postgresql/src/bin/pg_dump/pg_backup_db.c

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

663 lines
17 KiB
C
Raw Normal View History

/*-------------------------------------------------------------------------
*
* pg_backup_db.c
*
* Implements the basic DB functions used by the archiver.
*
* IDENTIFICATION
2010-09-20 22:08:53 +02:00
* src/bin/pg_dump/pg_backup_db.c
*
UUNET is looking into offering PostgreSQL as a part of a managed web hosting product, on both shared and dedicated machines. We currently offer Oracle and MySQL, and it would be a nice middle-ground. However, as shipped, PostgreSQL lacks the following features we need that MySQL has: 1. The ability to listen only on a particular IP address. Each hosting customer has their own IP address, on which all of their servers (http, ftp, real media, etc.) run. 2. The ability to place the Unix-domain socket in a mode 700 directory. This allows us to automatically create an empty database, with an empty DBA password, for new or upgrading customers without having to interactively set a DBA password and communicate it to (or from) the customer. This in turn cuts down our install and upgrade times. 3. The ability to connect to the Unix-domain socket from within a change-rooted environment. We run CGI programs chrooted to the user's home directory, which is another reason why we need to be able to specify where the Unix-domain socket is, instead of /tmp. 4. The ability to, if run as root, open a pid file in /var/run as root, and then setuid to the desired user. (mysqld -u can almost do this; I had to patch it, too). The patch below fixes problem 1-3. I plan to address #4, also, but haven't done so yet. These diffs are big enough that they should give the PG development team something to think about in the meantime :-) Also, I'm about to leave for 2 weeks' vacation, so I thought I'd get out what I have, which works (for the problems it tackles), now. With these changes, we can set up and run PostgreSQL with scripts the same way we can with apache or proftpd or mysql. In summary, this patch makes the following enhancements: 1. Adds an environment variable PGUNIXSOCKET, analogous to MYSQL_UNIX_PORT, and command line options -k --unix-socket to the relevant programs. 2. Adds a -h option to postmaster to set the hostname or IP address to listen on instead of the default INADDR_ANY. 3. Extends some library interfaces to support the above. 4. Fixes a few memory leaks in PQconnectdb(). The default behavior is unchanged from stock 7.0.2; if you don't use any of these new features, they don't change the operation. David J. MacKenzie
2000-11-13 16:18:15 +01:00
*-------------------------------------------------------------------------
*/
#include "postgres_fe.h"
#include <unistd.h>
#include <ctype.h>
#ifdef HAVE_TERMIOS_H
#include <termios.h>
#endif
#include "common/connect.h"
Remove arbitrary restrictions on password length. This patch started out with the goal of harmonizing various arbitrary limits on password length, but after awhile a better idea emerged: let's just get rid of those fixed limits. recv_password_packet() has an arbitrary limit on the packet size, which we don't really need, so just drop it. (Note that this doesn't really affect anything for MD5 or SCRAM password verification, since those will hash the user's password to something shorter anyway. It does matter for auth methods that require a cleartext password.) Likewise remove the arbitrary error condition in pg_saslprep(). The remaining limits are mostly in client-side code that prompts for passwords. To improve those, refactor simple_prompt() so that it allocates its own result buffer that can be made as big as necessary. Actually, it proves best to make a separate routine pