that could not occur in any serialized ordering of the transactions. This patch fixes that using a method called Serializable Snapshot Isolation, based on research papers by Michael J. Cahill (see README-SSI for full references). In Serializable Snapshot Isolation, transactions run like they do in Snapshot Isolation, but a predicate lock manager observes the reads and writes performed and aborts transactions if it detects that an anomaly might occur. This method produces some false positives, ie. it sometimes aborts transactions even though there is no anomaly. To track reads we implement predicate locking, see storage/lmgr/predicate.c. Whenever a tuple is read, a predicate lock is acquired on the tuple. Shared memory is finite, so when a transaction takes many tuple-level locks on a page, the locks are promoted to a single page-level lock, and further to a single relation level lock if necessary. To lock key values with no matching tuple, a sequential scan always takes a relation-level lock, and an index scan acquires a page-level lock that covers the search key, whether or not there are any matching keys at the moment. A predicate lock doesn't conflict with any regular locks or with another predicate locks in the normal sense. They're only used by the predicate lock manager to detect the danger of anomalies. Only serializable transactions participate in predicate locking, so there should be no extra overhead for for other transactions. Predicate locks can't be released at commit, but must be remembered until all the transactions that overlapped with it have completed. That means that we need to remember an unbounded amount of predicate locks, so we apply a lossy but conservative method of tracking locks for committed transactions. If we run short of shared memory, we overflow to a new "pg_serial" SLRU pool. We don't currently allow Serializable transactions in Hot Standby mode. That would be hard, because even read-only transactions can cause anomalies that wouldn't otherwise occur. Serializable isolation mode now means the new fully serializable level. Repeatable Read gives you the old Snapshot Isolation level that we have always had. Kevin Grittner and Dan Ports, reviewed by Jeff Davis, Heikki Linnakangas and Anssi Kääriäinen
2011-02-07 22:46:51 +01:00
}
}
/* If all the piles were empty, this permutation is completed. Run it */
if (!found)
run_permutation(testspec, nsteps, steps);
}
/*
* Run permutations given in the test spec
*/
static void
run_named_permutations(TestSpec *testspec)
{
Use annotations to reduce instability of isolation-test results. We've long contended with isolation test results that aren't entirely stable. Some test scripts insert long delays to try to force stable results, which is not terribly desirable; but other erratic failure modes remain, causing unrepeatable buildfarm failures. I've spent a fair amount of time trying to solve this by improving the server-side support code, without much success: that way is fundamentally unable to cope with diffs that stem from chance ordering of arrival of messages from different server processes. We can improve matters on the client side, however, by annotating the test scripts themselves to show the desired reporting order of events that might occur in different orders. This patch adds three types of annotations to deal with (a) test steps that might or might not complete their waits before the isolationtester can see them waiting; (b) test steps in different sessions that can legitimately complete in either order; and (c) NOTIFY messages that might arrive before or after the completion of a step in another session. We might need more annotation types later, but this seems to be enough to deal with the instabilities we've seen in the buildfarm. It also lets us get rid of all the long delays that were previously used, cutting more than a minute off the runtime of the isolation tests. Back-patch to all supported branches, because the buildfarm instabilities affect all the branches, and because it seems desirable to keep isolationtester's capabilities the same across all branches to simplify possible future back-patching of tests. Discussion: https://postgr.es/m/327948.1623725828@sss.pgh.pa.us
2021-06-23 03:43:12 +02:00
int i;
Implement genuine serializable isolation level. Until now, our Serializable mode has in fact been what's called Snapshot Isolation, which allows some anomalies that could not occur in any serialized ordering of the transactions. This patch fixes that using a method called Serializable Snapshot Isolation, based on research papers by Michael J. Cahill (see README-SSI for full references). In Serializable Snapshot Isolation, transactions run like they do in Snapshot Isolation, but a predicate lock manager observes the reads and writes performed and aborts transactions if it detects that an anomaly might occur. This method produces some false positives, ie. it sometimes aborts transactions even though there is no anomaly. To track reads we implement predicate locking, see storage/lmgr/predicate.c. Whenever a tuple is read, a predicate lock is acquired on the tuple. Shared memory is finite, so when a transaction takes many tuple-level locks on a page, the locks are promoted to a single page-level lock, and further to a single relation level lock if necessary. To lock key values with no matching tuple, a sequential scan always takes a relation-level lock, and an index scan acquires a page-level lock that covers the search key, whether or not there are any matching keys at the moment. A predicate lock doesn't conflict with any regular locks or with another predicate locks in the normal sense. They're only used by the predicate lock manager to detect the danger of anomalies. Only serializable transactions participate in predicate locking, so there should be no extra overhead for for other transactions. Predicate locks can't be released at commit, but must be remembered until all the transactions that overlapped with it have completed. That means that we need to remember an unbounded amount of predicate locks, so we apply a lossy but conservative method of tracking locks for committed transactions. If we run short of shared memory, we overflow to a new "pg_serial" SLRU pool. We don't currently allow Serializable transactions in Hot Standby mode. That would be hard, because even read-only transactions can cause anomalies that wouldn't otherwise occur. Serializable isolation mode now means the new fully serializable level. Repeatable Read gives you the old Snapshot Isolation level that we have always had. Kevin Grittner and Dan Ports, reviewed by Jeff Davis, Heikki Linnakangas and Anssi Kääriäinen
2011-02-07 22:46:51 +01:00
for (i = 0; i < testspec->npermutations; i++)
{
Permutation *p = testspec->permutations[i];
Add simple tests of EvalPlanQual using the isolationtester infrastructure. Much more could be done here, but at least now we have *some* automated test coverage of that mechanism. In particular this tests the writable-CTE case reported by Phil Sorber. In passing, remove isolationtester's arbitrary restriction on the number of steps in a permutation list. I used this so that a single spec file could be used to run several related test scenarios, but there are other possible reasons to want a step series that's not exactly a permutation. Improve documentation and fix a couple other nits as well.
2012-01-28 23:55:08 +01:00
Use annotations to reduce instability of isolation-test results. We've long contended with isolation test results that aren't entirely stable. Some test scripts insert long delays to try to force stable results, which is not terribly desirable; but other erratic failure modes remain, causing unrepeatable buildfarm failures. I've spent a fair amount of time trying to solve this by improving the server-side support code, without much success: that way is fundamentally unable to cope with diffs that stem from chance ordering of arrival of messages from different server processes. We can improve matters on the client side, however, by annotating the test scripts themselves to show the desired reporting order of events that might occur in different orders. This patch adds three types of annotations to deal with (a) test steps that might or might not complete their waits before the isolationtester can see them waiting; (b) test steps in different sessions that can legitimately complete in either order; and (c) NOTIFY messages that might arrive before or after the completion of a step in another session. We might need more annotation types later, but this seems to be enough to deal with the instabilities we've seen in the buildfarm. It also lets us get rid of all the long delays that were previously used, cutting more than a minute off the runtime of the isolation tests. Back-patch to all supported branches, because the buildfarm instabilities affect all the branches, and because it seems desirable to keep isolationtester's capabilities the same across all branches to simplify possible future back-patching of tests. Discussion: https://postgr.es/m/327948.1623725828@sss.pgh.pa.us
2021-06-23 03:43:12 +02:00
run_permutation(testspec, p->nsteps, p->steps);
Implement genuine serializable isolation level. Until now, our Serializable mode has in fact been what's called Snapshot Isolation, which allows some anomalies that could not occur in any serialized ordering of the transactions. This patch fixes that using a method called Serializable Snapshot Isolation, based on research papers by Michael J. Cahill (see README-SSI for full references). In Serializable Snapshot Isolation, transactions run like they do in Snapshot Isolation, but a predicate lock manager observes the reads and writes performed and aborts transactions if it detects that an anomaly might occur. This method produces some false positives, ie. it sometimes aborts transactions even though there is no anomaly. To track reads we implement predicate locking, see storage/lmgr/predicate.c. Whenever a tuple is read, a predicate lock is acquired on the tuple. Shared memory is finite, so when a transaction takes many tuple-level locks on a page, the locks are promoted to a single page-level lock, and further to a single relation level lock if necessary. To lock key values with no matching tuple, a sequential scan always takes a relation-level lock, and an index scan acquires a page-level lock that covers the search key, whether or not there are any matching keys at the moment. A predicate lock doesn't conflict with any regular locks or with another predicate locks in the normal sense. They're only used by the predicate lock manager to detect the danger of anomalies. Only serializable transactions participate in predicate locking, so there should be no extra overhead for for other transactions. Predicate locks can't be released at commit, but must be remembered until all the transactions that overlapped with it have completed. That means that we need to remember an unbounded amount of predicate locks, so we apply a lossy but conservative method of tracking locks for committed transactions. If we run short of shared memory, we overflow to a new "pg_serial" SLRU pool. We don't currently allow Serializable transactions in Hot Standby mode. That would be hard, because even read-only transactions can cause anomalies that wouldn't otherwise occur. Serializable isolation mode now means the new fully serializable level. Repeatable Read gives you the old Snapshot Isolation level that we have always had. Kevin Grittner and Dan Ports, reviewed by Jeff Davis, Heikki Linnakangas and Anssi Kääriäinen
2011-02-07 22:46:51 +01:00
}
}
static int
step_qsort_cmp(const void *a, const void *b)
{
Step *stepa = *((Step **) a);
Step *stepb = *((Step **) b);
return strcmp(stepa->name, stepb->name);
}
static int
step_bsearch_cmp(const void *a, const void *b)
{
char *stepname = (char *) a;
Step *step = *((Step **) b);
return strcmp(stepname, step->name);
}
/*
* Run one permutation
*/
static void
Use annotations to reduce instability of isolation-test results. We've long contended with isolation test results that aren't entirely stable. Some test scripts insert long delays to try to force stable results, which is not terribly desirable; but other erratic failure modes remain, causing unrepeatable buildfarm failures. I've spent a fair amount of time trying to solve this by improving the server-side support code, without much success: that way is fundamentally unable to cope with diffs that stem from chance ordering of arrival of messages from different server processes. We can improve matters on the client side, however, by annotating the test scripts themselves to show the desired reporting order of events that might occur in different orders. This patch adds three types of annotations to deal with (a) test steps that might or might not complete their waits before the isolationtester can see them waiting; (b) test steps in different sessions that can legitimately complete in either order; and (c) NOTIFY messages that might arrive before or after the completion of a step in another session. We might need more annotation types later, but this seems to be enough to deal with the instabilities we've seen in the buildfarm. It also lets us get rid of all the long delays that were previously used, cutting more than a minute off the runtime of the isolation tests. Back-patch to all supported branches, because the buildfarm instabilities affect all the branches, and because it seems desirable to keep isolationtester's capabilities the same across all branches to simplify possible future back-patching of tests. Discussion: https://postgr.es/m/327948.1623725828@sss.pgh.pa.us
2021-06-23 03:43:12 +02:00
run_permutation(TestSpec *testspec, int nsteps, PermutationStep **steps)
Implement genuine serializable isolation level. Until now, our Serializable mode has in fact been what's called Snapshot Isolation, which allows some anomalies that could not occur in any serialized ordering of the transactions. This patch fixes that using a method called Serializable Snapshot Isolation, based on research papers by Michael J. Cahill (see README-SSI for full references). In Serializable Snapshot Isolation, transactions run like they do in Snapshot Isolation, but a predicate lock manager observes the reads and writes performed and aborts transactions if it detects that an anomaly might occur. This method produces some false positives, ie. it sometimes aborts transactions even though there is no anomaly. To track reads we implement predicate locking, see storage/lmgr/predicate.c. Whenever a tuple is read, a predicate lock is acquired on the tuple. Shared memory is finite, so when a transaction takes many tuple-level locks on a page, the locks are promoted to a single page-level lock, and further to a single relation level lock if necessary. To lock key values with no matching tuple, a sequential scan always takes a relation-level lock, and an index scan acquires a page-level lock that covers the search key, whether or not there are any matching keys at the moment. A predicate lock doesn't conflict with any regular locks or with another predicate locks in the normal sense. They're only used by the predicate lock manager to detect the danger of anomalies. Only serializable transactions participate in predicate locking, so there should be no extra overhead for for other transactions. Predicate locks can't be released at commit, but must be remembered until all the transactions that overlapped with it have completed. That means that we need to remember an unbounded amount of predicate locks, so we apply a lossy but conservative method of tracking locks for committed transactions. If we run short of shared memory, we overflow to a new "pg_serial" SLRU pool. We don't currently allow Serializable transactions in Hot Standby mode. That would be hard, because even read-only transactions can cause anomalies that wouldn't otherwise occur. Serializable isolation mode now means the new fully serializable level. Repeatable Read gives you the old Snapshot Isolation level that we have always had. Kevin Grittner and Dan Ports, reviewed by Jeff Davis, Heikki Linnakangas and Anssi Kääriäinen
2011-02-07 22:46:51 +01:00
{
PGresult *res;
int i;
Modify the isolation tester so that multiple sessions can wait. This allows testing of deadlock scenarios. Scenarios that would previously have been considered invalid are now simply taken as a scenario in which more than one backend will wait.
2016-02-11 14:36:30 +01:00
int nwaiting = 0;
Use annotations to reduce instability of isolation-test results. We've long contended with isolation test results that aren't entirely stable. Some test scripts insert long delays to try to force stable results, which is not terribly desirable; but other erratic failure modes remain, causing unrepeatable buildfarm failures. I've spent a fair amount of time trying to solve this by improving the server-side support code, without much success: that way is fundamentally unable to cope with diffs that stem from chance ordering of arrival of messages from different server processes. We can improve matters on the client side, however, by annotating the test scripts themselves to show the desired reporting order of events that might occur in different orders. This patch adds three types of annotations to deal with (a) test steps that might or might not complete their waits before the isolationtester can see them waiting; (b) test steps in different sessions that can legitimately complete in either order; and (c) NOTIFY messages that might arrive before or after the completion of a step in another session. We might need more annotation types later, but this seems to be enough to deal with the instabilities we've seen in the buildfarm. It also lets us get rid of all the long delays that were previously used, cutting more than a minute off the runtime of the isolation tests. Back-patch to all supported branches, because the buildfarm instabilities affect all the branches, and because it seems desirable to keep isolationtester's capabilities the same across all branches to simplify possible future back-patching of tests. Discussion: https://postgr.es/m/327948.1623725828@sss.pgh.pa.us
2021-06-23 03:43:12 +02:00
PermutationStep **waiting;
Modify the isolation tester so that multiple sessions can wait. This allows testing of deadlock scenarios. Scenarios that would previously have been considered invalid are now simply taken as a scenario in which more than one backend will wait.
2016-02-11 14:36:30 +01:00
Use annotations to reduce instability of isolation-test results. We've long contended with isolation test results that aren't entirely stable. Some test scripts insert long delays to try to force stable results, which is not terribly desirable; but other erratic failure modes remain, causing unrepeatable buildfarm failures. I've spent a fair amount of time trying to solve this by improving the server-side support code, without much success: that way is fundamentally unable to cope with diffs that stem from chance ordering of arrival of messages from different server processes. We can improve matters on the client side, however, by annotating the test scripts themselves to show the desired reporting order of events that might occur in different orders. This patch adds three types of annotations to deal with (a) test steps that might or might not complete their waits before the isolationtester can see them waiting; (b) test steps in different sessions that can legitimately complete in either order; and (c) NOTIFY messages that might arrive before or after the completion of a step in another session. We might need more annotation types later, but this seems to be enough to deal with the instabilities we've seen in the buildfarm. It also lets us get rid of all the long delays that were previously used, cutting more than a minute off the runtime of the isolation tests. Back-patch to all supported branches, because the buildfarm instabilities affect all the branches, and because it seems desirable to keep isolationtester's capabilities the same across all branches to simplify possible future back-patching of tests. Discussion: https://postgr.es/m/327948.1623725828@sss.pgh.pa.us
2021-06-23 03:43:12 +02:00
waiting = pg_malloc(sizeof(PermutationStep *) * testspec->nsessions);
Code review for isolationtester changes. Fix a few oversights in 38f8bdcac4982215beb9f65a19debecaf22fd470: don't leak memory in run_permutation(), remember when we've issued a cancel rather than issuing another one every 10ms, fix some typos in comments.
2016-02-11 17:30:46 +01:00
Implement genuine serializable isolation level. Until now, our Serializable mode has in fact been what's called Snapshot Isolation, which allows some anomalies that could not occur in any serialized ordering of the transactions. This patch fixes that using a method called Serializable Snapshot Isolation, based on research papers by Michael J. Cahill (see README-SSI for full references). In Serializable Snapshot Isolation, transactions run like they do in Snapshot Isolation, but a predicate lock manager observes the reads and writes performed and aborts transactions if it detects that an anomaly might occur. This method produces some false positives, ie. it sometimes aborts transactions even though there is no anomaly. To track reads we implement predicate locking, see storage/lmgr/predicate.c. Whenever a tuple is read, a predicate lock is acquired on the tuple. Shared memory is finite, so when a transaction takes many tuple-level locks on a page, the locks are promoted to a single page-level lock, and further to a single relation level lock if necessary. To lock key values with no matching tuple, a sequential scan always takes a relation-level lock, and an index scan acquires a page-level lock that covers the search key, whether or not there are any matching keys at the moment. A predicate lock doesn't conflict with any regular locks or with another predicate locks in the normal sense. They're only used by the predicate lock manager to detect the danger of anomalies. Only serializable transactions participate in predicate locking, so there should be no extra overhead for for other transactions. Predicate locks can't be released at commit, but must be remembered until all the transactions that overlapped with it have completed. That means that we need to remember an unbounded amount of predicate locks, so we apply a lossy but conservative method of tracking locks for committed transactions. If we run short of shared memory, we overflow to a new "pg_serial" SLRU pool. We don't currently allow Serializable transactions in Hot Standby mode. That would be hard, because even read-only transactions can cause anomalies that wouldn't otherwise occur. Serializable isolation mode now means the new fully serializable level. Repeatable Read gives you the old Snapshot Isolation level that we have always had. Kevin Grittner and Dan Ports, reviewed by Jeff Davis, Heikki Linnakangas and Anssi Kääriäinen
2011-02-07 22:46:51 +01:00
printf("\nstarting permutation:");
for (i = 0; i < nsteps; i++)
printf(" %s", steps[i]->name);
printf("\n");
/* Perform setup */
Allow isolation tests to specify multiple setup blocks. Each setup block is run as a single PQexec submission, and some statements such as VACUUM cannot be combined with others in such a block. Backpatch to 9.2. Kevin Grittner and Tom Lane
2012-09-05 02:31:06 +02:00
for (i = 0; i < testspec->nsetupsqls; i++)
Implement genuine serializable isolation level. Until now, our Serializable mode has in fact been what's called Snapshot Isolation, which allows some anomalies that could not occur in any serialized ordering of the transactions. This patch fixes that using a method called Serializable Snapshot Isolation, based on research papers by Michael J. Cahill (see README-SSI for full references). In Serializable Snapshot Isolation, transactions run like they do in Snapshot Isolation, but a predicate lock manager observes the reads and writes performed and aborts transactions if it detects that an anomaly might occur. This method produces some false positives, ie. it sometimes aborts transactions even though there is no anomaly. To track reads we implement predicate locking, see storage/lmgr/predicate.c. Whenever a tuple is read, a predicate lock is acquired on the tuple. Shared memory is finite, so when a transaction takes many tuple-level locks on a page, the locks are promoted to a single page-level lock, and further to a single relation level lock if necessary. To lock key values with no matching tuple, a sequential scan always takes a relation-level lock, and an index scan acquires a page-level lock that covers the search key, whether or not there are any matching keys at the moment. A predicate lock doesn't conflict with any regular locks or with another predicate locks in the normal sense. They're only used by the predicate lock manager to detect the danger of anomalies. Only serializable transactions participate in predicate locking, so there should be no extra overhead for for other transactions. Predicate locks can't be released at commit, but must be remembered until all the transactions that overlapped with it have completed. That means that we need to remember an unbounded amount of predicate locks, so we apply a lossy but conservative method of tracking locks for committed transactions. If we run short of shared memory, we overflow to a new "pg_serial" SLRU pool. We don't currently allow Serializable transactions in Hot Standby mode. That would be hard, because even read-only transactions can cause anomalies that wouldn't otherwise occur. Serializable isolation mode now means the new fully serializable level. Repeatable Read gives you the old Snapshot Isolation level that we have always had. Kevin Grittner and Dan Ports, reviewed by Jeff Davis, Heikki Linnakangas and Anssi Kääriäinen
2011-02-07 22:46:51 +01:00
{
Use annotations to reduce instability of isolation-test results. We've long contended with isolation test results that aren't entirely stable. Some test scripts insert long delays to try to force stable results, which is not terribly desirable; but other erratic failure modes remain, causing unrepeatable buildfarm failures. I've spent a fair amount of time trying to solve this by improving the server-side support code, without much success: that way is fundamentally unable to cope with diffs that stem from chance ordering of arrival of messages from different server processes. We can improve matters on the client side, however, by annotating the test scripts themselves to show the desired reporting order of events that might occur in different orders. This patch adds three types of annotations to deal with (a) test steps that might or might not complete their waits before the isolationtester can see them waiting; (b) test steps in different sessions that can legitimately complete in either order; and (c) NOTIFY messages that might arrive before or after the completion of a step in another session. We might need more annotation types later, but this seems to be enough to deal with the instabilities we've seen in the buildfarm. It also lets us get rid of all the long delays that were previously used, cutting more than a minute off the runtime of the isolation tests. Back-patch to all supported branches, because the buildfarm instabilities affect all the branches, and because it seems desirable to keep isolationtester's capabilities the same across all branches to simplify possible future back-patching of tests. Discussion: https://postgr.es/m/327948.1623725828@sss.pgh.pa.us
2021-06-23 03:43:12 +02:00
res = PQexec(conns[0].conn, testspec->setupsqls[i]);
isolationtester: Allow tuples to be returned in more places Previously, isolationtester would forbid returning tuples in session-specific teardown (but not global teardown), as well as in global setup. Allow these places to return tuples, too.
2013-10-04 15:32:48 +02:00
if (PQresultStatus(res) == PGRES_TUPLES_OK)
{
printResultSet(res);
}
else if (PQresultStatus(res) != PGRES_COMMAND_OK)
Implement genuine serializable isolation level. Until now, our Serializable mode has in fact been what's called Snapshot Isolation, which allows some anomalies that could not occur in any serialized ordering of the transactions. This patch fixes that using a method called Serializable Snapshot Isolation, based on research papers by Michael J. Cahill (see README-SSI for full references). In Serializable Snapshot Isolation, transactions run like they do in Snapshot Isolation, but a predicate lock manager observes the reads and writes performed and aborts transactions if it detects that an anomaly might occur. This method produces some false positives, ie. it sometimes aborts transactions even though there is no anomaly. To track reads we implement predicate locking, see storage/lmgr/predicate.c. Whenever a tuple is read, a predicate lock is acquired on the tuple. Shared memory is finite, so when a transaction takes many tuple-level locks on a page, the locks are promoted to a single page-level lock, and further to a single relation level lock if necessary. To lock key values with no matching tuple, a sequential scan always takes a relation-level lock, and an index scan acquires a page-level lock that covers the search key, whether or not there are any matching keys at the moment. A predicate lock doesn't conflict with any regular locks or with another predicate locks in the normal sense. They're only used by the predicate lock manager to detect the danger of anomalies. Only serializable transactions participate in predicate locking, so there should be no extra overhead for for other transactions. Predicate locks can't be released at commit, but must be remembered until all the transactions that overlapped with it have completed. That means that we need to remember an unbounded amount of predicate locks, so we apply a lossy but conservative method of tracking locks for committed transactions. If we run short of shared memory, we overflow to a new "pg_serial" SLRU pool. We don't currently allow Serializable transactions in Hot Standby mode. That would be hard, because even read-only transactions can cause anomalies that wouldn't otherwise occur. Serializable isolation mode now means the new fully serializable level. Repeatable Read gives you the old Snapshot Isolation level that we have always had. Kevin Grittner and Dan Ports, reviewed by Jeff Davis, Heikki Linnakangas and Anssi Kääriäinen
2011-02-07 22:46:51 +01:00
{
Use annotations to reduce instability of isolation-test results. We've long contended with isolation test results that aren't entirely stable. Some test scripts insert long delays to try to force stable results, which is not terribly desirable; but other erratic failure modes remain, causing unrepeatable buildfarm failures. I've spent a fair amount of time trying to solve this by improving the server-side support code, without much success: that way is fundamentally unable to cope with diffs that stem from chance ordering of arrival of messages from different server processes. We can improve matters on the client side, however, by annotating the test scripts themselves to show the desired reporting order of events that might occur in different orders. This patch adds three types of annotations to deal with (a) test steps that might or might not complete their waits before the isolationtester can see them waiting; (b) test steps in different sessions that can legitimately complete in either order; and (c) NOTIFY messages that might arrive before or after the completion of a step in another session. We might need more annotation types later, but this seems to be enough to deal with the instabilities we've seen in the buildfarm. It also lets us get rid of all the long delays that were previously used, cutting more than a minute off the runtime of the isolation tests. Back-patch to all supported branches, because the buildfarm instabilities affect all the branches, and because it seems desirable to keep isolationtester's capabilities the same across all branches to simplify possible future back-patching of tests. Discussion: https://postgr.es/m/327948.1623725828@sss.pgh.pa.us
2021-06-23 03:43:12 +02:00
fprintf(stderr, "setup failed: %s", PQerrorMessage(conns[0].conn));
isolationtester: Use atexit() Replace exit_nicely() calls with standard exit() and register the cleanup actions using atexit(). Reviewed-by: Alvaro Herrera <alvherre@2ndquadrant.com> Discussion: https://www.postgresql.org/message-id/flat/ec4135ba-84e9-28bf-b584-0e78d47448d5@2ndquadrant.com/
2019-01-05 15:05:49 +01:00
exit(1);
Implement genuine serializable isolation level. Until now, our Serializable mode has in fact been what's called Snapshot Isolation, which allows some anomalies that could not occur in any serialized ordering of the transactions. This patch fixes that using a method called Serializable Snapshot Isolation, based on research papers by Michael J. Cahill (see README-SSI for full references). In Serializable Snapshot Isolation, transactions run like they do in Snapshot Isolation, but a predicate lock manager observes the reads and writes performed and aborts transactions if it detects that an anomaly might occur. This method produces some false positives, ie. it sometimes aborts transactions even though there is no anomaly. To track reads we implement predicate locking, see storage/lmgr/predicate.c. Whenever a tuple is read, a predicate lock is acquired on the tuple. Shared memory is finite, so when a transaction takes many tuple-level locks on a page, the locks are promoted to a single page-level lock, and further to a single relation level lock if necessary. To lock key values with no matching tuple, a sequential scan always takes a relation-level lock, and an index scan acquires a page-level lock that covers the search key, whether or not there are any matching keys at the moment. A predicate lock doesn't conflict with any regular locks or with another predicate locks in the normal sense. They're only used by the predicate lock manager to detect the danger of anomalies. Only serializable transactions participate in predicate locking, so there should be no extra overhead for for other transactions. Predicate locks can't be released at commit, but must be remembered until all the transactions that overlapped with it have completed. That means that we need to remember an unbounded amount of predicate locks, so we apply a lossy but conservative method of tracking locks for committed transactions. If we run short of shared memory, we overflow to a new "pg_serial" SLRU pool. We don't currently allow Serializable transactions in Hot Standby mode. That would be hard, because even read-only transactions can cause anomalies that wouldn't otherwise occur. Serializable isolation mode now means the new fully serializable level. Repeatable Read gives you the old Snapshot Isolation level that we have always had. Kevin Grittner and Dan Ports, reviewed by Jeff Davis, Heikki Linnakangas and Anssi Kääriäinen
2011-02-07 22:46:51 +01:00
}
PQclear(res);
}
/* Perform per-session setup */
for (i = 0; i < testspec->nsessions; i++)
{
if (testspec->sessions[i]->setupsql)
{
Use annotations to reduce instability of isolation-test results. We've long contended with isolation test results that aren't entirely stable. Some test scripts insert long delays to try to force stable results, which is not terribly desirable; but other erratic failure modes remain, causing unrepeatable buildfarm failures. I've spent a fair amount of time trying to solve this by improving the server-side support code, without much success: that way is fundamentally unable to cope with diffs that stem from chance ordering of arrival of messages from different server processes. We can improve matters on the client side, however, by annotating the test scripts themselves to show the desired reporting order of events that might occur in different orders. This patch adds three types of annotations to deal with (a) test steps that might or might not complete their waits before the isolationtester can see them waiting; (b) test steps in different sessions that can legitimately complete in either order; and (c) NOTIFY messages that might arrive before or after the completion of a step in another session. We might need more annotation types later, but this seems to be enough to deal with the instabilities we've seen in the buildfarm. It also lets us get rid of all the long delays that were previously used, cutting more than a minute off the runtime of the isolation tests. Back-patch to all supported branches, because the buildfarm instabilities affect all the branches, and because it seems desirable to keep isolationtester's capabilities the same across all branches to simplify possible future back-patching of tests. Discussion: https://postgr.es/m/327948.1623725828@sss.pgh.pa.us
2021-06-23 03:43:12 +02:00
res = PQexec(conns[i + 1].conn, testspec->sessions[i]->setupsql);
Add an SSI regression test that tests all interesting permutations in the order of begin, prepare, and commit of three concurrent transactions that have conflicts between them. The test runs for a quite long time, and the expected output file is huge, but this test caught some serious bugs during development, so seems worthwhile to keep. The test uses prepared transactions, so it fails if the server has max_prepared_transactions=0. Because of that, it's marked as "ignore" in the schedule file. Dan Ports
2011-08-18 12:36:37 +02:00
if (PQresultStatus(res) == PGRES_TUPLES_OK)
{
printResultSet(res);
}
else if (PQresultStatus(res) != PGRES_COMMAND_OK)
Implement genuine serializable isolation level. Until now, our Serializable mode has in fact been what's called Snapshot Isolation, which allows some anomalies that could not occur in any serialized ordering of the transactions. This patch fixes that using a method called Serializable Snapshot Isolation, based on research papers by Michael J. Cahill (see README-SSI for full references). In Serializable Snapshot Isolation, transactions run like they do in Snapshot Isolation, but a predicate lock manager observes the reads and writes performed and aborts transactions if it detects that an anomaly might occur. This method produces some false positives, ie. it sometimes aborts transactions even though there is no anomaly. To track reads we implement predicate locking, see storage/lmgr/predicate.c. Whenever a tuple is read, a predicate lock is acquired on the tuple. Shared memory is finite, so when a transaction takes many tuple-level locks on a page, the locks are promoted to a single page-level lock, and further to a single relation level lock if necessary. To lock key values with no matching tuple, a sequential scan always takes a relation-level lock, and an index scan acquires a page-level lock that covers the search key, whether or not there are any matching keys at the moment. A predicate lock doesn't conflict with any regular locks or with another predicate locks in the normal sense. They're only used by the predicate lock manager to detect the danger of anomalies. Only serializable transactions participate in predicate locking, so there should be no extra overhead for for other transactions. Predicate locks can't be released at commit, but must be remembered until all the transactions that overlapped with it have completed. That means that we need to remember an unbounded amount of predicate locks, so we apply a lossy but conservative method of tracking locks for committed transactions. If we run short of shared memory, we overflow to a new "pg_serial" SLRU pool. We don't currently allow Serializable transactions in Hot Standby mode. That would be hard, because even read-only transactions can cause anomalies that wouldn't otherwise occur. Serializable isolation mode now means the new fully serializable level. Repeatable Read gives you the old Snapshot Isolation level that we have always had. Kevin Grittner and Dan Ports, reviewed by Jeff Davis, Heikki Linnakangas and Anssi Kääriäinen
2011-02-07 22:46:51 +01:00
{
fprintf(stderr, "setup of session %s failed: %s",
Use annotations to reduce instability of isolation-test results. We've long contended with isolation test results that aren't entirely stable. Some test scripts insert long delays to try to force stable results, which is not terribly desirable; but other erratic failure modes remain, causing unrepeatable buildfarm failures. I've spent a fair amount of time trying to solve this by improving the server-side support code, without much success: that way is fundamentally unable to cope with diffs that stem from chance ordering of arrival of messages from different server processes. We can improve matters on the client side, however, by annotating the test scripts themselves to show the desired reporting order of events that might occur in different orders. This patch adds three types of annotations to deal with (a) test steps that might or might not complete their waits before the isolationtester can see them waiting; (b) test steps in different sessions that can legitimately complete in either order; and (c) NOTIFY messages that might arrive before or after the completion of a step in another session. We might need more annotation types later, but this seems to be enough to deal with the instabilities we've seen in the buildfarm. It also lets us get rid of all the long delays that were previously used, cutting more than a minute off the runtime of the isolation tests. Back-patch to all supported branches, because the buildfarm instabilities affect all the branches, and because it seems desirable to keep isolationtester's capabilities the same across all branches to simplify possible future back-patching of tests. Discussion: https://postgr.es/m/327948.1623725828@sss.pgh.pa.us
2021-06-23 03:43:12 +02:00
conns[i + 1].sessionname,
PQerrorMessage(conns[i + 1].conn));
isolationtester: Use atexit() Replace exit_nicely() calls with standard exit() and register the cleanup actions using atexit(). Reviewed-by: Alvaro Herrera <alvherre@2ndquadrant.com> Discussion: https://www.postgresql.org/message-id/flat/ec4135ba-84e9-28bf-b584-0e78d47448d5@2ndquadrant.com/
2019-01-05 15:05:49 +01:00
exit(1);
Implement genuine serializable isolation level. Until now, our Serializable mode has in fact been what's called Snapshot Isolation, which allows some anomalies that could not occur in any serialized ordering of the transactions. This patch fixes that using a method called Serializable Snapshot Isolation, based on research papers by Michael J. Cahill (see README-SSI for full references). In Serializable Snapshot Isolation, transactions run like they do in Snapshot Isolation, but a predicate lock manager observes the reads and writes performed and aborts transactions if it detects that an anomaly might occur. This method produces some false positives, ie. it sometimes aborts transactions even though there is no anomaly. To track reads we implement predicate locking, see storage/lmgr/predicate.c. Whenever a tuple is read, a predicate lock is acquired on the tuple. Shared memory is finite, so when a transaction takes many tuple-level locks on a page, the locks are promoted to a single page-level lock, and further to a single relation level lock if necessary. To lock key values with no matching tuple, a sequential scan always takes a relation-level lock, and an index scan acquires a page-level lock that covers the search key, whether or not there are any matching keys at the moment. A predicate lock doesn't conflict with any regular locks or with another predicate locks in the normal sense. They're only used by the predicate lock manager to detect the danger of anomalies. Only serializable transactions participate in predicate locking, so there should be no extra overhead for for other transactions. Predicate locks can't be released at commit, but must be remembered until all the transactions that overlapped with it have completed. That means that we need to remember an unbounded amount of predicate locks, so we apply a lossy but conservative method of tracking locks for committed transactions. If we run short of shared memory, we overflow to a new "pg_serial" SLRU pool. We don't currently allow Serializable transactions in Hot Standby mode. That would be hard, because even read-only transactions can cause anomalies that wouldn't otherwise occur. Serializable isolation mode now means the new fully serializable level. Repeatable Read gives you the old Snapshot Isolation level that we have always had. Kevin Grittner and Dan Ports, reviewed by Jeff Davis, Heikki Linnakangas and Anssi Kääriäinen
2011-02-07 22:46:51 +01:00
}
PQclear(res);
}
}
/* Perform steps */
for (i = 0; i < nsteps; i++)
{
Use annotations to reduce instability of isolation-test results. We've long contended with isolation test results that aren't entirely stable. Some test scripts insert long delays to try to force stable results, which is not terribly desirable; but other erratic failure modes remain, causing unrepeatable buildfarm failures. I've spent a fair amount of time trying to solve this by improving the server-side support code, without much success: that way is fundamentally unable to cope with diffs that stem from chance ordering of arrival of messages from different server processes. We can improve matters on the client side, however, by annotating the test scripts themselves to show the desired reporting order of events that might occur in different orders. This patch adds three types of annotations to deal with (a) test steps that might or might not complete their waits before the isolationtester can see them waiting; (b) test steps in different sessions that can legitimately complete in either order; and (c) NOTIFY messages that might arrive before or after the completion of a step in another session. We might need more annotation types later, but this seems to be enough to deal with the instabilities we've seen in the buildfarm. It also lets us get rid of all the long delays that were previously used, cutting more than a minute off the runtime of the isolation tests. Back-patch to all supported branches, because the buildfarm instabilities affect all the branches, and because it seems desirable to keep isolationtester's capabilities the same across all branches to simplify possible future back-patching of tests. Discussion: https://postgr.es/m/327948.1623725828@sss.pgh.pa.us
2021-06-23 03:43:12 +02:00
PermutationStep *pstep = steps[i];
Step *step = pstep->step;
IsoConnInfo *iconn = &conns[1 + step->session];
PGconn *conn = iconn->conn;
Modify the isolation tester so that multiple sessions can wait. This allows testing of deadlock scenarios. Scenarios that would previously have been considered invalid are now simply taken as a scenario in which more than one backend will wait.
2016-02-11 14:36:30 +01:00
bool mustwait;
Use annotations to reduce instability of isolation-test results. We've long contended with isolation test results that aren't entirely stable. Some test scripts insert long delays to try to force stable results, which is not terribly desirable; but other erratic failure modes remain, causing unrepeatable buildfarm failures. I've spent a fair amount of time trying to solve this by improving the server-side support code, without much success: that way is fundamentally unable to cope with diffs that stem from chance ordering of arrival of messages from different server processes. We can improve matters on the client side, however, by annotating the test scripts themselves to show the desired reporting order of events that might occur in different orders. This patch adds three types of annotations to deal with (a) test steps that might or might not complete their waits before the isolationtester can see them waiting; (b) test steps in different sessions that can legitimately complete in either order; and (c) NOTIFY messages that might arrive before or after the completion of a step in another session. We might need more annotation types later, but this seems to be enough to deal with the instabilities we've seen in the buildfarm. It also lets us get rid of all the long delays that were previously used, cutting more than a minute off the runtime of the isolation tests. Back-patch to all supported branches, because the buildfarm instabilities affect all the branches, and because it seems desirable to keep isolationtester's capabilities the same across all branches to simplify possible future back-patching of tests. Discussion: https://postgr.es/m/327948.1623725828@sss.pgh.pa.us
2021-06-23 03:43:12 +02:00
int j;
pgindent run before PG 9.1 beta 1.
2011-04-10 17:42:00 +02:00
Modify the isolation tester so that multiple sessions can wait. This allows testing of deadlock scenarios. Scenarios that would previously have been considered invalid are now simply taken as a scenario in which more than one backend will wait.
2016-02-11 14:36:30 +01:00
/*
Re-pgindent isolationtester.c. Need to do some more hacking on this, and got annoyed that it's not indent clean.
2016-02-12 19:36:13 +01:00
* Check whether the session that needs to perform the next step is
* still blocked on an earlier step. If so, wait for it to finish.
Modify the isolation tester so that multiple sessions can wait. This allows testing of deadlock scenarios. Scenarios that would previously have been considered invalid are now simply taken as a scenario in which more than one backend will wait.
2016-02-11 14:36:30 +01:00
*/
Use annotations to reduce instability of isolation-test results. We've long contended with isolation test results that aren't entirely stable. Some test scripts insert long delays to try to force stable results, which is not terribly desirable; but other erratic failure modes remain, causing unrepeatable buildfarm failures. I've spent a fair amount of time trying to solve this by improving the server-side support code, without much success: that way is fundamentally unable to cope with diffs that stem from chance ordering of arrival of messages from different server processes. We can improve matters on the client side, however, by annotating the test scripts themselves to show the desired reporting order of events that might occur in different orders. This patch adds three types of annotations to deal with (a) test steps that might or might not complete their waits before the isolationtester can see them waiting; (b) test steps in different sessions that can legitimately complete in either order; and (c) NOTIFY messages that might arrive before or after the completion of a step in another session. We might need more annotation types later, but this seems to be enough to deal with the instabilities we've seen in the buildfarm. It also lets us get rid of all the long delays that were previously used, cutting more than a minute off the runtime of the isolation tests. Back-patch to all supported branches, because the buildfarm instabilities affect all the branches, and because it seems desirable to keep isolationtester's capabilities the same across all branches to simplify possible future back-patching of tests. Discussion: https://postgr.es/m/327948.1623725828@sss.pgh.pa.us
2021-06-23 03:43:12 +02:00
if (iconn->active_step != NULL)
Detect invalid permutations in isolationtester isolationtester is now able to continue running other permutations when it detects that one of them is invalid, which is useful during initial development of spec files. Author: Alexander Shulgin
2012-01-14 23:36:39 +01:00
{
Use annotations to reduce instability of isolation-test results. We've long contended with isolation test results that aren't entirely stable. Some test scripts insert long delays to try to force stable results, which is not terribly desirable; but other erratic failure modes remain, causing unrepeatable buildfarm failures. I've spent a fair amount of time trying to solve this by improving the server-side support code, without much success: that way is fundamentally unable to cope with diffs that stem from chance ordering of arrival of messages from different server processes. We can improve matters on the client side, however, by annotating the test scripts themselves to show the desired reporting order of events that might occur in different orders. This patch adds three types of annotations to deal with (a) test steps that might or might not complete their waits before the isolationtester can see them waiting; (b) test steps in different sessions that can legitimately complete in either order; and (c) NOTIFY messages that might arrive before or after the completion of a step in another session. We might need more annotation types later, but this seems to be enough to deal with the instabilities we've seen in the buildfarm. It also lets us get rid of all the long delays that were previously used, cutting more than a minute off the runtime of the isolation tests. Back-patch to all supported branches, because the buildfarm instabilities affect all the branches, and because it seems desirable to keep isolationtester's capabilities the same across all branches to simplify possible future back-patching of tests. Discussion: https://postgr.es/m/327948.1623725828@sss.pgh.pa.us
2021-06-23 03:43:12 +02:00
struct timeval start_time;
isolationtester: don't repeat the is-it-waiting query when retrying a step. If we're retrying a step, then we already decided it was blocked on a lock, and there's no need to recheck that. The original coding of commit 38f8bdcac4982215beb9f65a19debecaf22fd470 resulted in a large number of is-it-waiting queries when dealing with multiple concurrently-blocked sessions, which is fairly pointless and also results in test failures in CLOBBER_CACHE_ALWAYS builds, where the is-it-waiting query is quite slow. This definition also permits appending pg_sleep() calls to steps where it's needed to control the order of finish of concurrent steps. Before, that did not work nicely because we'd decide that a step performing a sleep was not blocked and hang up waiting for it to finish, rather than noticing the completion of the concurrent step we're supposed to notice first. In passing, revise handling of removal of completed waiting steps to make it a bit less messy.
2016-02-12 20:10:36 +01:00
Use annotations to reduce instability of isolation-test results. We've long contended with isolation test results that aren't entirely stable. Some test scripts insert long delays to try to force stable results, which is not terribly desirable; but other erratic failure modes remain, causing unrepeatable buildfarm failures. I've spent a fair amount of time trying to solve this by improving the server-side support code, without much success: that way is fundamentally unable to cope with diffs that stem from chance ordering of arrival of messages from different server processes. We can improve matters on the client side, however, by annotating the test scripts themselves to show the desired reporting order of events that might occur in different orders. This patch adds three types of annotations to deal with (a) test steps that might or might not complete their waits before the isolationtester can see them waiting; (b) test steps in different sessions that can legitimately complete in either order; and (c) NOTIFY messages that might arrive before or after the completion of a step in another session. We might need more annotation types later, but this seems to be enough to deal with the instabilities we've seen in the buildfarm. It also lets us get rid of all the long delays that were previously used, cutting more than a minute off the runtime of the isolation tests. Back-patch to all supported branches, because the buildfarm instabilities affect all the branches, and because it seems desirable to keep isolationtester's capabilities the same across all branches to simplify possible future back-patching of tests. Discussion: https://postgr.es/m/327948.1623725828@sss.pgh.pa.us
2021-06-23 03:43:12 +02:00
gettimeofday(&start_time, NULL);
isolationtester: don't repeat the is-it-waiting query when retrying a step. If we're retrying a step, then we already decided it was blocked on a lock, and there's no need to recheck that. The original coding of commit 38f8bdcac4982215beb9f65a19debecaf22fd470 resulted in a large number of is-it-waiting queries when dealing with multiple concurrently-blocked sessions, which is fairly pointless and also results in test failures in CLOBBER_CACHE_ALWAYS builds, where the is-it-waiting query is quite slow. This definition also permits appending pg_sleep() calls to steps where it's needed to control the order of finish of concurrent steps. Before, that did not work nicely because we'd decide that a step performing a sleep was not blocked and hang up waiting for it to finish, rather than noticing the completion of the concurrent step we're supposed to notice first. In passing, revise handling of removal of completed waiting steps to make it a bit less messy.
2016-02-12 20:10:36 +01:00
Use annotations to reduce instability of isolation-test results. We've long contended with isolation test results that aren't entirely stable. Some test scripts insert long delays to try to force stable results, which is not terribly desirable; but other erratic failure modes remain, causing unrepeatable buildfarm failures. I've spent a fair amount of time trying to solve this by improving the server-side support code, without much success: that way is fundamentally unable to cope with diffs that stem from chance ordering of arrival of messages from different server processes. We can improve matters on the client side, however, by annotating the test scripts themselves to show the desired reporting order of events that might occur in different orders. This patch adds three types of annotations to deal with (a) test steps that might or might not complete their waits before the isolationtester can see them waiting; (b) test steps in different sessions that can legitimately complete in either order; and (c) NOTIFY messages that might arrive before or after the completion of a step in another session. We might need more annotation types later, but this seems to be enough to deal with the instabilities we've seen in the buildfarm. It also lets us get rid of all the long delays that were previously used, cutting more than a minute off the runtime of the isolation tests. Back-patch to all supported branches, because the buildfarm instabilities affect all the branches, and because it seems desirable to keep isolationtester's capabilities the same across all branches to simplify possible future back-patching of tests. Discussion: https://postgr.es/m/327948.1623725828@sss.pgh.pa.us
2021-06-23 03:43:12 +02:00
while (iconn->active_step != NULL)
Detect invalid permutations in isolationtester isolationtester is now able to continue running other permutations when it detects that one of them is invalid, which is useful during initial development of spec files. Author: Alexander Shulgin
2012-01-14 23:36:39 +01:00
{
Use annotations to reduce instability of isolation-test results. We've long contended with isolation test results that aren't entirely stable. Some test scripts insert long delays to try to force stable results, which is not terribly desirable; but other erratic failure modes remain, causing unrepeatable buildfarm failures. I've spent a fair amount of time trying to solve this by improving the server-side support code, without much success: that way is fundamentally unable to cope with diffs that stem from chance ordering of arrival of messages from different server processes. We can improve matters on the client side, however, by annotating the test scripts themselves to show the desired reporting order of events that might occur in different orders. This patch adds three types of annotations to deal with (a) test steps that might or might not complete their waits before the isolationtester can see them waiting; (b) test steps in different sessions that can legitimately complete in either order; and (c) NOTIFY messages that might arrive before or after the completion of a step in another session. We might need more annotation types later, but this seems to be enough to deal with the instabilities we've seen in the buildfarm. It also lets us get rid of all the long delays that were previously used, cutting more than a minute off the runtime of the isolation tests. Back-patch to all supported branches, because the buildfarm instabilities affect all the branches, and because it seems desirable to keep isolationtester's capabilities the same across all branches to simplify possible future back-patching of tests. Discussion: https://postgr.es/m/327948.1623725828@sss.pgh.pa.us
2021-06-23 03:43:12 +02:00
PermutationStep *oldstep = iconn->active_step;
/*
* Wait for oldstep. But even though we don't use
* STEP_NONBLOCK, it might not complete because of blocker
* conditions.
*/
if (!try_complete_step(testspec, oldstep, STEP_RETRY))
Modify the isolation tester so that multiple sessions can wait. This allows testing of deadlock scenarios. Scenarios that would previously have been considered invalid are now simply taken as a scenario in which more than one backend will wait.
2016-02-11 14:36:30 +01:00
{
Use annotations to reduce instability of isolation-test results. We've long contended with isolation test results that aren't entirely stable. Some test scripts insert long delays to try to force stable results, which is not terribly desirable; but other erratic failure modes remain, causing unrepeatable buildfarm failures. I've spent a fair amount of time trying to solve this by improving the server-side support code, without much success: that way is fundamentally unable to cope with diffs that stem from chance ordering of arrival of messages from different server processes. We can improve matters on the client side, however, by annotating the test scripts themselves to show the desired reporting order of events that might occur in different orders. This patch adds three types of annotations to deal with (a) test steps that might or might not complete their waits before the isolationtester can see them waiting; (b) test steps in different sessions that can legitimately complete in either order; and (c) NOTIFY messages that might arrive before or after the completion of a step in another session. We might need more annotation types later, but this seems to be enough to deal with the instabilities we've seen in the buildfarm. It also lets us get rid of all the long delays that were previously used, cutting more than a minute off the runtime of the isolation tests. Back-patch to all supported branches, because the buildfarm instabilities affect all the branches, and because it seems desirable to keep isolationtester's capabilities the same across all branches to simplify possible future back-patching of tests. Discussion: https://postgr.es/m/327948.1623725828@sss.pgh.pa.us
2021-06-23 03:43:12 +02:00
/* Done, so remove oldstep from the waiting[] array. */
int w;
for (w = 0; w < nwaiting; w++)
{
if (oldstep == waiting[w])
break;
}
if (w >= nwaiting)
abort(); /* can't happen */
Modify the isolation tester so that multiple sessions can wait. This allows testing of deadlock scenarios. Scenarios that would previously have been considered invalid are now simply taken as a scenario in which more than one backend will wait.
2016-02-11 14:36:30 +01:00
if (w + 1 < nwaiting)
Use memmove() not memcpy() to slide some pointers down. The previous coding here was formally undefined, though it seems to accidentally work on most platforms in the buildfarm. Caught by some OpenBSD platforms in which libc contains an assertion check for overlapping areas passed to memcpy(). Thomas Munro
2016-04-28 00:19:28 +02:00
memmove(&waiting[w], &waiting[w + 1],
Use annotations to reduce instability of isolation-test results. We've long contended with isolation test results that aren't entirely stable. Some test scripts insert long delays to try to force stable results, which is not terribly desirable; but other erratic failure modes remain, causing unrepeatable buildfarm failures. I've spent a fair amount of time trying to solve this by improving the server-side support code, without much success: that way is fundamentally unable to cope with diffs that stem from chance ordering of arrival of messages from different server processes. We can improve matters on the client side, however, by annotating the test scripts themselves to show the desired reporting order of events that might occur in different orders. This patch adds three types of annotations to deal with (a) test steps that might or might not complete their waits before the isolationtester can see them waiting; (b) test steps in different sessions that can legitimately complete in either order; and (c) NOTIFY messages that might arrive before or after the completion of a step in another session. We might need more annotation types later, but this seems to be enough to deal with the instabilities we've seen in the buildfarm. It also lets us get rid of all the long delays that were previously used, cutting more than a minute off the runtime of the isolation tests. Back-patch to all supported branches, because the buildfarm instabilities affect all the branches, and because it seems desirable to keep isolationtester's capabilities the same across all branches to simplify possible future back-patching of tests. Discussion: https://postgr.es/m/327948.1623725828@sss.pgh.pa.us
2021-06-23 03:43:12 +02:00
(nwaiting - (w + 1)) * sizeof(PermutationStep *));
Modify the isolation tester so that multiple sessions can wait. This allows testing of deadlock scenarios. Scenarios that would previously have been considered invalid are now simply taken as a scenario in which more than one backend will wait.
2016-02-11 14:36:30 +01:00
nwaiting--;
}
Detect invalid permutations in isolationtester isolationtester is now able to continue running other permutations when it detects that one of them is invalid, which is useful during initial development of spec files. Author: Alexander Shulgin
2012-01-14 23:36:39 +01:00
Use annotations to reduce instability of isolation-test results. We've long contended with isolation test results that aren't entirely stable. Some test scripts insert long delays to try to force stable results, which is not terribly desirable; but other erratic failure modes remain, causing unrepeatable buildfarm failures. I've spent a fair amount of time trying to solve this by improving the server-side support code, without much success: that way is fundamentally unable to cope with diffs that stem from chance ordering of arrival of messages from different server processes. We can improve matters on the client side, however, by annotating the test scripts themselves to show the desired reporting order of events that might occur in different orders. This patch adds three types of annotations to deal with (a) test steps that might or might not complete their waits before the isolationtester can see them waiting; (b) test steps in different sessions that can legitimately complete in either order; and (c) NOTIFY messages that might arrive before or after the completion of a step in another session. We might need more annotation types later, but this seems to be enough to deal with the instabilities we've seen in the buildfarm. It also lets us get rid of all the long delays that were previously used, cutting more than a minute off the runtime of the isolation tests. Back-patch to all supported branches, because the buildfarm instabilities affect all the branches, and because it seems desirable to keep isolationtester's capabilities the same across all branches to simplify possible future back-patching of tests. Discussion: https://postgr.es/m/327948.1623725828@sss.pgh.pa.us
2021-06-23 03:43:12 +02:00
/*
* Check for other steps that have finished. We should do
* this if oldstep completed, as it might have unblocked
* something. On the other hand, if oldstep hasn't completed,
* we must poll all the active steps in hopes of unblocking
* oldstep. So either way, poll them.
*/
nwaiting = try_complete_steps(testspec, waiting, nwaiting,
STEP_NONBLOCK | STEP_RETRY);
/*
* If the target session is still busy, apply a timeout to
* keep from hanging indefinitely, which could happen with
* incorrect blocker annotations. Use the same 2 *
* max_step_wait limit as try_complete_step does for deciding
* to die. (We don't bother with trying to cancel anything,
* since it's unclear what to cancel in this case.)
*/
if (iconn->active_step != NULL)
{
struct timeval current_time;
int64 td;
gettimeofday(&current_time, NULL);
td = (int64) current_time.tv_sec - (int64) start_time.tv_sec;
td *= USECS_PER_SEC;
td += (int64) current_time.tv_usec - (int64) start_time.tv_usec;
if (td > 2 * max_step_wait)
{
fprintf(stderr, "step %s timed out after %d seconds\n",
iconn->active_step->name,
(int) (td / USECS_PER_SEC));
fprintf(stderr, "active steps are:");
for (j = 1; j < nconns; j++)
{
IsoConnInfo *oconn = &conns[j];
if (oconn->active_step != NULL)
fprintf(stderr, " %s",
oconn->active_step->name);
}
fprintf(stderr, "\n");
exit(1);
}
}
}
Detect invalid permutations in isolationtester isolationtester is now able to continue running other permutations when it detects that one of them is invalid, which is useful during initial development of spec files. Author: Alexander Shulgin
2012-01-14 23:36:39 +01:00
}
Modify the isolation tester so that multiple sessions can wait. This allows testing of deadlock scenarios. Scenarios that would previously have been considered invalid are now simply taken as a scenario in which more than one backend will wait.
2016-02-11 14:36:30 +01:00
/* Send the query for this step. */
Detect invalid permutations in isolationtester isolationtester is now able to continue running other permutations when it detects that one of them is invalid, which is useful during initial development of spec files. Author: Alexander Shulgin
2012-01-14 23:36:39 +01:00
if (!PQsendQuery(conn, step->sql))
Implement genuine serializable isolation level. Until now, our Serializable mode has in fact been what's called Snapshot Isolation, which allows some anomalies that could not occur in any serialized ordering of the transactions. This patch fixes that using a method called Serializable Snapshot Isolation, based on research papers by Michael J. Cahill (see README-SSI for full references). In Serializable Snapshot Isolation, transactions run like they do in Snapshot Isolation, but a predicate lock manager observes the reads and writes performed and aborts transactions if it detects that an anomaly might occur. This method produces some false positives, ie. it sometimes aborts transactions even though there is no anomaly. To track reads we implement predicate locking, see storage/lmgr/predicate.c. Whenever a tuple is read, a predicate lock is acquired on the tuple. Shared memory is finite, so when a transaction takes many tuple-level locks on a page, the locks are promoted to a single page-level lock, and further to a single relation level lock if necessary. To lock key values with no matching tuple, a sequential scan always takes a relation-level lock, and an index scan acquires a page-level lock that covers the search key, whether or not there are any matching keys at the moment. A predicate lock doesn't conflict with any regular locks or with another predicate locks in the normal sense. They're only used by the predicate lock manager to detect the danger of anomalies. Only serializable transactions participate in predicate locking, so there should be no extra overhead for for other transactions. Predicate locks can't be released at commit, but must be remembered until all the transactions that overlapped with it have completed. That means that we need to remember an unbounded amount of predicate locks, so we apply a lossy but conservative method of tracking locks for committed transactions. If we run short of shared memory, we overflow to a new "pg_serial" SLRU pool. We don't currently allow Serializable transactions in Hot Standby mode. That would be hard, because even read-only transactions can cause anomalies that wouldn't otherwise occur. Serializable isolation mode now means the new fully serializable level. Repeatable Read gives you the old Snapshot Isolation level that we have always had. Kevin Grittner and Dan Ports, reviewed by Jeff Davis, Heikki Linnakangas and Anssi Kääriäinen
2011-02-07 22:46:51 +01:00
{
Add debugging aid in isolationtester
2011-10-25 03:11:10 +02:00
fprintf(stdout, "failed to send query for step %s: %s\n",
Fix minor bug in isolationtester. If the lock wait query failed, isolationtester would report the PQerrorMessage from some other connection, meaning there would be no message or an unrelated one. This seems like a pretty unlikely occurrence, but if it did happen, this bug could make it really difficult/confusing to figure out what happened. That seems to justify patching all the way back. In passing, clean up another place where the "wrong" conn was used for an error report. That one's not actually buggy because it's a different alias for the same connection, but it's still confusing to the reader.
2018-10-17 21:06:38 +02:00
step->name, PQerrorMessage(conn));
isolationtester: Use atexit() Replace exit_nicely() calls with standard exit() and register the cleanup actions using atexit(). Reviewed-by: Alvaro Herrera <alvherre@2ndquadrant.com> Discussion: https://www.postgresql.org/message-id/flat/ec4135ba-84e9-28bf-b584-0e78d47448d5@2ndquadrant.com/
2019-01-05 15:05:49 +01:00
exit(1);
Add support for blocked commands in isolationtester This enables us to test that blocking commands (such as foreign keys checks that conflict with some other lock) act as intended. The set of tests that this adds is pretty minimal, but can easily be extended by adding new specs. The intention is that this will serve as a basis for ensuring that further tweaks of locking implementation preserve (or improve) existing behavior. Author: Noah Misch
2011-07-12 22:35:09 +02:00
}
Implement genuine serializable isolation level. Until now, our Serializable mode has in fact been what's called Snapshot Isolation, which allows some anomalies that could not occur in any serialized ordering of the transactions. This patch fixes that using a method called Serializable Snapshot Isolation, based on research papers by Michael J. Cahill (see README-SSI for full references). In Serializable Snapshot Isolation, transactions run like they do in Snapshot Isolation, but a predicate lock manager observes the reads and writes performed and aborts transactions if it detects that an anomaly might occur. This method produces some false positives, ie. it sometimes aborts transactions even though there is no anomaly. To track reads we implement predicate locking, see storage/lmgr/predicate.c. Whenever a tuple is read, a predicate lock is acquired on the tuple. Shared memory is finite, so when a transaction takes many tuple-level locks on a page, the locks are promoted to a single page-level lock, and further to a single relation level lock if necessary. To lock key values with no matching tuple, a sequential scan always takes a relation-level lock, and an index scan acquires a page-level lock that covers the search key, whether or not there are any matching keys at the moment. A predicate lock doesn't conflict with any regular locks or with another predicate locks in the normal sense. They're only used by the predicate lock manager to detect the danger of anomalies. Only serializable transactions participate in predicate locking, so there should be no extra overhead for for other transactions. Predicate locks can't be released at commit, but must be remembered until all the transactions that overlapped with it have completed. That means that we need to remember an unbounded amount of predicate locks, so we apply a lossy but conservative method of tracking locks for committed transactions. If we run short of shared memory, we overflow to a new "pg_serial" SLRU pool. We don't currently allow Serializable transactions in Hot Standby mode. That would be hard, because even read-only transactions can cause anomalies that wouldn't otherwise occur. Serializable isolation mode now means the new fully serializable level. Repeatable Read gives you the old Snapshot Isolation level that we have always had. Kevin Grittner and Dan Ports, reviewed by Jeff Davis, Heikki Linnakangas and Anssi Kääriäinen
2011-02-07 22:46:51 +01:00
Use annotations to reduce instability of isolation-test results. We've long contended with isolation test results that aren't entirely stable. Some test scripts insert long delays to try to force stable results, which is not terribly desirable; but other erratic failure modes remain, causing unrepeatable buildfarm failures. I've spent a fair amount of time trying to solve this by improving the server-side support code, without much success: that way is fundamentally unable to cope with diffs that stem from chance ordering of arrival of messages from different server processes. We can improve matters on the client side, however, by annotating the test scripts themselves to show the desired reporting order of events that might occur in different orders. This patch adds three types of annotations to deal with (a) test steps that might or might not complete their waits before the isolationtester can see them waiting; (b) test steps in different sessions that can legitimately complete in either order; and (c) NOTIFY messages that might arrive before or after the completion of a step in another session. We might need more annotation types later, but this seems to be enough to deal with the instabilities we've seen in the buildfarm. It also lets us get rid of all the long delays that were previously used, cutting more than a minute off the runtime of the isolation tests. Back-patch to all supported branches, because the buildfarm instabilities affect all the branches, and because it seems desirable to keep isolationtester's capabilities the same across all branches to simplify possible future back-patching of tests. Discussion: https://postgr.es/m/327948.1623725828@sss.pgh.pa.us
2021-06-23 03:43:12 +02:00
/* Remember we launched a step. */
iconn->active_step = pstep;
Implement genuine serializable isolation level. Until now, our Serializable mode has in fact been what's called Snapshot Isolation, which allows some anomalies that could not occur in any serialized ordering of the transactions. This patch fixes that using a method called Serializable Snapshot Isolation, based on research papers by Michael J. Cahill (see README-SSI for full references). In Serializable Snapshot Isolation, transactions run like they do in Snapshot Isolation, but a predicate lock manager observes the reads and writes performed and aborts transactions if it detects that an anomaly might occur. This method produces some false positives, ie. it sometimes aborts transactions even though there is no anomaly. To track reads we implement predicate locking, see storage/lmgr/predicate.c. Whenever a tuple is read, a predicate lock is acquired on the tuple. Shared memory is finite, so when a transaction takes many tuple-level locks on a page, the locks are promoted to a single page-level lock, and further to a single relation level lock if necessary. To lock key values with no matching tuple, a sequential scan always takes a relation-level lock, and an index scan acquires a page-level lock that covers the search key, whether or not there are any matching keys at the moment. A predicate lock doesn't conflict with any regular locks or with another predicate locks in the normal sense. They're only used by the predicate lock manager to detect the danger of anomalies. Only serializable transactions participate in predicate locking, so there should be no extra overhead for for other transactions. Predicate locks can't be released at commit, but must be remembered until all the transactions that overlapped with it have completed. That means that we need to remember an unbounded amount of predicate locks, so we apply a lossy but conservative method of tracking locks for committed transactions. If we run short of shared memory, we overflow to a new "pg_serial" SLRU pool. We don't currently allow Serializable transactions in Hot Standby mode. That would be hard, because even read-only transactions can cause anomalies that wouldn't otherwise occur. Serializable isolation mode now means the new fully serializable level. Repeatable Read gives you the old Snapshot Isolation level that we have always had. Kevin Grittner and Dan Ports, reviewed by Jeff Davis, Heikki Linnakangas and Anssi Kääriäinen
2011-02-07 22:46:51 +01:00
Use annotations to reduce instability of isolation-test results. We've long contended with isolation test results that aren't entirely stable. Some test scripts insert long delays to try to force stable results, which is not terribly desirable; but other erratic failure modes remain, causing unrepeatable buildfarm failures. I've spent a fair amount of time trying to solve this by improving the server-side support code, without much success: that way is fundamentally unable to cope with diffs that stem from chance ordering of arrival of messages from different server processes. We can improve matters on the client side, however, by annotating the test scripts themselves to show the desired reporting order of events that might occur in different orders. This patch adds three types of annotations to deal with (a) test steps that might or might not complete their waits before the isolationtester can see them waiting; (b) test steps in different sessions that can legitimately complete in either order; and (c) NOTIFY messages that might arrive before or after the completion of a step in another session. We might need more annotation types later, but this seems to be enough to deal with the instabilities we've seen in the buildfarm. It also lets us get rid of all the long delays that were previously used, cutting more than a minute off the runtime of the isolation tests. Back-patch to all supported branches, because the buildfarm instabilities affect all the branches, and because it seems desirable to keep isolationtester's capabilities the same across all branches to simplify possible future back-patching of tests. Discussion: https://postgr.es/m/327948.1623725828@sss.pgh.pa.us
2021-06-23 03:43:12 +02:00
/* Remember target number of NOTICEs for any blocker conditions. */
for (j = 0; j < pstep->nblockers; j++)
Modify the isolation tester so that multiple sessions can wait. This allows testing of deadlock scenarios. Scenarios that would previously have been considered invalid are now simply taken as a scenario in which more than one backend will wait.
2016-02-11 14:36:30 +01:00
{
Use annotations to reduce instability of isolation-test results. We've long contended with isolation test results that aren't entirely stable. Some test scripts insert long delays to try to force stable results, which is not terribly desirable; but other erratic failure modes remain, causing unrepeatable buildfarm failures. I've spent a fair amount of time trying to solve this by improving the server-side support code, without much success: that way is fundamentally unable to cope with diffs that stem from chance ordering of arrival of messages from different server processes. We can improve matters on the client side, however, by annotating the test scripts themselves to show the desired reporting order of events that might occur in different orders. This patch adds three types of annotations to deal with (a) test steps that might or might not complete their waits before the isolationtester can see them waiting; (b) test steps in different sessions that can legitimately complete in either order; and (c) NOTIFY messages that might arrive before or after the completion of a step in another session. We might need more annotation types later, but this seems to be enough to deal with the instabilities we've seen in the buildfarm. It also lets us get rid of all the long delays that were previously used, cutting more than a minute off the runtime of the isolation tests. Back-patch to all supported branches, because the buildfarm instabilities affect all the branches, and because it seems desirable to keep isolationtester's capabilities the same across all branches to simplify possible future back-patching of tests. Discussion: https://postgr.es/m/327948.1623725828@sss.pgh.pa.us
2021-06-23 03:43:12 +02:00
PermutationStepBlocker *blocker = pstep->blockers[j];
if (blocker->blocktype == PSB_NUM_NOTICES)
blocker->target_notices = blocker->num_notices +
conns[blocker->step->session + 1].total_notices;
Implement genuine serializable isolation level. Until now, our Serializable mode has in fact been what's called Snapshot Isolation, which allows some anomalies that could not occur in any serialized ordering of the transactions. This patch fixes that using a method called Serializable Snapshot Isolation, based on research papers by Michael J. Cahill (see README-SSI for full references). In Serializable Snapshot Isolation, transactions run like they do in Snapshot Isolation, but a predicate lock manager observes the reads and writes performed and aborts transactions if it detects that an anomaly might occur. This method produces some false positives, ie. it sometimes aborts transactions even though there is no anomaly. To track reads we implement predicate locking, see storage/lmgr/predicate.c. Whenever a tuple is read, a predicate lock is acquired on the tuple. Shared memory is finite, so when a transaction takes many tuple-level locks on a page, the locks are promoted to a single page-level lock, and further to a single relation level lock if necessary. To lock key values with no matching tuple, a sequential scan always takes a relation-level lock, and an index scan acquires a page-level lock that covers the search key, whether or not there are any matching keys at the moment. A predicate lock doesn't conflict with any regular locks or with another predicate locks in the normal sense. They're only used by the predicate lock manager to detect the danger of anomalies. Only serializable transactions participate in predicate locking, so there should be no extra overhead for for other transactions. Predicate locks can't be released at commit, but must be remembered until all the transactions that overlapped with it have completed. That means that we need to remember an unbounded amount of predicate locks, so we apply a lossy but conservative method of tracking locks for committed transactions. If we run short of shared memory, we overflow to a new "pg_serial" SLRU pool. We don't currently allow Serializable transactions in Hot Standby mode. That would be hard, because even read-only transactions can cause anomalies that wouldn't otherwise occur. Serializable isolation mode now means the new fully serializable level. Repeatable Read gives you the old Snapshot Isolation level that we have always had. Kevin Grittner and Dan Ports, reviewed by Jeff Davis, Heikki Linnakangas and Anssi Kääriäinen
2011-02-07 22:46:51 +01:00
}
Modify the isolation tester so that multiple sessions can wait. This allows testing of deadlock scenarios. Scenarios that would previously have been considered invalid are now simply taken as a scenario in which more than one backend will wait.
2016-02-11 14:36:30 +01:00
Use annotations to reduce instability of isolation-test results. We've long contended with isolation test results that aren't entirely stable. Some test scripts insert long delays to try to force stable results, which is not terribly desirable; but other erratic failure modes remain, causing unrepeatable buildfarm failures. I've spent a fair amount of time trying to solve this by improving the server-side support code, without much success: that way is fundamentally unable to cope with diffs that stem from chance ordering of arrival of messages from different server processes. We can improve matters on the client side, however, by annotating the test scripts themselves to show the desired reporting order of events that might occur in different orders. This patch adds three types of annotations to deal with (a) test steps that might or might not complete their waits before the isolationtester can see them waiting; (b) test steps in different sessions that can legitimately complete in either order; and (c) NOTIFY messages that might arrive before or after the completion of a step in another session. We might need more annotation types later, but this seems to be enough to deal with the instabilities we've seen in the buildfarm. It also lets us get rid of all the long delays that were previously used, cutting more than a minute off the runtime of the isolation tests. Back-patch to all supported branches, because the buildfarm instabilities affect all the branches, and because it seems desirable to keep isolationtester's capabilities the same across all branches to simplify possible future back-patching of tests. Discussion: https://postgr.es/m/327948.1623725828@sss.pgh.pa.us
2021-06-23 03:43:12 +02:00
/* Try to complete this step without blocking. */
mustwait = try_complete_step(testspec, pstep, STEP_NONBLOCK);
/* Check for completion of any steps that were previously waiting. */
nwaiting = try_complete_steps(testspec, waiting, nwaiting,
STEP_NONBLOCK | STEP_RETRY);
Modify the isolation tester so that multiple sessions can wait. This allows testing of deadlock scenarios. Scenarios that would previously have been considered invalid are now simply taken as a scenario in which more than one backend will wait.
2016-02-11 14:36:30 +01:00
/* If this step is waiting, add it to the array of waiters. */
if (mustwait)
Use annotations to reduce instability of isolation-test results. We've long contended with isolation test results that aren't entirely stable. Some test scripts insert long delays to try to force stable results, which is not terribly desirable; but other erratic failure modes remain, causing unrepeatable buildfarm failures. I've spent a fair amount of time trying to solve this by improving the server-side support code, without much success: that way is fundamentally unable to cope with diffs that stem from chance ordering of arrival of messages from different server processes. We can improve matters on the client side, however, by annotating the test scripts themselves to show the desired reporting order of events that might occur in different orders. This patch adds three types of annotations to deal with (a) test steps that might or might not complete their waits before the isolationtester can see them waiting; (b) test steps in different sessions that can legitimately complete in either order; and (c) NOTIFY messages that might arrive before or after the completion of a step in another session. We might need more annotation types later, but this seems to be enough to deal with the instabilities we've seen in the buildfarm. It also lets us get rid of all the long delays that were previously used, cutting more than a minute off the runtime of the isolation tests. Back-patch to all supported branches, because the buildfarm instabilities affect all the branches, and because it seems desirable to keep isolationtester's capabilities the same across all branches to simplify possible future back-patching of tests. Discussion: https://postgr.es/m/327948.1623725828@sss.pgh.pa.us
2021-06-23 03:43:12 +02:00
waiting[nwaiting++] = pstep;
Implement genuine serializable isolation level. Until now, our Serializable mode has in fact been what's called Snapshot Isolation, which allows some anomalies that could not occur in any serialized ordering of the transactions. This patch fixes that using a method called Serializable Snapshot Isolation, based on research papers by Michael J. Cahill (see README-SSI for full references). In Serializable Snapshot Isolation, transactions run like they do in Snapshot Isolation, but a predicate lock manager observes the reads and writes performed and aborts transactions if it detects that an anomaly might occur. This method produces some false positives, ie. it sometimes aborts transactions even though there is no anomaly. To track reads we implement predicate locking, see storage/lmgr/predicate.c. Whenever a tuple is read, a predicate lock is acquired on the tuple. Shared memory is finite, so when a transaction takes many tuple-level locks on a page, the locks are promoted to a single page-level lock, and further to a single relation level lock if necessary. To lock key values with no matching tuple, a sequential scan always takes a relation-level lock, and an index scan acquires a page-level lock that covers the search key, whether or not there are any matching keys at the moment. A predicate lock doesn't conflict with any regular locks or with another predicate locks in the normal sense. They're only used by the predicate lock manager to detect the danger of anomalies. Only serializable transactions participate in predicate locking, so there should be no extra overhead for for other transactions. Predicate locks can't be released at commit, but must be remembered until all the transactions that overlapped with it have completed. That means that we need to remember an unbounded amount of predicate locks, so we apply a lossy but conservative method of tracking locks for committed transactions. If we run short of shared memory, we overflow to a new "pg_serial" SLRU pool. We don't currently allow Serializable transactions in Hot Standby mode. That would be hard, because even read-only transactions can cause anomalies that wouldn't otherwise occur. Serializable isolation mode now means the new fully serializable level. Repeatable Read gives you the old Snapshot Isolation level that we have always had. Kevin Grittner and Dan Ports, reviewed by Jeff Davis, Heikki Linnakangas and Anssi Kääriäinen
2011-02-07 22:46:51 +01:00
}
isolationtester: don't repeat the is-it-waiting query when retrying a step. If we're retrying a step, then we already decided it was blocked on a lock, and there's no need to recheck that. The original coding of commit 38f8bdcac4982215beb9f65a19debecaf22fd470 resulted in a large number of is-it-waiting queries when dealing with multiple concurrently-blocked sessions, which is fairly pointless and also results in test failures in CLOBBER_CACHE_ALWAYS builds, where the is-it-waiting query is quite slow. This definition also permits appending pg_sleep() calls to steps where it's needed to control the order of finish of concurrent steps. Before, that did not work nicely because we'd decide that a step performing a sleep was not blocked and hang up waiting for it to finish, rather than noticing the completion of the concurrent step we're supposed to notice first. In passing, revise handling of removal of completed waiting steps to make it a bit less messy.
2016-02-12 20:10:36 +01:00
/* Wait for any remaining queries. */
Use annotations to reduce instability of isolation-test results. We've long contended with isolation test results that aren't entirely stable. Some test scripts insert long delays to try to force stable results, which is not terribly desirable; but other erratic failure modes remain, causing unrepeatable buildfarm failures. I've spent a fair amount of time trying to solve this by improving the server-side support code, without much success: that way is fundamentally unable to cope with diffs that stem from chance ordering of arrival of messages from different server processes. We can improve matters on the client side, however, by annotating the test scripts themselves to show the desired reporting order of events that might occur in different orders. This patch adds three types of annotations to deal with (a) test steps that might or might not complete their waits before the isolationtester can see them waiting; (b) test steps in different sessions that can legitimately complete in either order; and (c) NOTIFY messages that might arrive before or after the completion of a step in another session. We might need more annotation types later, but this seems to be enough to deal with the instabilities we've seen in the buildfarm. It also lets us get rid of all the long delays that were previously used, cutting more than a minute off the runtime of the isolation tests. Back-patch to all supported branches, because the buildfarm instabilities affect all the branches, and because it seems desirable to keep isolationtester's capabilities the same across all branches to simplify possible future back-patching of tests. Discussion: https://postgr.es/m/327948.1623725828@sss.pgh.pa.us
2021-06-23 03:43:12 +02:00
nwaiting = try_complete_steps(testspec, waiting, nwaiting, STEP_RETRY);
if (nwaiting != 0)
Remove dependency on error ordering in isolation tests We now report errors reported by the just-unblocked and unblocking transactions identically; this should fix relatively common buildfarm failures reported by animals that are failing the "wrong" session.
2011-09-27 21:08:31 +02:00
{
Use annotations to reduce instability of isolation-test results. We've long contended with isolation test results that aren't entirely stable. Some test scripts insert long delays to try to force stable results, which is not terribly desirable; but other erratic failure modes remain, causing unrepeatable buildfarm failures. I've spent a fair amount of time trying to solve this by improving the server-side support code, without much success: that way is fundamentally unable to cope with diffs that stem from chance ordering of arrival of messages from different server processes. We can improve matters on the client side, however, by annotating the test scripts themselves to show the desired reporting order of events that might occur in different orders. This patch adds three types of annotations to deal with (a) test steps that might or might not complete their waits before the isolationtester can see them waiting; (b) test steps in different sessions that can legitimately complete in either order; and (c) NOTIFY messages that might arrive before or after the completion of a step in another session. We might need more annotation types later, but this seems to be enough to deal with the instabilities we've seen in the buildfarm. It also lets us get rid of all the long delays that were previously used, cutting more than a minute off the runtime of the isolation tests. Back-patch to all supported branches, because the buildfarm instabilities affect all the branches, and because it seems desirable to keep isolationtester's capabilities the same across all branches to simplify possible future back-patching of tests. Discussion: https://postgr.es/m/327948.1623725828@sss.pgh.pa.us
2021-06-23 03:43:12 +02:00
fprintf(stderr, "failed to complete permutation due to mutually-blocking steps\n");
exit(1);
Remove dependency on error ordering in isolation tests We now report errors reported by the just-unblocked and unblocking transactions identically; this should fix relatively common buildfarm failures reported by animals that are failing the "wrong" session.
2011-09-27 21:08:31 +02:00
}
Add support for blocked commands in isolationtester This enables us to test that blocking commands (such as foreign keys checks that conflict with some other lock) act as intended. The set of tests that this adds is pretty minimal, but can easily be extended by adding new specs. The intention is that this will serve as a basis for ensuring that further tweaks of locking implementation preserve (or improve) existing behavior. Author: Noah Misch
2011-07-12 22:35:09 +02:00
Implement genuine serializable isolation level. Until now, our Serializable mode has in fact been what's called Snapshot Isolation, which allows some anomalies that could not occur in any serialized ordering of the transactions. This patch fixes that using a method called Serializable Snapshot Isolation, based on research papers by Michael J. Cahill (see README-SSI for full references). In Serializable Snapshot Isolation, transactions run like they do in Snapshot Isolation, but a predicate lock manager observes the reads and writes performed and aborts transactions if it detects that an anomaly might occur. This method produces some false positives, ie. it sometimes aborts transactions even though there is no anomaly. To track reads we implement predicate locking, see storage/lmgr/predicate.c. Whenever a tuple is read, a predicate lock is acquired on the tuple. Shared memory is finite, so when a transaction takes many tuple-level locks on a page, the locks are promoted to a single page-level lock, and further to a single relation level lock if necessary. To lock key values with no matching tuple, a sequential scan always takes a relation-level lock, and an index scan acquires a page-level lock that covers the search key, whether or not there are any matching keys at the moment. A predicate lock doesn't conflict with any regular locks or with another predicate locks in the normal sense. They're only used by the predicate lock manager to detect the danger of anomalies. Only serializable transactions participate in predicate locking, so there should be no extra overhead for for other transactions. Predicate locks can't be released at commit, but must be remembered until all the transactions that overlapped with it have completed. That means that we need to remember an unbounded amount of predicate locks, so we apply a lossy but conservative method of tracking locks for committed transactions. If we run short of shared memory, we overflow to a new "pg_serial" SLRU pool. We don't currently allow Serializable transactions in Hot Standby mode. That would be hard, because even read-only transactions can cause anomalies that wouldn't otherwise occur. Serializable isolation mode now means the new fully serializable level. Repeatable Read gives you the old Snapshot Isolation level that we have always had. Kevin Grittner and Dan Ports, reviewed by Jeff Davis, Heikki Linnakangas and Anssi Kääriäinen
2011-02-07 22:46:51 +01:00
/* Perform per-session teardown */
for (i = 0; i < testspec->nsessions; i++)
{
if (testspec->sessions[i]->teardownsql)
{
Use annotations to reduce instability of isolation-test results. We've long contended with isolation test results that aren't entirely stable. Some test scripts insert long delays to try to force stable results, which is not terribly desirable; but other erratic failure modes remain, causing unrepeatable buildfarm failures. I've spent a fair amount of time trying to solve this by improving the server-side support code, without much success: that way is fundamentally unable to cope with diffs that stem from chance ordering of arrival of messages from different server processes. We can improve matters on the client side, however, by annotating the test scripts themselves to show the desired reporting order of events that might occur in different orders. This patch adds three types of annotations to deal with (a) test steps that might or might not complete their waits before the isolationtester can see them waiting; (b) test steps in different sessions that can legitimately complete in either order; and (c) NOTIFY messages that might arrive before or after the completion of a step in another session. We might need more annotation types later, but this seems to be enough to deal with the instabilities we've seen in the buildfarm. It also lets us get rid of all the long delays that were previously used, cutting more than a minute off the runtime of the isolation tests. Back-patch to all supported branches, because the buildfarm instabilities affect all the branches, and because it seems desirable to keep isolationtester's capabilities the same across all branches to simplify possible future back-patching of tests. Discussion: https://postgr.es/m/327948.1623725828@sss.pgh.pa.us
2021-06-23 03:43:12 +02:00
res = PQexec(conns[i + 1].conn, testspec->sessions[i]->teardownsql);
isolationtester: Allow tuples to be returned in more places Previously, isolationtester would forbid returning tuples in session-specific teardown (but not global teardown), as well as in global setup. Allow these places to return tuples, too.
2013-10-04 15:32:48 +02:00
if (PQresultStatus(res) == PGRES_TUPLES_OK)
{
printResultSet(res);
}
else if (PQresultStatus(res) != PGRES_COMMAND_OK)
Implement genuine serializable isolation level. Until now, our Serializable mode has in fact been what's called Snapshot Isolation, which allows some anomalies that could not occur in any serialized ordering of the transactions. This patch fixes that using a method called Serializable Snapshot Isolation, based on research papers by Michael J. Cahill (see README-SSI for full references). In Serializable Snapshot Isolation, transactions run like they do in Snapshot Isolation, but a predicate lock manager observes the reads and writes performed and aborts transactions if it detects that an anomaly might occur. This method produces some false positives, ie. it sometimes aborts transactions even though there is no anomaly. To track reads we implement predicate locking, see storage/lmgr/predicate.c. Whenever a tuple is read, a predicate lock is acquired on the tuple. Shared memory is finite, so when a transaction takes many tuple-level locks on a page, the locks are promoted to a single page-level lock, and further to a single relation level lock if necessary. To lock key values with no matching tuple, a sequential scan always takes a relation-level lock, and an index scan acquires a page-level lock that covers the search key, whether or not there are any matching keys at the moment. A predicate lock doesn't conflict with any regular locks or with another predicate locks in the normal sense. They're only used by the predicate lock manager to detect the danger of anomalies. Only serializable transactions participate in predicate locking, so there should be no extra overhead for for other transactions. Predicate locks can't be released at commit, but must be remembered until all the transactions that overlapped with it have completed. That means that we need to remember an unbounded amount of predicate locks, so we apply a lossy but conservative method of tracking locks for committed transactions. If we run short of shared memory, we overflow to a new "pg_serial" SLRU pool. We don't currently allow Serializable transactions in Hot Standby mode. That would be hard, because even read-only transactions can cause anomalies that wouldn't otherwise occur. Serializable isolation mode now means the new fully serializable level. Repeatable Read gives you the old Snapshot Isolation level that we have always had. Kevin Grittner and Dan Ports, reviewed by Jeff Davis, Heikki Linnakangas and Anssi Kääriäinen
2011-02-07 22:46:51 +01:00
{
fprintf(stderr, "teardown of session %s failed: %s",
Use annotations to reduce instability of isolation-test results. We've long contended with isolation test results that aren't entirely stable. Some test scripts insert long delays to try to force stable results, which is not terribly desirable; but other erratic failure modes remain, causing unrepeatable buildfarm failures. I've spent a fair amount of time trying to solve this by improving the server-side support code, without much success: that way is fundamentally unable to cope with diffs that stem from chance ordering of arrival of messages from different server processes. We can improve matters on the client side, however, by annotating the test scripts themselves to show the desired reporting order of events that might occur in different orders. This patch adds three types of annotations to deal with (a) test steps that might or might not complete their waits before the isolationtester can see them waiting; (b) test steps in different sessions that can legitimately complete in either order; and (c) NOTIFY messages that might arrive before or after the completion of a step in another session. We might need more annotation types later, but this seems to be enough to deal with the instabilities we've seen in the buildfarm. It also lets us get rid of all the long delays that were previously used, cutting more than a minute off the runtime of the isolation tests. Back-patch to all supported branches, because the buildfarm instabilities affect all the branches, and because it seems desirable to keep isolationtester's capabilities the same across all branches to simplify possible future back-patching of tests. Discussion: https://postgr.es/m/327948.1623725828@sss.pgh.pa.us
2021-06-23 03:43:12 +02:00
conns[i + 1].sessionname,
PQerrorMessage(conns[i + 1].conn));
Implement genuine serializable isolation level. Until now, our Serializable mode has in fact been what's called Snapshot Isolation, which allows some anomalies that could not occur in any serialized ordering of the transactions. This patch fixes that using a method called Serializable Snapshot Isolation, based on research papers by Michael J. Cahill (see README-SSI for full references). In Serializable Snapshot Isolation, transactions run like they do in Snapshot Isolation, but a predicate lock manager observes the reads and writes performed and aborts transactions if it detects that an anomaly might occur. This method produces some false positives, ie. it sometimes aborts transactions even though there is no anomaly. To track reads we implement predicate locking, see storage/lmgr/predicate.c. Whenever a tuple is read, a predicate lock is acquired on the tuple. Shared memory is finite, so when a transaction takes many tuple-level locks on a page, the locks are promoted to a single page-level lock, and further to a single relation level lock if necessary. To lock key values with no matching tuple, a sequential scan always takes a relation-level lock, and an index scan acquires a page-level lock that covers the search key, whether or not there are any matching keys at the moment. A predicate lock doesn't conflict with any regular locks or with another predicate locks in the normal sense. They're only used by the predicate lock manager to detect the danger of anomalies. Only serializable transactions participate in predicate locking, so there should be no extra overhead for for other transactions. Predicate locks can't be released at commit, but must be remembered until all the transactions that overlapped with it have completed. That means that we need to remember an unbounded amount of predicate locks, so we apply a lossy but conservative method of tracking locks for committed transactions. If we run short of shared memory, we overflow to a new "pg_serial" SLRU pool. We don't currently allow Serializable transactions in Hot Standby mode. That would be hard, because even read-only transactions can cause anomalies that wouldn't otherwise occur. Serializable isolation mode now means the new fully serializable level. Repeatable Read gives you the old Snapshot Isolation level that we have always had. Kevin Grittner and Dan Ports, reviewed by Jeff Davis, Heikki Linnakangas and Anssi Kääriäinen
2011-02-07 22:46:51 +01:00
/* don't exit on teardown failure */
}
PQclear(res);
}
}
/* Perform teardown */
if (testspec->teardownsql)
{
Use annotations to reduce instability of isolation-test results. We've long contended with isolation test results that aren't entirely stable. Some test scripts insert long delays to try to force stable results, which is not terribly desirable; but other erratic failure modes remain, causing unrepeatable buildfarm failures. I've spent a fair amount of time trying to solve this by improving the server-side support code, without much success: that way is fundamentally unable to cope with diffs that stem from chance ordering of arrival of messages from different server processes. We can improve matters on the client side, however, by annotating the test scripts themselves to show the desired reporting order of events that might occur in different orders. This patch adds three types of annotations to deal with (a) test steps that might or might not complete their waits before the isolationtester can see them waiting; (b) test steps in different sessions that can legitimately complete in either order; and (c) NOTIFY messages that might arrive before or after the completion of a step in another session. We might need more annotation types later, but this seems to be enough to deal with the instabilities we've seen in the buildfarm. It also lets us get rid of all the long delays that were previously used, cutting more than a minute off the runtime of the isolation tests. Back-patch to all supported branches, because the buildfarm instabilities affect all the branches, and because it seems desirable to keep isolationtester's capabilities the same across all branches to simplify possible future back-patching of tests. Discussion: https://postgr.es/m/327948.1623725828@sss.pgh.pa.us
2021-06-23 03:43:12 +02:00
res = PQexec(conns[0].conn, testspec->teardownsql);
Add an SSI regression test that tests all interesting permutations in the order of begin, prepare, and commit of three concurrent transactions that have conflicts between them. The test runs for a quite long time, and the expected output file is huge, but this test caught some serious bugs during development, so seems worthwhile to keep. The test uses prepared transactions, so it fails if the server has max_prepared_transactions=0. Because of that, it's marked as "ignore" in the schedule file. Dan Ports
2011-08-18 12:36:37 +02:00
if (PQresultStatus(res) == PGRES_TUPLES_OK)
{
printResultSet(res);
}
else if (PQresultStatus(res) != PGRES_COMMAND_OK)
Implement genuine serializable isolation level. Until now, our Serializable mode has in fact been what's called Snapshot Isolation, which allows some anomalies that could not occur in any serialized ordering of the transactions. This patch fixes that using a method called Serializable Snapshot Isolation, based on research papers by Michael J. Cahill (see README-SSI for full references). In Serializable Snapshot Isolation, transactions run like they do in Snapshot Isolation, but a predicate lock manager observes the reads and writes performed and aborts transactions if it detects that an anomaly might occur. This method produces some false positives, ie. it sometimes aborts transactions even though there is no anomaly. To track reads we implement predicate locking, see storage/lmgr/predicate.c. Whenever a tuple is read, a predicate lock is acquired on the tuple. Shared memory is finite, so when a transaction takes many tuple-level locks on a page, the locks are promoted to a single page-level lock, and further to a single relation level lock if necessary. To lock key values with no matching tuple, a sequential scan always takes a relation-level lock, and an index scan acquires a page-level lock that covers the search key, whether or not there are any matching keys at the moment. A predicate lock doesn't conflict with any regular locks or with another predicate locks in the normal sense. They're only used by the predicate lock manager to detect the danger of anomalies. Only serializable transactions participate in predicate locking, so there should be no extra overhead for for other transactions. Predicate locks can't be released at commit, but must be remembered until all the transactions that overlapped with it have completed. That means that we need to remember an unbounded amount of predicate locks, so we apply a lossy but conservative method of tracking locks for committed transactions. If we run short of shared memory, we overflow to a new "pg_serial" SLRU pool. We don't currently allow Serializable transactions in Hot Standby mode. That would be hard, because even read-only transactions can cause anomalies that wouldn't otherwise occur. Serializable isolation mode now means the new fully serializable level. Repeatable Read gives you the old Snapshot Isolation level that we have always had. Kevin Grittner and Dan Ports, reviewed by Jeff Davis, Heikki Linnakangas and Anssi Kääriäinen
2011-02-07 22:46:51 +01:00
{
fprintf(stderr, "teardown failed: %s",
Use annotations to reduce instability of isolation-test results. We've long contended with isolation test results that aren't entirely stable. Some test scripts insert long delays to try to force stable results, which is not terribly desirable; but other erratic failure modes remain, causing unrepeatable buildfarm failures. I've spent a fair amount of time trying to solve this by improving the server-side support code, without much success: that way is fundamentally unable to cope with diffs that stem from chance ordering of arrival of messages from different server processes. We can improve matters on the client side, however, by annotating the test scripts themselves to show the desired reporting order of events that might occur in different orders. This patch adds three types of annotations to deal with (a) test steps that might or might not complete their waits before the isolationtester can see them waiting; (b) test steps in different sessions that can legitimately complete in either order; and (c) NOTIFY messages that might arrive before or after the completion of a step in another session. We might need more annotation types later, but this seems to be enough to deal with the instabilities we've seen in the buildfarm. It also lets us get rid of all the long delays that were previously used, cutting more than a minute off the runtime of the isolation tests. Back-patch to all supported branches, because the buildfarm instabilities affect all the branches, and because it seems desirable to keep isolationtester's capabilities the same across all branches to simplify possible future back-patching of tests. Discussion: https://postgr.es/m/327948.1623725828@sss.pgh.pa.us
2021-06-23 03:43:12 +02:00
PQerrorMessage(conns[0].conn));
Implement genuine serializable isolation level. Until now, our Serializable mode has in fact been what's called Snapshot Isolation, which allows some anomalies that could not occur in any serialized ordering of the transactions. This patch fixes that using a method called Serializable Snapshot Isolation, based on research papers by Michael J. Cahill (see README-SSI for full references). In Serializable Snapshot Isolation, transactions run like they do in Snapshot Isolation, but a predicate lock manager observes the reads and writes performed and aborts transactions if it detects that an anomaly might occur. This method produces some false positives, ie. it sometimes aborts transactions even though there is no anomaly. To track reads we implement predicate locking, see storage/lmgr/predicate.c. Whenever a tuple is read, a predicate lock is acquired on the tuple. Shared memory is finite, so when a transaction takes many tuple-level locks on a page, the locks are promoted to a single page-level lock, and further to a single relation level lock if necessary. To lock key values with no matching tuple, a sequential scan always takes a relation-level lock, and an index scan acquires a page-level lock that covers the search key, whether or not there are any matching keys at the moment. A predicate lock doesn't conflict with any regular locks or with another predicate locks in the normal sense. They're only used by the predicate lock manager to detect the danger of anomalies. Only serializable transactions participate in predicate locking, so there should be no extra overhead for for other transactions. Predicate locks can't be released at commit, but must be remembered until all the transactions that overlapped with it have completed. That means that we need to remember an unbounded amount of predicate locks, so we apply a lossy but conservative method of tracking locks for committed transactions. If we run short of shared memory, we overflow to a new "pg_serial" SLRU pool. We don't currently allow Serializable transactions in Hot Standby mode. That would be hard, because even read-only transactions can cause anomalies that wouldn't otherwise occur. Serializable isolation mode now means the new fully serializable level. Repeatable Read gives you the old Snapshot Isolation level that we have always had. Kevin Grittner and Dan Ports, reviewed by Jeff Davis, Heikki Linnakangas and Anssi Kääriäinen
2011-02-07 22:46:51 +01:00
/* don't exit on teardown failure */
}
PQclear(res);
}
Code review for isolationtester changes. Fix a few oversights in 38f8bdcac4982215beb9f65a19debecaf22fd470: don't leak memory in run_permutation(), remember when we've issued a cancel rather than issuing another one every 10ms, fix some typos in comments.
2016-02-11 17:30:46 +01:00
free(waiting);
Use annotations to reduce instability of isolation-test results. We've long contended with isolation test results that aren't entirely stable. Some test scripts insert long delays to try to force stable results, which is not terribly desirable; but other erratic failure modes remain, causing unrepeatable buildfarm failures. I've spent a fair amount of time trying to solve this by improving the server-side support code, without much success: that way is fundamentally unable to cope with diffs that stem from chance ordering of arrival of messages from different server processes. We can improve matters on the client side, however, by annotating the test scripts themselves to show the desired reporting order of events that might occur in different orders. This patch adds three types of annotations to deal with (a) test steps that might or might not complete their waits before the isolationtester can see them waiting; (b) test steps in different sessions that can legitimately complete in either order; and (c) NOTIFY messages that might arrive before or after the completion of a step in another session. We might need more annotation types later, but this seems to be enough to deal with the instabilities we've seen in the buildfarm. It also lets us get rid of all the long delays that were previously used, cutting more than a minute off the runtime of the isolation tests. Back-patch to all supported branches, because the buildfarm instabilities affect all the branches, and because it seems desirable to keep isolationtester's capabilities the same across all branches to simplify possible future back-patching of tests. Discussion: https://postgr.es/m/327948.1623725828@sss.pgh.pa.us
2021-06-23 03:43:12 +02:00
}
/*
* Check for completion of any waiting step(s).
* Remove completed ones from the waiting[] array,
* and return the new value of nwaiting.
* See try_complete_step for the meaning of the flags.
*/
static int
try_complete_steps(TestSpec *testspec, PermutationStep **waiting,
int nwaiting, int flags)
{
int old_nwaiting;
bool have_blocker;
do
{
int w = 0;
/* Reset latch; we only care about notices received within loop. */
any_new_notice = false;
/* Likewise, these variables reset for each retry. */
old_nwaiting = nwaiting;
have_blocker = false;
/* Scan the array, try to complete steps. */
while (w < nwaiting)
{
if (try_complete_step(testspec, waiting[w], flags))
{
/* Still blocked, leave it alone. */
if (waiting[w]->nblockers > 0)
have_blocker = true;
w++;
}
else
{
/* Done, remove it from array. */
if (w + 1 < nwaiting)
memmove(&waiting[w], &waiting[w + 1],
(nwaiting - (w + 1)) * sizeof(PermutationStep *));
nwaiting--;
}
}
/*
* If any of the still-waiting steps have blocker conditions attached,
* it's possible that one of the steps we examined afterwards has
* released them (either by completing, or by sending a NOTICE). If
* any step completions or NOTICEs happened, repeat the loop until
* none occurs. Without this provision, completion timing could vary
* depending on the order in which the steps appear in the array.
*/
} while (have_blocker && (nwaiting < old_nwaiting || any_new_notice));
return nwaiting;
Implement genuine serializable isolation level. Until now, our Serializable mode has in fact been what's called Snapshot Isolation, which allows some anomalies that could not occur in any serialized ordering of the transactions. This patch fixes that using a method called Serializable Snapshot Isolation, based on research papers by Michael J. Cahill (see README-SSI for full references). In Serializable Snapshot Isolation, transactions run like they do in Snapshot Isolation, but a predicate lock manager observes the reads and writes performed and aborts transactions if it detects that an anomaly might occur. This method produces some false positives, ie. it sometimes aborts transactions even though there is no anomaly. To track reads we implement predicate locking, see storage/lmgr/predicate.c. Whenever a tuple is read, a predicate lock is acquired on the tuple. Shared memory is finite, so when a transaction takes many tuple-level locks on a page, the locks are promoted to a single page-level lock, and further to a single relation level lock if necessary. To lock key values with no matching tuple, a sequential scan always takes a relation-level lock, and an index scan acquires a page-level lock that covers the search key, whether or not there are any matching keys at the moment. A predicate lock doesn't conflict with any regular locks or with another predicate locks in the normal sense. They're only used by the predicate lock manager to detect the danger of anomalies. Only serializable transactions participate in predicate locking, so there should be no extra overhead for for other transactions. Predicate locks can't be released at commit, but must be remembered until all the transactions that overlapped with it have completed. That means that we need to remember an unbounded amount of predicate locks, so we apply a lossy but conservative method of tracking locks for committed transactions. If we run short of shared memory, we overflow to a new "pg_serial" SLRU pool. We don't currently allow Serializable transactions in Hot Standby mode. That would be hard, because even read-only transactions can cause anomalies that wouldn't otherwise occur. Serializable isolation mode now means the new fully serializable level. Repeatable Read gives you the old Snapshot Isolation level that we have always had. Kevin Grittner and Dan Ports, reviewed by Jeff Davis, Heikki Linnakangas and Anssi Kääriäinen
2011-02-07 22:46:51 +01:00
}
Add support for blocked commands in isolationtester This enables us to test that blocking commands (such as foreign keys checks that conflict with some other lock) act as intended. The set of tests that this adds is pretty minimal, but can easily be extended by adding new specs. The intention is that this will serve as a basis for ensuring that further tweaks of locking implementation preserve (or improve) existing behavior. Author: Noah Misch
2011-07-12 22:35:09 +02:00
/*
* Our caller already sent the query associated with this step. Wait for it
Use annotations to reduce instability of isolation-test results. We've long contended with isolation test results that aren't entirely stable. Some test scripts insert long delays to try to force stable results, which is not terribly desirable; but other erratic failure modes remain, causing unrepeatable buildfarm failures. I've spent a fair amount of time trying to solve this by improving the server-side support code, without much success: that way is fundamentally unable to cope with diffs that stem from chance ordering of arrival of messages from different server processes. We can improve matters on the client side, however, by annotating the test scripts themselves to show the desired reporting order of events that might occur in different orders. This patch adds three types of annotations to deal with (a) test steps that might or might not complete their waits before the isolationtester can see them waiting; (b) test steps in different sessions that can legitimately complete in either order; and (c) NOTIFY messages that might arrive before or after the completion of a step in another session. We might need more annotation types later, but this seems to be enough to deal with the instabilities we've seen in the buildfarm. It also lets us get rid of all the long delays that were previously used, cutting more than a minute off the runtime of the isolation tests. Back-patch to all supported branches, because the buildfarm instabilities affect all the branches, and because it seems desirable to keep isolationtester's capabilities the same across all branches to simplify possible future back-patching of tests. Discussion: https://postgr.es/m/327948.1623725828@sss.pgh.pa.us
2021-06-23 03:43:12 +02:00
* to either complete, or hit a blocking condition.
Add support for blocked commands in isolationtester This enables us to test that blocking commands (such as foreign keys checks that conflict with some other lock) act as intended. The set of tests that this adds is pretty minimal, but can easily be extended by adding new specs. The intention is that this will serve as a basis for ensuring that further tweaks of locking implementation preserve (or improve) existing behavior. Author: Noah Misch
2011-07-12 22:35:09 +02:00
*
* When calling this function on behalf of a given step for a second or later
Use annotations to reduce instability of isolation-test results. We've long contended with isolation test results that aren't entirely stable. Some test scripts insert long delays to try to force stable results, which is not terribly desirable; but other erratic failure modes remain, causing unrepeatable buildfarm failures. I've spent a fair amount of time trying to solve this by improving the server-side support code, without much success: that way is fundamentally unable to cope with diffs that stem from chance ordering of arrival of messages from different server processes. We can improve matters on the client side, however, by annotating the test scripts themselves to show the desired reporting order of events that might occur in different orders. This patch adds three types of annotations to deal with (a) test steps that might or might not complete their waits before the isolationtester can see them waiting; (b) test steps in different sessions that can legitimately complete in either order; and (c) NOTIFY messages that might arrive before or after the completion of a step in another session. We might need more annotation types later, but this seems to be enough to deal with the instabilities we've seen in the buildfarm. It also lets us get rid of all the long delays that were previously used, cutting more than a minute off the runtime of the isolation tests. Back-patch to all supported branches, because the buildfarm instabilities affect all the branches, and because it seems desirable to keep isolationtester's capabilities the same across all branches to simplify possible future back-patching of tests. Discussion: https://postgr.es/m/327948.1623725828@sss.pgh.pa.us
2021-06-23 03:43:12 +02:00
* time, pass the STEP_RETRY flag. Do not pass it on the first call.
Remove dependency on error ordering in isolation tests We now report errors reported by the just-unblocked and unblocking transactions identically; this should fix relatively common buildfarm failures reported by animals that are failing the "wrong" session.
2011-09-27 21:08:31 +02:00
*
Use annotations to reduce instability of isolation-test results. We've long contended with isolation test results that aren't entirely stable. Some test scripts insert long delays to try to force stable results, which is not terribly desirable; but other erratic failure modes remain, causing unrepeatable buildfarm failures. I've spent a fair amount of time trying to solve this by improving the server-side support code, without much success: that way is fundamentally unable to cope with diffs that stem from chance ordering of arrival of messages from different server processes. We can improve matters on the client side, however, by annotating the test scripts themselves to show the desired reporting order of events that might occur in different orders. This patch adds three types of annotations to deal with (a) test steps that might or might not complete their waits before the isolationtester can see them waiting; (b) test steps in different sessions that can legitimately complete in either order; and (c) NOTIFY messages that might arrive before or after the completion of a step in another session. We might need more annotation types later, but this seems to be enough to deal with the instabilities we've seen in the buildfarm. It also lets us get rid of all the long delays that were previously used, cutting more than a minute off the runtime of the isolation tests. Back-patch to all supported branches, because the buildfarm instabilities affect all the branches, and because it seems desirable to keep isolationtester's capabilities the same across all branches to simplify possible future back-patching of tests. Discussion: https://postgr.es/m/327948.1623725828@sss.pgh.pa.us
2021-06-23 03:43:12 +02:00
* Returns true if the step was *not* completed, false if it was completed.
* Reasons for non-completion are (a) the STEP_NONBLOCK flag was specified
* and the query is waiting to acquire a lock, or (b) the step has an
* unsatisfied blocker condition. When STEP_NONBLOCK is given, we assume
* that any lock wait will persist until we have executed additional steps.
Add support for blocked commands in isolationtester This enables us to test that blocking commands (such as foreign keys checks that conflict with some other lock) act as intended. The set of tests that this adds is pretty minimal, but can easily be extended by adding new specs. The intention is that this will serve as a basis for ensuring that further tweaks of locking implementation preserve (or improve) existing behavior. Author: Noah Misch
2011-07-12 22:35:09 +02:00
*/
static bool
Use annotations to reduce instability of isolation-test results. We've long contended with isolation test results that aren't entirely stable. Some test scripts insert long delays to try to force stable results, which is not terribly desirable; but other erratic failure modes remain, causing unrepeatable buildfarm failures. I've spent a fair amount of time trying to solve this by improving the server-side support code, without much success: that way is fundamentally unable to cope with diffs that stem from chance ordering of arrival of messages from different server processes. We can improve matters on the client side, however, by annotating the test scripts themselves to show the desired reporting order of events that might occur in different orders. This patch adds three types of annotations to deal with (a) test steps that might or might not complete their waits before the isolationtester can see them waiting; (b) test steps in different sessions that can legitimately complete in either order; and (c) NOTIFY messages that might arrive before or after the completion of a step in another session. We might need more annotation types later, but this seems to be enough to deal with the instabilities we've seen in the buildfarm. It also lets us get rid of all the long delays that were previously used, cutting more than a minute off the runtime of the isolation tests. Back-patch to all supported branches, because the buildfarm instabilities affect all the branches, and because it seems desirable to keep isolationtester's capabilities the same across all branches to simplify possible future back-patching of tests. Discussion: https://postgr.es/m/327948.1623725828@sss.pgh.pa.us
2021-06-23 03:43:12 +02:00
try_complete_step(TestSpec *testspec, PermutationStep *pstep, int flags)
Add support for blocked commands in isolationtester This enables us to test that blocking commands (such as foreign keys checks that conflict with some other lock) act as intended. The set of tests that this adds is pretty minimal, but can easily be extended by adding new specs. The intention is that this will serve as a basis for ensuring that further tweaks of locking implementation preserve (or improve) existing behavior. Author: Noah Misch
2011-07-12 22:35:09 +02:00
{
Use annotations to reduce instability of isolation-test results. We've long contended with isolation test results that aren't entirely stable. Some test scripts insert long delays to try to force stable results, which is not terribly desirable; but other erratic failure modes remain, causing unrepeatable buildfarm failures. I've spent a fair amount of time trying to solve this by improving the server-side support code, without much success: that way is fundamentally unable to cope with diffs that stem from chance ordering of arrival of messages from different server processes. We can improve matters on the client side, however, by annotating the test scripts themselves to show the desired reporting order of events that might occur in different orders. This patch adds three types of annotations to deal with (a) test steps that might or might not complete their waits before the isolationtester can see them waiting; (b) test steps in different sessions that can legitimately complete in either order; and (c) NOTIFY messages that might arrive before or after the completion of a step in another session. We might need more annotation types later, but this seems to be enough to deal with the instabilities we've seen in the buildfarm. It also lets us get rid of all the long delays that were previously used, cutting more than a minute off the runtime of the isolation tests. Back-patch to all supported branches, because the buildfarm instabilities affect all the branches, and because it seems desirable to keep isolationtester's capabilities the same across all branches to simplify possible future back-patching of tests. Discussion: https://postgr.es/m/327948.1623725828@sss.pgh.pa.us
2021-06-23 03:43:12 +02:00
Step *step = pstep->step;
IsoConnInfo *iconn = &conns[1 + step->session];
PGconn *conn = iconn->conn;
Add support for blocked commands in isolationtester This enables us to test that blocking commands (such as foreign keys checks that conflict with some other lock) act as intended. The set of tests that this adds is pretty minimal, but can easily be extended by adding new specs. The intention is that this will serve as a basis for ensuring that further tweaks of locking implementation preserve (or improve) existing behavior. Author: Noah Misch
2011-07-12 22:35:09 +02:00
fd_set read_set;
Modify the isolation tester so that multiple sessions can wait. This allows testing of deadlock scenarios. Scenarios that would previously have been considered invalid are now simply taken as a scenario in which more than one backend will wait.
2016-02-11 14:36:30 +01:00
struct timeval start_time;
Add support for blocked commands in isolationtester This enables us to test that blocking commands (such as foreign keys checks that conflict with some other lock) act as intended. The set of tests that this adds is pretty minimal, but can easily be extended by adding new specs. The intention is that this will serve as a basis for ensuring that further tweaks of locking implementation preserve (or improve) existing behavior. Author: Noah Misch
2011-07-12 22:35:09 +02:00
struct timeval timeout;
int sock = PQsocket(conn);
int ret;
PGresult *res;
Improve test coverage for LISTEN/NOTIFY. We had no actual end-to-end test of NOTIFY message delivery. In the core async.sql regression test, testing this is problematic because psql traditionally prints the PID of the sending backend, making the output unstable. We also have an isolation test script, but it likewise failed to prove that delivery worked, because isolationtester.c had no provisions for detecting/reporting NOTIFY messages. Hence, add such provisions to isolationtester.c, and extend async-notify.spec to include direct tests of basic NOTIFY functionality. I also added tests showing that NOTIFY de-duplicates messages normally, but not across subtransaction boundaries. (That's the historical behavior since we introduced subtransactions, though perhaps we ought to change it.) Patch by me, with suggestions/review by Andres Freund. Discussion: https://postgr.es/m/31304.1564246011@sss.pgh.pa.us
2019-07-28 18:02:27 +02:00
PGnotify *notify;
Modify the isolation tester so that multiple sessions can wait. This allows testing of deadlock scenarios. Scenarios that would previously have been considered invalid are now simply taken as a scenario in which more than one backend will wait.
2016-02-11 14:36:30 +01:00
bool canceled = false;
Add support for blocked commands in isolationtester This enables us to test that blocking commands (such as foreign keys checks that conflict with some other lock) act as intended. The set of tests that this adds is pretty minimal, but can easily be extended by adding new specs. The intention is that this will serve as a basis for ensuring that further tweaks of locking implementation preserve (or improve) existing behavior. Author: Noah Misch
2011-07-12 22:35:09 +02:00
Use annotations to reduce instability of isolation-test results. We've long contended with isolation test results that aren't entirely stable. Some test scripts insert long delays to try to force stable results, which is not terribly desirable; but other erratic failure modes remain, causing unrepeatable buildfarm failures. I've spent a fair amount of time trying to solve this by improving the server-side support code, without much success: that way is fundamentally unable to cope with diffs that stem from chance ordering of arrival of messages from different server processes. We can improve matters on the client side, however, by annotating the test scripts themselves to show the desired reporting order of events that might occur in different orders. This patch adds three types of annotations to deal with (a) test steps that might or might not complete their waits before the isolationtester can see them waiting; (b) test steps in different sessions that can legitimately complete in either order; and (c) NOTIFY messages that might arrive before or after the completion of a step in another session. We might need more annotation types later, but this seems to be enough to deal with the instabilities we've seen in the buildfarm. It also lets us get rid of all the long delays that were previously used, cutting more than a minute off the runtime of the isolation tests. Back-patch to all supported branches, because the buildfarm instabilities affect all the branches, and because it seems desirable to keep isolationtester's capabilities the same across all branches to simplify possible future back-patching of tests. Discussion: https://postgr.es/m/327948.1623725828@sss.pgh.pa.us
2021-06-23 03:43:12 +02:00
/*
* If the step is annotated with (*), then on the first call, force it to
* wait. This is useful for ensuring consistent output when the step
* might or might not complete so fast that we don't observe it waiting.
*/
if (!(flags & STEP_RETRY))
{
int i;
for (i = 0; i < pstep->nblockers; i++)
{
PermutationStepBlocker *blocker = pstep->blockers[i];
if (blocker->blocktype == PSB_ONCE)
{
printf("step %s: %s <waiting ...>\n",
step->name, step->sql);
return true;
}
}
}
Handle invalid libpq sockets in more places Also, make error messages consistent. From: Michael Paquier <michael.paquier@gmail.com>
2016-03-09 03:10:33 +01:00
if (sock < 0)
{
fprintf(stderr, "invalid socket: %s", PQerrorMessage(conn));
isolationtester: Use atexit() Replace exit_nicely() calls with standard exit() and register the cleanup actions using atexit(). Reviewed-by: Alvaro Herrera <alvherre@2ndquadrant.com> Discussion: https://www.postgresql.org/message-id/flat/ec4135ba-84e9-28bf-b584-0e78d47448d5@2ndquadrant.com/
2019-01-05 15:05:49 +01:00
exit(1);
Handle invalid libpq sockets in more places Also, make error messages consistent. From: Michael Paquier <michael.paquier@gmail.com>
2016-03-09 03:10:33 +01:00
}
Modify the isolation tester so that multiple sessions can wait. This allows testing of deadlock scenarios. Scenarios that would previously have been considered invalid are now simply taken as a scenario in which more than one backend will wait.
2016-02-11 14:36:30 +01:00
gettimeofday(&start_time, NULL);
Add support for blocked commands in isolationtester This enables us to test that blocking commands (such as foreign keys checks that conflict with some other lock) act as intended. The set of tests that this adds is pretty minimal, but can easily be extended by adding new specs. The intention is that this will serve as a basis for ensuring that further tweaks of locking implementation preserve (or improve) existing behavior. Author: Noah Misch
2011-07-12 22:35:09 +02:00
FD_ZERO(&read_set);
Modify the isolation tester so that multiple sessions can wait. This allows testing of deadlock scenarios. Scenarios that would previously have been considered invalid are now simply taken as a scenario in which more than one backend will wait.
2016-02-11 14:36:30 +01:00
while (PQisBusy(conn))
Add support for blocked commands in isolationtester This enables us to test that blocking commands (such as foreign keys checks that conflict with some other lock) act as intended. The set of tests that this adds is pretty minimal, but can easily be extended by adding new specs. The intention is that this will serve as a basis for ensuring that further tweaks of locking implementation preserve (or improve) existing behavior. Author: Noah Misch
2011-07-12 22:35:09 +02:00
{
FD_SET(sock, &read_set);
timeout.tv_sec = 0;
timeout.tv_usec = 10000; /* Check for lock waits every 10ms. */
ret = select(sock + 1, &read_set, NULL, NULL, &timeout);
if (ret < 0) /* error in select() */
{
In isolationtester, retry after EINTR return from select(2). Per report from Jaime Casanova. Very curious that no one else has seen this failure ... but the code is clearly wrong as-is.
2013-04-07 04:28:49 +02:00
if (errno == EINTR)
continue;
Add support for blocked commands in isolationtester This enables us to test that blocking commands (such as foreign keys checks that conflict with some other lock) act as intended. The set of tests that this adds is pretty minimal, but can easily be extended by adding new specs. The intention is that this will serve as a basis for ensuring that further tweaks of locking implementation preserve (or improve) existing behavior. Author: Noah Misch
2011-07-12 22:35:09 +02:00
fprintf(stderr, "select failed: %s\n", strerror(errno));
isolationtester: Use atexit() Replace exit_nicely() calls with standard exit() and register the cleanup actions using atexit(). Reviewed-by: Alvaro Herrera <alvherre@2ndquadrant.com> Discussion: https://www.postgresql.org/message-id/flat/ec4135ba-84e9-28bf-b584-0e78d47448d5@2ndquadrant.com/
2019-01-05 15:05:49 +01:00
exit(1);
Add support for blocked commands in isolationtester This enables us to test that blocking commands (such as foreign keys checks that conflict with some other lock) act as intended. The set of tests that this adds is pretty minimal, but can easily be extended by adding new specs. The intention is that this will serve as a basis for ensuring that further tweaks of locking implementation preserve (or improve) existing behavior. Author: Noah Misch
2011-07-12 22:35:09 +02:00
}
else if (ret == 0) /* select() timeout: check for lock wait */
{
Modify the isolation tester so that multiple sessions can wait. This allows testing of deadlock scenarios. Scenarios that would previously have been considered invalid are now simply taken as a scenario in which more than one backend will wait.
2016-02-11 14:36:30 +01:00
struct timeval current_time;
int64 td;
Add support for blocked commands in isolationtester This enables us to test that blocking commands (such as foreign keys checks that conflict with some other lock) act as intended. The set of tests that this adds is pretty minimal, but can easily be extended by adding new specs. The intention is that this will serve as a basis for ensuring that further tweaks of locking implementation preserve (or improve) existing behavior. Author: Noah Misch
2011-07-12 22:35:09 +02:00
Modify the isolation tester so that multiple sessions can wait. This allows testing of deadlock scenarios. Scenarios that would previously have been considered invalid are now simply taken as a scenario in which more than one backend will wait.
2016-02-11 14:36:30 +01:00
/* If it's OK for the step to block, check whether it has. */
if (flags & STEP_NONBLOCK)
Add support for blocked commands in isolationtester This enables us to test that blocking commands (such as foreign keys checks that conflict with some other lock) act as intended. The set of tests that this adds is pretty minimal, but can easily be extended by adding new specs. The intention is that this will serve as a basis for ensuring that further tweaks of locking implementation preserve (or improve) existing behavior. Author: Noah Misch
2011-07-12 22:35:09 +02:00
{
Create a function to reliably identify which sessions block which others. This patch introduces "pg_blocking_pids(int) returns int[]", which returns the PIDs of any sessions that are blocking the session with the given PID. Historically people have obtained such information using a self-join on the pg_locks view, but it's unreasonably tedious to do it that way with any modicum of correctness, and the addition of parallel queries has pretty much broken that approach altogether. (Given some more columns in the view than there are today, you could imagine handling parallel-query cases with a 4-way join; but ugh.) The new function has the following behaviors that are painful or impossible to get right via pg_locks: 1. Correctly understands which lock modes block which other ones. 2. In soft-block situations (two processes both waiting for conflicting lock modes), only the one that's in front in the wait queue is reported to block the other. 3. In parallel-query cases, reports all sessions blocking any member of the given PID's lock group, and reports a session by naming its leader process's PID, which will be the pg_backend_pid() value visible to clients. The motivation for doing this right now is mostly to fix the isolation tests. Commit 38f8bdcac4982215beb9f65a19debecaf22fd470 lobotomized isolationtester's is-it-waiting query by removing its ability to recognize nonconflicting lock modes, as a crude workaround for the inability to handle soft-block situations properly. But even without the lock mode tests, the old query was excessively slow, particularly in CLOBBER_CACHE_ALWAYS builds; some of our buildfarm animals fail the new deadlock-hard test because the deadlock timeout elapses before they can probe the waiting status of all eight sessions. Replacing the pg_locks self-join with use of pg_blocking_pids() is not only much more correct, but a lot faster: I measure it at about 9X faster in a typical dev build with Asserts, and 3X faster in CLOBBER_CACHE_ALWAYS builds. That should provide enough headroom for the slower CLOBBER_CACHE_ALWAYS animals to pass the test, without having to lengthen deadlock_timeout yet more and thus slow down the test for everyone else.
2016-02-22 20:31:43 +01:00
bool waiting;
Modify the isolation tester so that multiple sessions can wait. This allows testing of deadlock scenarios. Scenarios that would previously have been considered invalid are now simply taken as a scenario in which more than one backend will wait.
2016-02-11 14:36:30 +01:00
Use annotations to reduce instability of isolation-test results. We've long contended with isolation test results that aren't entirely stable. Some test scripts insert long delays to try to force stable results, which is not terribly desirable; but other erratic failure modes remain, causing unrepeatable buildfarm failures. I've spent a fair amount of time trying to solve this by improving the server-side support code, without much success: that way is fundamentally unable to cope with diffs that stem from chance ordering of arrival of messages from different server processes. We can improve matters on the client side, however, by annotating the test scripts themselves to show the desired reporting order of events that might occur in different orders. This patch adds three types of annotations to deal with (a) test steps that might or might not complete their waits before the isolationtester can see them waiting; (b) test steps in different sessions that can legitimately complete in either order; and (c) NOTIFY messages that might arrive before or after the completion of a step in another session. We might need more annotation types later, but this seems to be enough to deal with the instabilities we've seen in the buildfarm. It also lets us get rid of all the long delays that were previously used, cutting more than a minute off the runtime of the isolation tests. Back-patch to all supported branches, because the buildfarm instabilities affect all the branches, and because it seems desirable to keep isolationtester's capabilities the same across all branches to simplify possible future back-patching of tests. Discussion: https://postgr.es/m/327948.1623725828@sss.pgh.pa.us
2021-06-23 03:43:12 +02:00
res = PQexecPrepared(conns[0].conn, PREP_WAITING, 1,
&conns[step->session + 1].backend_pid_str,
Modify the isolation tester so that multiple sessions can wait. This allows testing of deadlock scenarios. Scenarios that would previously have been considered invalid are now simply taken as a scenario in which more than one backend will wait.
2016-02-11 14:36:30 +01:00
NULL, NULL, 0);
Create a function to reliably identify which sessions block which others. This patch introduces "pg_blocking_pids(int) returns int[]", which returns the PIDs of any sessions that are blocking the session with the given PID. Historically people have obtained such information using a self-join on the pg_locks view, but it's unreasonably tedious to do it that way with any modicum of correctness, and the addition of parallel queries has pretty much broken that approach altogether. (Given some more columns in the view than there are today, you could imagine handling parallel-query cases with a 4-way join; but ugh.) The new function has the following behaviors that are painful or impossible to get right via pg_locks: 1. Correctly understands which lock modes block which other ones. 2. In soft-block situations (two processes both waiting for conflicting lock modes), only the one that's in front in the wait queue is reported to block the other. 3. In parallel-query cases, reports all sessions blocking any member of the given PID's lock group, and reports a session by naming its leader process's PID, which will be the pg_backend_pid() value visible to clients. The motivation for doing this right now is mostly to fix the isolation tests. Commit 38f8bdcac4982215beb9f65a19debecaf22fd470 lobotomized isolationtester's is-it-waiting query by removing its ability to recognize nonconflicting lock modes, as a crude workaround for the inability to handle soft-block situations properly. But even without the lock mode tests, the old query was excessively slow, particularly in CLOBBER_CACHE_ALWAYS builds; some of our buildfarm animals fail the new deadlock-hard test because the deadlock timeout elapses before they can probe the waiting status of all eight sessions. Replacing the pg_locks self-join with use of pg_blocking_pids() is not only much more correct, but a lot faster: I measure it at about 9X faster in a typical dev build with Asserts, and 3X faster in CLOBBER_CACHE_ALWAYS builds. That should provide enough headroom for the slower CLOBBER_CACHE_ALWAYS animals to pass the test, without having to lengthen deadlock_timeout yet more and thus slow down the test for everyone else.
2016-02-22 20:31:43 +01:00
if (PQresultStatus(res) != PGRES_TUPLES_OK ||
PQntuples(res) != 1)
Modify the isolation tester so that multiple sessions can wait. This allows testing of deadlock scenarios. Scenarios that would previously have been considered invalid are now simply taken as a scenario in which more than one backend will wait.
2016-02-11 14:36:30 +01:00
{
fprintf(stderr, "lock wait query failed: %s",
Use annotations to reduce instability of isolation-test results. We've long contended with isolation test results that aren't entirely stable. Some test scripts insert long delays to try to force stable results, which is not terribly desirable; but other erratic failure modes remain, causing unrepeatable buildfarm failures. I've spent a fair amount of time trying to solve this by improving the server-side support code, without much success: that way is fundamentally unable to cope with diffs that stem from chance ordering of arrival of messages from different server processes. We can improve matters on the client side, however, by annotating the test scripts themselves to show the desired reporting order of events that might occur in different orders. This patch adds three types of annotations to deal with (a) test steps that might or might not complete their waits before the isolationtester can see them waiting; (b) test steps in different sessions that can legitimately complete in either order; and (c) NOTIFY messages that might arrive before or after the completion of a step in another session. We might need more annotation types later, but this seems to be enough to deal with the instabilities we've seen in the buildfarm. It also lets us get rid of all the long delays that were previously used, cutting more than a minute off the runtime of the isolation tests. Back-patch to all supported branches, because the buildfarm instabilities affect all the branches, and because it seems desirable to keep isolationtester's capabilities the same across all branches to simplify possible future back-patching of tests. Discussion: https://postgr.es/m/327948.1623725828@sss.pgh.pa.us
2021-06-23 03:43:12 +02:00
PQerrorMessage(conns[0].conn));
isolationtester: Use atexit() Replace exit_nicely() calls with standard exit() and register the cleanup actions using atexit(). Reviewed-by: Alvaro Herrera <alvherre@2ndquadrant.com> Discussion: https://www.postgresql.org/message-id/flat/ec4135ba-84e9-28bf-b584-0e78d47448d5@2ndquadrant.com/
2019-01-05 15:05:49 +01:00
exit(1);
Modify the isolation tester so that multiple sessions can wait. This allows testing of deadlock scenarios. Scenarios that would previously have been considered invalid are now simply taken as a scenario in which more than one backend will wait.
2016-02-11 14:36:30 +01:00
}
Create a function to reliably identify which sessions block which others. This patch introduces "pg_blocking_pids(int) returns int[]", which returns the PIDs of any sessions that are blocking the session with the given PID. Historically people have obtained such information using a self-join on the pg_locks view, but it's unreasonably tedious to do it that way with any modicum of correctness, and the addition of parallel queries has pretty much broken that approach altogether. (Given some more columns in the view than there are today, you could imagine handling parallel-query cases with a 4-way join; but ugh.) The new function has the following behaviors that are painful or impossible to get right via pg_locks: 1. Correctly understands which lock modes block which other ones. 2. In soft-block situations (two processes both waiting for conflicting lock modes), only the one that's in front in the wait queue is reported to block the other. 3. In parallel-query cases, reports all sessions blocking any member of the given PID's lock group, and reports a session by naming its leader process's PID, which will be the pg_backend_pid() value visible to clients. The motivation for doing this right now is mostly to fix the isolation tests. Commit 38f8bdcac4982215beb9f65a19debecaf22fd470 lobotomized isolationtester's is-it-waiting query by removing its ability to recognize nonconflicting lock modes, as a crude workaround for the inability to handle soft-block situations properly. But even without the lock mode tests, the old query was excessively slow, particularly in CLOBBER_CACHE_ALWAYS builds; some of our buildfarm animals fail the new deadlock-hard test because the deadlock timeout elapses before they can probe the waiting status of all eight sessions. Replacing the pg_locks self-join with use of pg_blocking_pids() is not only much more correct, but a lot faster: I measure it at about 9X faster in a typical dev build with Asserts, and 3X faster in CLOBBER_CACHE_ALWAYS builds. That should provide enough headroom for the slower CLOBBER_CACHE_ALWAYS animals to pass the test, without having to lengthen deadlock_timeout yet more and thus slow down the test for everyone else.
2016-02-22 20:31:43 +01:00
waiting = ((PQgetvalue(res, 0, 0))[0] == 't');
Modify the isolation tester so that multiple sessions can wait. This allows testing of deadlock scenarios. Scenarios that would previously have been considered invalid are now simply taken as a scenario in which more than one backend will wait.
2016-02-11 14:36:30 +01:00
PQclear(res);
Create a function to reliably identify which sessions block which others. This patch introduces "pg_blocking_pids(int) returns int[]", which returns the PIDs of any sessions that are blocking the session with the given PID. Historically people have obtained such information using a self-join on the pg_locks view, but it's unreasonably tedious to do it that way with any modicum of correctness, and the addition of parallel queries has pretty much broken that approach altogether. (Given some more columns in the view than there are today, you could imagine handling parallel-query cases with a 4-way join; but ugh.) The new function has the following behaviors that are painful or impossible to get right via pg_locks: 1. Correctly understands which lock modes block which other ones. 2. In soft-block situations (two processes both waiting for conflicting lock modes), only the one that's in front in the wait queue is reported to block the other. 3. In parallel-query cases, reports all sessions blocking any member of the given PID's lock group, and reports a session by naming its leader process's PID, which will be the pg_backend_pid() value visible to clients. The motivation for doing this right now is mostly to fix the isolation tests. Commit 38f8bdcac4982215beb9f65a19debecaf22fd470 lobotomized isolationtester's is-it-waiting query by removing its ability to recognize nonconflicting lock modes, as a crude workaround for the inability to handle soft-block situations properly. But even without the lock mode tests, the old query was excessively slow, particularly in CLOBBER_CACHE_ALWAYS builds; some of our buildfarm animals fail the new deadlock-hard test because the deadlock timeout elapses before they can probe the waiting status of all eight sessions. Replacing the pg_locks self-join with use of pg_blocking_pids() is not only much more correct, but a lot faster: I measure it at about 9X faster in a typical dev build with Asserts, and 3X faster in CLOBBER_CACHE_ALWAYS builds. That should provide enough headroom for the slower CLOBBER_CACHE_ALWAYS animals to pass the test, without having to lengthen deadlock_timeout yet more and thus slow down the test for everyone else.
2016-02-22 20:31:43 +01:00
if (waiting) /* waiting to acquire a lock */
Modify the isolation tester so that multiple sessions can wait. This allows testing of deadlock scenarios. Scenarios that would previously have been considered invalid are now simply taken as a scenario in which more than one backend will wait.
2016-02-11 14:36:30 +01:00
{
Fix isolationtester race condition for notices sent before blocking. If a test sends a notice just before blocking, it's possible on slow machines for isolationtester to detect the blocked state before it's consumed the notice. (For this to happen, the notice would have to arrive after isolationtester has waited for data for 10ms, so on fast/lightly-loaded machines it's hard to reproduce the failure.) But, if we have seen the backend as blocked, it's certainly already sent any notices it's going to send. Therefore, one more round of PQconsumeInput and PQisBusy should be enough to collect and process any such notices. This appears to explain the instability noted in commit ebd499282, so undo the hack therein to not print notices from insert-conflict-specconflict. Patch by me, diagnosis by Andres Freund. Discussion: https://postgr.es/m/14616.1564251339@sss.pgh.pa.us
2019-07-28 02:21:54 +02:00
/*
* Since it takes time to perform the lock-check query,
* some data --- notably, NOTICE messages --- might have
* arrived since we looked. We must call PQconsumeInput
* and then PQisBusy to collect and process any such
* messages. In the (unlikely) case that PQisBusy then
* returns false, we might as well go examine the
* available result.
*/
if (!PQconsumeInput(conn))
{
fprintf(stderr, "PQconsumeInput failed: %s\n",
PQerrorMessage(conn));
exit(1);
}
if (!PQisBusy(conn))
break;
/*
* conn is still busy, so conclude that the step really is
* waiting.
*/
Revert "isolationtester: don't repeat the is-it-waiting query when retrying a step." This mostly reverts commit 9c9782f066e0ce5424b8706df2cce147cb78170f. I left in the parts that rearranged removal of completed waiting steps; but the idea of not rechecking a step's blocked-ness isn't working.
2016-02-12 23:12:23 +01:00
if (!(flags & STEP_RETRY))
printf("step %s: %s <waiting ...>\n",
step->name, step->sql);
Modify the isolation tester so that multiple sessions can wait. This allows testing of deadlock scenarios. Scenarios that would previously have been considered invalid are now simply taken as a scenario in which more than one backend will wait.
2016-02-11 14:36:30 +01:00
return true;
}
/* else, not waiting */
Add support for blocked commands in isolationtester This enables us to test that blocking commands (such as foreign keys checks that conflict with some other lock) act as intended. The set of tests that this adds is pretty minimal, but can easily be extended by adding new specs. The intention is that this will serve as a basis for ensuring that further tweaks of locking implementation preserve (or improve) existing behavior. Author: Noah Misch
2011-07-12 22:35:09 +02:00
}
Modify the isolation tester so that multiple sessions can wait. This allows testing of deadlock scenarios. Scenarios that would previously have been considered invalid are now simply taken as a scenario in which more than one backend will wait.
2016-02-11 14:36:30 +01:00
/* Figure out how long we've been waiting for this step. */
gettimeofday(&current_time, NULL);
td = (int64) current_time.tv_sec - (int64) start_time.tv_sec;
td *= USECS_PER_SEC;
td += (int64) current_time.tv_usec - (int64) start_time.tv_usec;
/*
Improve isolationtester's timeout management. isolationtester.c had a hard-wired limit of 3 minutes per test step. It now emerges that this isn't quite enough for some of the slowest buildfarm animals. This isn't the first time we've had to raise this limit (cf. 1db439ad4), so let's make it configurable. This patch raises the default to 5 minutes, and introduces an environment variable PGISOLATIONTIMEOUT that can be set if more time is needed, following the precedent of PGCTLTIMEOUT. Also, modify isolationtester so that when the timeout is hit, it explicitly reports having sent a cancel. This makes the regression failure log considerably more intelligible. (In the worst case, a timed-out test might actually be reported as "passing" without this extra output, so arguably this is a bug fix in itself.) In passing, update the README file, which had apparently not gotten touched when we added "make check" support here. Back-patch to 9.6; older versions don't have comparable timeout logic. Discussion: https://postgr.es/m/22964.1575842935@sss.pgh.pa.us
2019-12-09 20:31:57 +01:00
* After max_step_wait microseconds, try to cancel the query.
Modify the isolation tester so that multiple sessions can wait. This allows testing of deadlock scenarios. Scenarios that would previously have been considered invalid are now simply taken as a scenario in which more than one backend will wait.
2016-02-11 14:36:30 +01:00
*
Re-pgindent isolationtester.c. Need to do some more hacking on this, and got annoyed that it's not indent clean.
2016-02-12 19:36:13 +01:00
* If the user tries to test an invalid permutation, we don't want
* to hang forever, especially when this is running in the
Raise some timeouts to 180s, in test code. Slow runs of buildfarm members chipmunk, hornet and mandrill saw the shorter timeouts expire. The 180s timeout in poll_query_until has been trouble-free since 2a0f89cd717ce6d49cdc47850577823682167e87 introduced it two years ago, so use 180s more widely. Back-patch to 9.6, where the first of these timeouts was introduced. Reviewed by Michael Paquier. Discussion: https://postgr.es/m/20181209001601.GC2973271@rfd.leadboat.com
2018-12-11 05:15:42 +01:00
* buildfarm. This will presumably lead to this permutation
* failing, but remaining permutations and tests should still be
* OK.
Modify the isolation tester so that multiple sessions can wait. This allows testing of deadlock scenarios. Scenarios that would previously have been considered invalid are now simply taken as a scenario in which more than one backend will wait.
2016-02-11 14:36:30 +01:00
*/
Improve isolationtester's timeout management. isolationtester.c had a hard-wired limit of 3 minutes per test step. It now emerges that this isn't quite enough for some of the slowest buildfarm animals. This isn't the first time we've had to raise this limit (cf. 1db439ad4), so let's make it configurable. This patch raises the default to 5 minutes, and introduces an environment variable PGISOLATIONTIMEOUT that can be set if more time is needed, following the precedent of PGCTLTIMEOUT. Also, modify isolationtester so that when the timeout is hit, it explicitly reports having sent a cancel. This makes the regression failure log considerably more intelligible. (In the worst case, a timed-out test might actually be reported as "passing" without this extra output, so arguably this is a bug fix in itself.) In passing, update the README file, which had apparently not gotten touched when we added "make check" support here. Back-patch to 9.6; older versions don't have comparable timeout logic. Discussion: https://postgr.es/m/22964.1575842935@sss.pgh.pa.us
2019-12-09 20:31:57 +01:00
if (td > max_step_wait && !canceled)
Add support for blocked commands in isolationtester This enables us to test that blocking commands (such as foreign keys checks that conflict with some other lock) act as intended. The set of tests that this adds is pretty minimal, but can easily be extended by adding new specs. The intention is that this will serve as a basis for ensuring that further tweaks of locking implementation preserve (or improve) existing behavior. Author: Noah Misch
2011-07-12 22:35:09 +02:00
{
Re-pgindent isolationtester.c. Need to do some more hacking on this, and got annoyed that it's not indent clean.
2016-02-12 19:36:13 +01:00
PGcancel *cancel = PQgetCancel(conn);
Modify the isolation tester so that multiple sessions can wait. This allows testing of deadlock scenarios. Scenarios that would previously have been considered invalid are now simply taken as a scenario in which more than one backend will wait.
2016-02-11 14:36:30 +01:00
if (cancel != NULL)
Code review for isolationtester changes. Fix a few oversights in 38f8bdcac4982215beb9f65a19debecaf22fd470: don't leak memory in run_permutation(), remember when we've issued a cancel rather than issuing another one every 10ms, fix some typos in comments.
2016-02-11 17:30:46 +01:00
{
Re-pgindent isolationtester.c. Need to do some more hacking on this, and got annoyed that it's not indent clean.
2016-02-12 19:36:13 +01:00
char buf[256];
Code review for isolationtester changes. Fix a few oversights in 38f8bdcac4982215beb9f65a19debecaf22fd470: don't leak memory in run_permutation(), remember when we've issued a cancel rather than issuing another one every 10ms, fix some typos in comments.
2016-02-11 17:30:46 +01:00
if (PQcancel(cancel, buf, sizeof(buf)))
Improve isolationtester's timeout management. isolationtester.c had a hard-wired limit of 3 minutes per test step. It now emerges that this isn't quite enough for some of the slowest buildfarm animals. This isn't the first time we've had to raise this limit (cf. 1db439ad4), so let's make it configurable. This patch raises the default to 5 minutes, and introduces an environment variable PGISOLATIONTIMEOUT that can be set if more time is needed, following the precedent of PGCTLTIMEOUT. Also, modify isolationtester so that when the timeout is hit, it explicitly reports having sent a cancel. This makes the regression failure log considerably more intelligible. (In the worst case, a timed-out test might actually be reported as "passing" without this extra output, so arguably this is a bug fix in itself.) In passing, update the README file, which had apparently not gotten touched when we added "make check" support here. Back-patch to 9.6; older versions don't have comparable timeout logic. Discussion: https://postgr.es/m/22964.1575842935@sss.pgh.pa.us
2019-12-09 20:31:57 +01:00
{
/*
* print to stdout not stderr, as this should appear
* in the test case's results
*/
printf("isolationtester: canceling step %s after %d seconds\n",
step->name, (int) (td / USECS_PER_SEC));
Code review for isolationtester changes. Fix a few oversights in 38f8bdcac4982215beb9f65a19debecaf22fd470: don't leak memory in run_permutation(), remember when we've issued a cancel rather than issuing another one every 10ms, fix some typos in comments.
2016-02-11 17:30:46 +01:00
canceled = true;
Improve isolationtester's timeout management. isolationtester.c had a hard-wired limit of 3 minutes per test step. It now emerges that this isn't quite enough for some of the slowest buildfarm animals. This isn't the first time we've had to raise this limit (cf. 1db439ad4), so let's make it configurable. This patch raises the default to 5 minutes, and introduces an environment variable PGISOLATIONTIMEOUT that can be set if more time is needed, following the precedent of PGCTLTIMEOUT. Also, modify isolationtester so that when the timeout is hit, it explicitly reports having sent a cancel. This makes the regression failure log considerably more intelligible. (In the worst case, a timed-out test might actually be reported as "passing" without this extra output, so arguably this is a bug fix in itself.) In passing, update the README file, which had apparently not gotten touched when we added "make check" support here. Back-patch to 9.6; older versions don't have comparable timeout logic. Discussion: https://postgr.es/m/22964.1575842935@sss.pgh.pa.us
2019-12-09 20:31:57 +01:00
}
Code review for isolationtester changes. Fix a few oversights in 38f8bdcac4982215beb9f65a19debecaf22fd470: don't leak memory in run_permutation(), remember when we've issued a cancel rather than issuing another one every 10ms, fix some typos in comments.
2016-02-11 17:30:46 +01:00
else
fprintf(stderr, "PQcancel failed: %s\n", buf);
Modify the isolation tester so that multiple sessions can wait. This allows testing of deadlock scenarios. Scenarios that would previously have been considered invalid are now simply taken as a scenario in which more than one backend will wait.
2016-02-11 14:36:30 +01:00
PQfreeCancel(cancel);
Code review for isolationtester changes. Fix a few oversights in 38f8bdcac4982215beb9f65a19debecaf22fd470: don't leak memory in run_permutation(), remember when we've issued a cancel rather than issuing another one every 10ms, fix some typos in comments.
2016-02-11 17:30:46 +01:00
}
Modify the isolation tester so that multiple sessions can wait. This allows testing of deadlock scenarios. Scenarios that would previously have been considered invalid are now simply taken as a scenario in which more than one backend will wait.
2016-02-11 14:36:30 +01:00
}
/*
Improve isolationtester's timeout management. isolationtester.c had a hard-wired limit of 3 minutes per test step. It now emerges that this isn't quite enough for some of the slowest buildfarm animals. This isn't the first time we've had to raise this limit (cf. 1db439ad4), so let's make it configurable. This patch raises the default to 5 minutes, and introduces an environment variable PGISOLATIONTIMEOUT that can be set if more time is needed, following the precedent of PGCTLTIMEOUT. Also, modify isolationtester so that when the timeout is hit, it explicitly reports having sent a cancel. This makes the regression failure log considerably more intelligible. (In the worst case, a timed-out test might actually be reported as "passing" without this extra output, so arguably this is a bug fix in itself.) In passing, update the README file, which had apparently not gotten touched when we added "make check" support here. Back-patch to 9.6; older versions don't have comparable timeout logic. Discussion: https://postgr.es/m/22964.1575842935@sss.pgh.pa.us
2019-12-09 20:31:57 +01:00
* After twice max_step_wait, just give up and die.
Modify the isolation tester so that multiple sessions can wait. This allows testing of deadlock scenarios. Scenarios that would previously have been considered invalid are now simply taken as a scenario in which more than one backend will wait.
2016-02-11 14:36:30 +01:00
*
* Since cleanup steps won't be run in this case, this may cause
* later tests to fail. That stinks, but it's better than waiting
* forever for the server to respond to the cancel.
*/
Improve isolationtester's timeout management. isolationtester.c had a hard-wired limit of 3 minutes per test step. It now emerges that this isn't quite enough for some of the slowest buildfarm animals. This isn't the first time we've had to raise this limit (cf. 1db439ad4), so let's make it configurable. This patch raises the default to 5 minutes, and introduces an environment variable PGISOLATIONTIMEOUT that can be set if more time is needed, following the precedent of PGCTLTIMEOUT. Also, modify isolationtester so that when the timeout is hit, it explicitly reports having sent a cancel. This makes the regression failure log considerably more intelligible. (In the worst case, a timed-out test might actually be reported as "passing" without this extra output, so arguably this is a bug fix in itself.) In passing, update the README file, which had apparently not gotten touched when we added "make check" support here. Back-patch to 9.6; older versions don't have comparable timeout logic. Discussion: https://postgr.es/m/22964.1575842935@sss.pgh.pa.us
2019-12-09 20:31:57 +01:00
if (td > 2 * max_step_wait)
Modify the isolation tester so that multiple sessions can wait. This allows testing of deadlock scenarios. Scenarios that would previously have been considered invalid are now simply taken as a scenario in which more than one backend will wait.
2016-02-11 14:36:30 +01:00
{
Improve isolationtester's timeout management. isolationtester.c had a hard-wired limit of 3 minutes per test step. It now emerges that this isn't quite enough for some of the slowest buildfarm animals. This isn't the first time we've had to raise this limit (cf. 1db439ad4), so let's make it configurable. This patch raises the default to 5 minutes, and introduces an environment variable PGISOLATIONTIMEOUT that can be set if more time is needed, following the precedent of PGCTLTIMEOUT. Also, modify isolationtester so that when the timeout is hit, it explicitly reports having sent a cancel. This makes the regression failure log considerably more intelligible. (In the worst case, a timed-out test might actually be reported as "passing" without this extra output, so arguably this is a bug fix in itself.) In passing, update the README file, which had apparently not gotten touched when we added "make check" support here. Back-patch to 9.6; older versions don't have comparable timeout logic. Discussion: https://postgr.es/m/22964.1575842935@sss.pgh.pa.us
2019-12-09 20:31:57 +01:00
fprintf(stderr, "step %s timed out after %d seconds\n",
step->name, (int) (td / USECS_PER_SEC));
isolationtester: Use atexit() Replace exit_nicely() calls with standard exit() and register the cleanup actions using atexit(). Reviewed-by: Alvaro Herrera <alvherre@2ndquadrant.com> Discussion: https://www.postgresql.org/message-id/flat/ec4135ba-84e9-28bf-b584-0e78d47448d5@2ndquadrant.com/
2019-01-05 15:05:49 +01:00
exit(1);
Add support for blocked commands in isolationtester This enables us to test that blocking commands (such as foreign keys checks that conflict with some other lock) act as intended. The set of tests that this adds is pretty minimal, but can easily be extended by adding new specs. The intention is that this will serve as a basis for ensuring that further tweaks of locking implementation preserve (or improve) existing behavior. Author: Noah Misch
2011-07-12 22:35:09 +02:00
}
}
else if (!PQconsumeInput(conn)) /* select(): data available */
{
Add simple tests of EvalPlanQual using the isolationtester infrastructure. Much more could be done here, but at least now we have *some* automated test coverage of that mechanism. In particular this tests the writable-CTE case reported by Phil Sorber. In passing, remove isolationtester's arbitrary restriction on the number of steps in a permutation list. I used this so that a single spec file could be used to run several related test scenarios, but there are other possible reasons to want a step series that's not exactly a permutation. Improve documentation and fix a couple other nits as well.
2012-01-28 23:55:08 +01:00
fprintf(stderr, "PQconsumeInput failed: %s\n",
PQerrorMessage(conn));
isolationtester: Use atexit() Replace exit_nicely() calls with standard exit() and register the cleanup actions using atexit(). Reviewed-by: Alvaro Herrera <alvherre@2ndquadrant.com> Discussion: https://www.postgresql.org/message-id/flat/ec4135ba-84e9-28bf-b584-0e78d47448d5@2ndquadrant.com/
2019-01-05 15:05:49 +01:00
exit(1);
Add support for blocked commands in isolationtester This enables us to test that blocking commands (such as foreign keys checks that conflict with some other lock) act as intended. The set of tests that this adds is pretty minimal, but can easily be extended by adding new specs. The intention is that this will serve as a basis for ensuring that further tweaks of locking implementation preserve (or improve) existing behavior. Author: Noah Misch
2011-07-12 22:35:09 +02:00
}
}
Use annotations to reduce instability of isolation-test results. We've long contended with isolation test results that aren't entirely stable. Some test scripts insert long delays to try to force stable results, which is not terribly desirable; but other erratic failure modes remain, causing unrepeatable buildfarm failures. I've spent a fair amount of time trying to solve this by improving the server-side support code, without much success: that way is fundamentally unable to cope with diffs that stem from chance ordering of arrival of messages from different server processes. We can improve matters on the client side, however, by annotating the test scripts themselves to show the desired reporting order of events that might occur in different orders. This patch adds three types of annotations to deal with (a) test steps that might or might not complete their waits before the isolationtester can see them waiting; (b) test steps in different sessions that can legitimately complete in either order; and (c) NOTIFY messages that might arrive before or after the completion of a step in another session. We might need more annotation types later, but this seems to be enough to deal with the instabilities we've seen in the buildfarm. It also lets us get rid of all the long delays that were previously used, cutting more than a minute off the runtime of the isolation tests. Back-patch to all supported branches, because the buildfarm instabilities affect all the branches, and because it seems desirable to keep isolationtester's capabilities the same across all branches to simplify possible future back-patching of tests. Discussion: https://postgr.es/m/327948.1623725828@sss.pgh.pa.us
2021-06-23 03:43:12 +02:00
/*
* The step is done, but we won't report it as complete so long as there
* are blockers.
*/
if (step_has_blocker(pstep))
{
if (!(flags & STEP_RETRY))
printf("step %s: %s <waiting ...>\n",
step->name, step->sql);
return true;
}
/* Otherwise, go ahead and complete it. */
Add support for blocked commands in isolationtester This enables us to test that blocking commands (such as foreign keys checks that conflict with some other lock) act as intended. The set of tests that this adds is pretty minimal, but can easily be extended by adding new specs. The intention is that this will serve as a basis for ensuring that further tweaks of locking implementation preserve (or improve) existing behavior. Author: Noah Misch
2011-07-12 22:35:09 +02:00
if (flags & STEP_RETRY)
printf("step %s: <... completed>\n", step->name);
else
printf("step %s: %s\n", step->name, step->sql);
while ((res = PQgetResult(conn)))
{
switch (PQresultStatus(res))
{
case PGRES_COMMAND_OK:
Use annotations to reduce instability of isolation-test results. We've long contended with isolation test results that aren't entirely stable. Some test scripts insert long delays to try to force stable results, which is not terribly desirable; but other erratic failure modes remain, causing unrepeatable buildfarm failures. I've spent a fair amount of time trying to solve this by improving the server-side support code, without much success: that way is fundamentally unable to cope with diffs that stem from chance ordering of arrival of messages from different server processes. We can improve matters on the client side, however, by annotating the test scripts themselves to show the desired reporting order of events that might occur in different orders. This patch adds three types of annotations to deal with (a) test steps that might or might not complete their waits before the isolationtester can see them waiting; (b) test steps in different sessions that can legitimately complete in either order; and (c) NOTIFY messages that might arrive before or after the completion of a step in another session. We might need more annotation types later, but this seems to be enough to deal with the instabilities we've seen in the buildfarm. It also lets us get rid of all the long delays that were previously used, cutting more than a minute off the runtime of the isolation tests. Back-patch to all supported branches, because the buildfarm instabilities affect all the branches, and because it seems desirable to keep isolationtester's capabilities the same across all branches to simplify possible future back-patching of tests. Discussion: https://postgr.es/m/327948.1623725828@sss.pgh.pa.us
2021-06-23 03:43:12 +02:00
case PGRES_EMPTY_QUERY:
Add support for blocked commands in isolationtester This enables us to test that blocking commands (such as foreign keys checks that conflict with some other lock) act as intended. The set of tests that this adds is pretty minimal, but can easily be extended by adding new specs. The intention is that this will serve as a basis for ensuring that further tweaks of locking implementation preserve (or improve) existing behavior. Author: Noah Misch
2011-07-12 22:35:09 +02:00
break;
case PGRES_TUPLES_OK:
printResultSet(res);
break;
case PGRES_FATAL_ERROR:
Minor robustness improvements for isolationtester. Notice and complain about PQcancel() failures. Also, don't dump core if an error PGresult doesn't contain severity and message subfields, as it might not if it was generated by libpq itself. (We have a longstanding TODO item to improve that, but in the meantime isolationtester had better cope.) I tripped across the latter item while investigating a trouble report on buildfarm member spoonbill. As for the former, there's no evidence that PQcancel failure is actually involved in spoonbill's problem, but it still seems like a bad idea to ignore an error return code.
2013-04-03 03:15:37 +02:00
/*
* Detail may contain XID values, so we want to just show
* primary. Beware however that libpq-generated error results
* may not contain subfields, only an old-style message.
*/
{
const char *sev = PQresultErrorField(res,
PG_DIAG_SEVERITY);
const char *msg = PQresultErrorField(res,
PG_DIAG_MESSAGE_PRIMARY);
if (sev && msg)
Use annotations to reduce instability of isolation-test results. We've long contended with isolation test results that aren't entirely stable. Some test scripts insert long delays to try to force stable results, which is not terribly desirable; but other erratic failure modes remain, causing unrepeatable buildfarm failures. I've spent a fair amount of time trying to solve this by improving the server-side support code, without much success: that way is fundamentally unable to cope with diffs that stem from chance ordering of arrival of messages from different server processes. We can improve matters on the client side, however, by annotating the test scripts themselves to show the desired reporting order of events that might occur in different orders. This patch adds three types of annotations to deal with (a) test steps that might or might not complete their waits before the isolationtester can see them waiting; (b) test steps in different sessions that can legitimately complete in either order; and (c) NOTIFY messages that might arrive before or after the completion of a step in another session. We might need more annotation types later, but this seems to be enough to deal with the instabilities we've seen in the buildfarm. It also lets us get rid of all the long delays that were previously used, cutting more than a minute off the runtime of the isolation tests. Back-patch to all supported branches, because the buildfarm instabilities affect all the branches, and because it seems desirable to keep isolationtester's capabilities the same across all branches to simplify possible future back-patching of tests. Discussion: https://postgr.es/m/327948.1623725828@sss.pgh.pa.us
2021-06-23 03:43:12 +02:00
printf("%s: %s\n", sev, msg);
Minor robustness improvements for isolationtester. Notice and complain about PQcancel() failures. Also, don't dump core if an error PGresult doesn't contain severity and message subfields, as it might not if it was generated by libpq itself. (We have a longstanding TODO item to improve that, but in the meantime isolationtester had better cope.) I tripped across the latter item while investigating a trouble report on buildfarm member spoonbill. As for the former, there's no evidence that PQcancel failure is actually involved in spoonbill's problem, but it still seems like a bad idea to ignore an error return code.
2013-04-03 03:15:37 +02:00
else
Use annotations to reduce instability of isolation-test results. We've long contended with isolation test results that aren't entirely stable. Some test scripts insert long delays to try to force stable results, which is not terribly desirable; but other erratic failure modes remain, causing unrepeatable buildfarm failures. I've spent a fair amount of time trying to solve this by improving the server-side support code, without much success: that way is fundamentally unable to cope with diffs that stem from chance ordering of arrival of messages from different server processes. We can improve matters on the client side, however, by annotating the test scripts themselves to show the desired reporting order of events that might occur in different orders. This patch adds three types of annotations to deal with (a) test steps that might or might not complete their waits before the isolationtester can see them waiting; (b) test steps in different sessions that can legitimately complete in either order; and (c) NOTIFY messages that might arrive before or after the completion of a step in another session. We might need more annotation types later, but this seems to be enough to deal with the instabilities we've seen in the buildfarm. It also lets us get rid of all the long delays that were previously used, cutting more than a minute off the runtime of the isolation tests. Back-patch to all supported branches, because the buildfarm instabilities affect all the branches, and because it seems desirable to keep isolationtester's capabilities the same across all branches to simplify possible future back-patching of tests. Discussion: https://postgr.es/m/327948.1623725828@sss.pgh.pa.us
2021-06-23 03:43:12 +02:00
printf("%s\n", PQresultErrorMessage(res));
Minor robustness improvements for isolationtester. Notice and complain about PQcancel() failures. Also, don't dump core if an error PGresult doesn't contain severity and message subfields, as it might not if it was generated by libpq itself. (We have a longstanding TODO item to improve that, but in the meantime isolationtester had better cope.) I tripped across the latter item while investigating a trouble report on buildfarm member spoonbill. As for the former, there's no evidence that PQcancel failure is actually involved in spoonbill's problem, but it still seems like a bad idea to ignore an error return code.
2013-04-03 03:15:37 +02:00
}
Add support for blocked commands in isolationtester This enables us to test that blocking commands (such as foreign keys checks that conflict with some other lock) act as intended. The set of tests that this adds is pretty minimal, but can easily be extended by adding new specs. The intention is that this will serve as a basis for ensuring that further tweaks of locking implementation preserve (or improve) existing behavior. Author: Noah Misch
2011-07-12 22:35:09 +02:00
break;
default:
printf("unexpected result status: %s\n",
PQresStatus(PQresultStatus(res)));
}
PQclear(res);
}
Improve test coverage for LISTEN/NOTIFY. We had no actual end-to-end test of NOTIFY message delivery. In the core async.sql regression test, testing this is problematic because psql traditionally prints the PID of the sending backend, making the output unstable. We also have an isolation test script, but it likewise failed to prove that delivery worked, because isolationtester.c had no provisions for detecting/reporting NOTIFY messages. Hence, add such provisions to isolationtester.c, and extend async-notify.spec to include direct tests of basic NOTIFY functionality. I also added tests showing that NOTIFY de-duplicates messages normally, but not across subtransaction boundaries. (That's the historical behavior since we introduced subtransactions, though perhaps we ought to change it.) Patch by me, with suggestions/review by Andres Freund. Discussion: https://postgr.es/m/31304.1564246011@sss.pgh.pa.us
2019-07-28 18:02:27 +02:00
/* Report any available NOTIFY messages, too */
PQconsumeInput(conn);
while ((notify = PQnotifies(conn)) != NULL)
{
/* Try to identify which session it came from */
const char *sendername = NULL;
char pidstring[32];
Use annotations to reduce instability of isolation-test results. We've long contended with isolation test results that aren't entirely stable. Some test scripts insert long delays to try to force stable results, which is not terribly desirable; but other erratic failure modes remain, causing unrepeatable buildfarm failures. I've spent a fair amount of time trying to solve this by improving the server-side support code, without much success: that way is fundamentally unable to cope with diffs that stem from chance ordering of arrival of messages from different server processes. We can improve matters on the client side, however, by annotating the test scripts themselves to show the desired reporting order of events that might occur in different orders. This patch adds three types of annotations to deal with (a) test steps that might or might not complete their waits before the isolationtester can see them waiting; (b) test steps in different sessions that can legitimately complete in either order; and (c) NOTIFY messages that might arrive before or after the completion of a step in another session. We might need more annotation types later, but this seems to be enough to deal with the instabilities we've seen in the buildfarm. It also lets us get rid of all the long delays that were previously used, cutting more than a minute off the runtime of the isolation tests. Back-patch to all supported branches, because the buildfarm instabilities affect all the branches, and because it seems desirable to keep isolationtester's capabilities the same across all branches to simplify possible future back-patching of tests. Discussion: https://postgr.es/m/327948.1623725828@sss.pgh.pa.us
2021-06-23 03:43:12 +02:00
int i;
Improve test coverage for LISTEN/NOTIFY. We had no actual end-to-end test of NOTIFY message delivery. In the core async.sql regression test, testing this is problematic because psql traditionally prints the PID of the sending backend, making the output unstable. We also have an isolation test script, but it likewise failed to prove that delivery worked, because isolationtester.c had no provisions for detecting/reporting NOTIFY messages. Hence, add such provisions to isolationtester.c, and extend async-notify.spec to include direct tests of basic NOTIFY functionality. I also added tests showing that NOTIFY de-duplicates messages normally, but not across subtransaction boundaries. (That's the historical behavior since we introduced subtransactions, though perhaps we ought to change it.) Patch by me, with suggestions/review by Andres Freund. Discussion: https://postgr.es/m/31304.1564246011@sss.pgh.pa.us
2019-07-28 18:02:27 +02:00
Use annotations to reduce instability of isolation-test results. We've long contended with isolation test results that aren't entirely stable. Some test scripts insert long delays to try to force stable results, which is not terribly desirable; but other erratic failure modes remain, causing unrepeatable buildfarm failures. I've spent a fair amount of time trying to solve this by improving the server-side support code, without much success: that way is fundamentally unable to cope with diffs that stem from chance ordering of arrival of messages from different server processes. We can improve matters on the client side, however, by annotating the test scripts themselves to show the desired reporting order of events that might occur in different orders. This patch adds three types of annotations to deal with (a) test steps that might or might not complete their waits before the isolationtester can see them waiting; (b) test steps in different sessions that can legitimately complete in either order; and (c) NOTIFY messages that might arrive before or after the completion of a step in another session. We might need more annotation types later, but this seems to be enough to deal with the instabilities we've seen in the buildfarm. It also lets us get rid of all the long delays that were previously used, cutting more than a minute off the runtime of the isolation tests. Back-patch to all supported branches, because the buildfarm instabilities affect all the branches, and because it seems desirable to keep isolationtester's capabilities the same across all branches to simplify possible future back-patching of tests. Discussion: https://postgr.es/m/327948.1623725828@sss.pgh.pa.us
2021-06-23 03:43:12 +02:00
for (i = 0; i < testspec->nsessions; i++)
Improve test coverage for LISTEN/NOTIFY. We had no actual end-to-end test of NOTIFY message delivery. In the core async.sql regression test, testing this is problematic because psql traditionally prints the PID of the sending backend, making the output unstable. We also have an isolation test script, but it likewise failed to prove that delivery worked, because isolationtester.c had no provisions for detecting/reporting NOTIFY messages. Hence, add such provisions to isolationtester.c, and extend async-notify.spec to include direct tests of basic NOTIFY functionality. I also added tests showing that NOTIFY de-duplicates messages normally, but not across subtransaction boundaries. (That's the historical behavior since we introduced subtransactions, though perhaps we ought to change it.) Patch by me, with suggestions/review by Andres Freund. Discussion: https://postgr.es/m/31304.1564246011@sss.pgh.pa.us
2019-07-28 18:02:27 +02:00
{
Use annotations to reduce instability of isolation-test results. We've long contended with isolation test results that aren't entirely stable. Some test scripts insert long delays to try to force stable results, which is not terribly desirable; but other erratic failure modes remain, causing unrepeatable buildfarm failures. I've spent a fair amount of time trying to solve this by improving the server-side support code, without much success: that way is fundamentally unable to cope with diffs that stem from chance ordering of arrival of messages from different server processes. We can improve matters on the client side, however, by annotating the test scripts themselves to show the desired reporting order of events that might occur in different orders. This patch adds three types of annotations to deal with (a) test steps that might or might not complete their waits before the isolationtester can see them waiting; (b) test steps in different sessions that can legitimately complete in either order; and (c) NOTIFY messages that might arrive before or after the completion of a step in another session. We might need more annotation types later, but this seems to be enough to deal with the instabilities we've seen in the buildfarm. It also lets us get rid of all the long delays that were previously used, cutting more than a minute off the runtime of the isolation tests. Back-patch to all supported branches, because the buildfarm instabilities affect all the branches, and because it seems desirable to keep isolationtester's capabilities the same across all branches to simplify possible future back-patching of tests. Discussion: https://postgr.es/m/327948.1623725828@sss.pgh.pa.us
2021-06-23 03:43:12 +02:00
if (notify->be_pid == conns[i + 1].backend_pid)
Improve test coverage for LISTEN/NOTIFY. We had no actual end-to-end test of NOTIFY message delivery. In the core async.sql regression test, testing this is problematic because psql traditionally prints the PID of the sending backend, making the output unstable. We also have an isolation test script, but it likewise failed to prove that delivery worked, because isolationtester.c had no provisions for detecting/reporting NOTIFY messages. Hence, add such provisions to isolationtester.c, and extend async-notify.spec to include direct tests of basic NOTIFY functionality. I also added tests showing that NOTIFY de-duplicates messages normally, but not across subtransaction boundaries. (That's the historical behavior since we introduced subtransactions, though perhaps we ought to change it.) Patch by me, with suggestions/review by Andres Freund. Discussion: https://postgr.es/m/31304.1564246011@sss.pgh.pa.us
2019-07-28 18:02:27 +02:00
{
Use annotations to reduce instability of isolation-test results. We've long contended with isolation test results that aren't entirely stable. Some test scripts insert long delays to try to force stable results, which is not terribly desirable; but other erratic failure modes remain, causing unrepeatable buildfarm failures. I've spent a fair amount of time trying to solve this by improving the server-side support code, without much success: that way is fundamentally unable to cope with diffs that stem from chance ordering of arrival of messages from different server processes. We can improve matters on the client side, however, by annotating the test scripts themselves to show the desired reporting order of events that might occur in different orders. This patch adds three types of annotations to deal with (a) test steps that might or might not complete their waits before the isolationtester can see them waiting; (b) test steps in different sessions that can legitimately complete in either order; and (c) NOTIFY messages that might arrive before or after the completion of a step in another session. We might need more annotation types later, but this seems to be enough to deal with the instabilities we've seen in the buildfarm. It also lets us get rid of all the long delays that were previously used, cutting more than a minute off the runtime of the isolation tests. Back-patch to all supported branches, because the buildfarm instabilities affect all the branches, and because it seems desirable to keep isolationtester's capabilities the same across all branches to simplify possible future back-patching of tests. Discussion: https://postgr.es/m/327948.1623725828@sss.pgh.pa.us
2021-06-23 03:43:12 +02:00
sendername = conns[i + 1].sessionname;
Improve test coverage for LISTEN/NOTIFY. We had no actual end-to-end test of NOTIFY message delivery. In the core async.sql regression test, testing this is problematic because psql traditionally prints the PID of the sending backend, making the output unstable. We also have an isolation test script, but it likewise failed to prove that delivery worked, because isolationtester.c had no provisions for detecting/reporting NOTIFY messages. Hence, add such provisions to isolationtester.c, and extend async-notify.spec to include direct tests of basic NOTIFY functionality. I also added tests showing that NOTIFY de-duplicates messages normally, but not across subtransaction boundaries. (That's the historical behavior since we introduced subtransactions, though perhaps we ought to change it.) Patch by me, with suggestions/review by Andres Freund. Discussion: https://postgr.es/m/31304.1564246011@sss.pgh.pa.us
2019-07-28 18:02:27 +02:00
break;
}
}
if (sendername == NULL)
{
/* Doesn't seem to be any test session, so show the hard way */
snprintf(pidstring, sizeof(pidstring), "PID %d", notify->be_pid);
sendername = pidstring;
}
printf("%s: NOTIFY \"%s\" with payload \"%s\" from %s\n",
testspec->sessions[step->session]->name,
notify->relname, notify->extra, sendername);
PQfreemem(notify);
PQconsumeInput(conn);
}
Use annotations to reduce instability of isolation-test results. We've long contended with isolation test results that aren't entirely stable. Some test scripts insert long delays to try to force stable results, which is not terribly desirable; but other erratic failure modes remain, causing unrepeatable buildfarm failures. I've spent a fair amount of time trying to solve this by improving the server-side support code, without much success: that way is fundamentally unable to cope with diffs that stem from chance ordering of arrival of messages from different server processes. We can improve matters on the client side, however, by annotating the test scripts themselves to show the desired reporting order of events that might occur in different orders. This patch adds three types of annotations to deal with (a) test steps that might or might not complete their waits before the isolationtester can see them waiting; (b) test steps in different sessions that can legitimately complete in either order; and (c) NOTIFY messages that might arrive before or after the completion of a step in another session. We might need more annotation types later, but this seems to be enough to deal with the instabilities we've seen in the buildfarm. It also lets us get rid of all the long delays that were previously used, cutting more than a minute off the runtime of the isolation tests. Back-patch to all supported branches, because the buildfarm instabilities affect all the branches, and because it seems desirable to keep isolationtester's capabilities the same across all branches to simplify possible future back-patching of tests. Discussion: https://postgr.es/m/327948.1623725828@sss.pgh.pa.us
2021-06-23 03:43:12 +02:00
/* Connection is now idle. */
iconn->active_step = NULL;
return false;
}
/* Detect whether a step has any unsatisfied blocker conditions */
static bool
step_has_blocker(PermutationStep *pstep)
{
int i;
for (i = 0; i < pstep->nblockers; i++)
{
PermutationStepBlocker *blocker = pstep->blockers[i];
IsoConnInfo *iconn;
switch (blocker->blocktype)
{
case PSB_ONCE:
/* Ignore; try_complete_step handles this specially */
break;
case PSB_OTHER_STEP:
/* Block if referenced step is active */
iconn = &conns[1 + blocker->step->session];
if (iconn->active_step &&
iconn->active_step->step == blocker->step)
return true;
break;
case PSB_NUM_NOTICES:
/* Block if not enough notices received yet */
iconn = &conns[1 + blocker->step->session];
if (iconn->total_notices < blocker->target_notices)
return true;
break;
}
}
Add support for blocked commands in isolationtester This enables us to test that blocking commands (such as foreign keys checks that conflict with some other lock) act as intended. The set of tests that this adds is pretty minimal, but can easily be extended by adding new specs. The intention is that this will serve as a basis for ensuring that further tweaks of locking implementation preserve (or improve) existing behavior. Author: Noah Misch
2011-07-12 22:35:09 +02:00
return false;
}
Implement genuine serializable isolation level. Until now, our Serializable mode has in fact been what's called Snapshot Isolation, which allows some anomalies that could not occur in any serialized ordering of the transactions. This patch fixes that using a method called Serializable Snapshot Isolation, based on research papers by Michael J. Cahill (see README-SSI for full references). In Serializable Snapshot Isolation, transactions run like they do in Snapshot Isolation, but a predicate lock manager observes the reads and writes performed and aborts transactions if it detects that an anomaly might occur. This method produces some false positives, ie. it sometimes aborts transactions even though there is no anomaly. To track reads we implement predicate locking, see storage/lmgr/predicate.c. Whenever a tuple is read, a predicate lock is acquired on the tuple. Shared memory is finite, so when a transaction takes many tuple-level locks on a page, the locks are promoted to a single page-level lock, and further to a single relation level lock if necessary. To lock key values with no matching tuple, a sequential scan always takes a relation-level lock, and an index scan acquires a page-level lock that covers the search key, whether or not there are any matching keys at the moment. A predicate lock doesn't conflict with any regular locks or with another predicate locks in the normal sense. They're only used by the predicate lock manager to detect the danger of anomalies. Only serializable transactions participate in predicate locking, so there should be no extra overhead for for other transactions. Predicate locks can't be released at commit, but must be remembered until all the transactions that overlapped with it have completed. That means that we need to remember an unbounded amount of predicate locks, so we apply a lossy but conservative method of tracking locks for committed transactions. If we run short of shared memory, we overflow to a new "pg_serial" SLRU pool. We don't currently allow Serializable transactions in Hot Standby mode. That would be hard, because even read-only transactions can cause anomalies that wouldn't otherwise occur. Serializable isolation mode now means the new fully serializable level. Repeatable Read gives you the old Snapshot Isolation level that we have always had. Kevin Grittner and Dan Ports, reviewed by Jeff Davis, Heikki Linnakangas and Anssi Kääriäinen
2011-02-07 22:46:51 +01:00
static void
printResultSet(PGresult *res)
{
Improve display of query results in isolation tests. Previously, isolationtester displayed SQL query results using some ad-hoc code that clearly hadn't had much effort expended on it. Field values longer than 14 characters weren't separated from the next field, and usually caused misalignment of the columns too. Also there was no visual separation of a query's result from subsequent isolationtester output. This made test result files confusing and hard to read. To improve matters, let's use libpq's PQprint() function. Although that's long since unused by psql, it's still plenty good enough for the purpose here. Like 741d7f104, back-patch to all supported branches, so that this isn't a stumbling block for back-patching isolation test changes. Discussion: https://postgr.es/m/582362.1623798221@sss.pgh.pa.us
2021-06-23 17:12:31 +02:00
PQprintOpt popt;
pgindent run before PG 9.1 beta 1.
2011-04-10 17:42:00 +02:00
Improve display of query results in isolation tests. Previously, isolationtester displayed SQL query results using some ad-hoc code that clearly hadn't had much effort expended on it. Field values longer than 14 characters weren't separated from the next field, and usually caused misalignment of the columns too. Also there was no visual separation of a query's result from subsequent isolationtester output. This made test result files confusing and hard to read. To improve matters, let's use libpq's PQprint() function. Although that's long since unused by psql, it's still plenty good enough for the purpose here. Like 741d7f104, back-patch to all supported branches, so that this isn't a stumbling block for back-patching isolation test changes. Discussion: https://postgr.es/m/582362.1623798221@sss.pgh.pa.us
2021-06-23 17:12:31 +02:00
memset(&popt, 0, sizeof(popt));
popt.header = true;
popt.align = true;
popt.fieldSep = "|";
PQprint(stdout, res, &popt);
Implement genuine serializable isolation level. Until now, our Serializable mode has in fact been what's called Snapshot Isolation, which allows some anomalies that could not occur in any serialized ordering of the transactions. This patch fixes that using a method called Serializable Snapshot Isolation, based on research papers by Michael J. Cahill (see README-SSI for full references). In Serializable Snapshot Isolation, transactions run like they do in Snapshot Isolation, but a predicate lock manager observes the reads and writes performed and aborts transactions if it detects that an anomaly might occur. This method produces some false positives, ie. it sometimes aborts transactions even though there is no anomaly. To track reads we implement predicate locking, see storage/lmgr/predicate.c. Whenever a tuple is read, a predicate lock is acquired on the tuple. Shared memory is finite, so when a transaction takes many tuple-level locks on a page, the locks are promoted to a single page-level lock, and further to a single relation level lock if necessary. To lock key values with no matching tuple, a sequential scan always takes a relation-level lock, and an index scan acquires a page-level lock that covers the search key, whether or not there are any matching keys at the moment. A predicate lock doesn't conflict with any regular locks or with another predicate locks in the normal sense. They're only used by the predicate lock manager to detect the danger of anomalies. Only serializable transactions participate in predicate locking, so there should be no extra overhead for for other transactions. Predicate locks can't be released at commit, but must be remembered until all the transactions that overlapped with it have completed. That means that we need to remember an unbounded amount of predicate locks, so we apply a lossy but conservative method of tracking locks for committed transactions. If we run short of shared memory, we overflow to a new "pg_serial" SLRU pool. We don't currently allow Serializable transactions in Hot Standby mode. That would be hard, because even read-only transactions can cause anomalies that wouldn't otherwise occur. Serializable isolation mode now means the new fully serializable level. Repeatable Read gives you the old Snapshot Isolation level that we have always had. Kevin Grittner and Dan Ports, reviewed by Jeff Davis, Heikki Linnakangas and Anssi Kääriäinen
2011-02-07 22:46:51 +01:00
}
Indicate session name in isolationtester notices When a session under isolationtester produces printable notices (NOTICE, WARNING) we were just printing them unadorned, which can be confusing when debugging. Prefix them with the session name, which makes things clearer. Author: Álvaro Herrera Reviewed-by: Hari Babu Kommi Discussion: https://postgr.es/m/20181024213451.75nh3f3dctmcdbfq@alvherre.pgsql
2018-11-09 17:08:00 +01:00
Use annotations to reduce instability of isolation-test results. We've long contended with isolation test results that aren't entirely stable. Some test scripts insert long delays to try to force stable results, which is not terribly desirable; but other erratic failure modes remain, causing unrepeatable buildfarm failures. I've spent a fair amount of time trying to solve this by improving the server-side support code, without much success: that way is fundamentally unable to cope with diffs that stem from chance ordering of arrival of messages from different server processes. We can improve matters on the client side, however, by annotating the test scripts themselves to show the desired reporting order of events that might occur in different orders. This patch adds three types of annotations to deal with (a) test steps that might or might not complete their waits before the isolationtester can see them waiting; (b) test steps in different sessions that can legitimately complete in either order; and (c) NOTIFY messages that might arrive before or after the completion of a step in another session. We might need more annotation types later, but this seems to be enough to deal with the instabilities we've seen in the buildfarm. It also lets us get rid of all the long delays that were previously used, cutting more than a minute off the runtime of the isolation tests. Back-patch to all supported branches, because the buildfarm instabilities affect all the branches, and because it seems desirable to keep isolationtester's capabilities the same across all branches to simplify possible future back-patching of tests. Discussion: https://postgr.es/m/327948.1623725828@sss.pgh.pa.us
2021-06-23 03:43:12 +02:00
/* notice processor for regular user sessions */
Indicate session name in isolationtester notices When a session under isolationtester produces printable notices (NOTICE, WARNING) we were just printing them unadorned, which can be confusing when debugging. Prefix them with the session name, which makes things clearer. Author: Álvaro Herrera Reviewed-by: Hari Babu Kommi Discussion: https://postgr.es/m/20181024213451.75nh3f3dctmcdbfq@alvherre.pgsql
2018-11-09 17:08:00 +01:00
static void
isotesterNoticeProcessor(void *arg, const char *message)
{
Use annotations to reduce instability of isolation-test results. We've long contended with isolation test results that aren't entirely stable. Some test scripts insert long delays to try to force stable results, which is not terribly desirable; but other erratic failure modes remain, causing unrepeatable buildfarm failures. I've spent a fair amount of time trying to solve this by improving the server-side support code, without much success: that way is fundamentally unable to cope with diffs that stem from chance ordering of arrival of messages from different server processes. We can improve matters on the client side, however, by annotating the test scripts themselves to show the desired reporting order of events that might occur in different orders. This patch adds three types of annotations to deal with (a) test steps that might or might not complete their waits before the isolationtester can see them waiting; (b) test steps in different sessions that can legitimately complete in either order; and (c) NOTIFY messages that might arrive before or after the completion of a step in another session. We might need more annotation types later, but this seems to be enough to deal with the instabilities we've seen in the buildfarm. It also lets us get rid of all the long delays that were previously used, cutting more than a minute off the runtime of the isolation tests. Back-patch to all supported branches, because the buildfarm instabilities affect all the branches, and because it seems desirable to keep isolationtester's capabilities the same across all branches to simplify possible future back-patching of tests. Discussion: https://postgr.es/m/327948.1623725828@sss.pgh.pa.us
2021-06-23 03:43:12 +02:00
IsoConnInfo *myconn = (IsoConnInfo *) arg;
/* Prefix the backend's message with the session name. */
printf("%s: %s", myconn->sessionname, message);
/* Record notices, since we may need this to decide to unblock a step. */
myconn->total_notices++;
any_new_notice = true;
Indicate session name in isolationtester notices When a session under isolationtester produces printable notices (NOTICE, WARNING) we were just printing them unadorned, which can be confusing when debugging. Prefix them with the session name, which makes things clearer. Author: Álvaro Herrera Reviewed-by: Hari Babu Kommi Discussion: https://postgr.es/m/20181024213451.75nh3f3dctmcdbfq@alvherre.pgsql
2018-11-09 17:08:00 +01:00
}
/* notice processor, hides the message */
static void
blackholeNoticeProcessor(void *arg, const char *message)
{
/* do nothing */
}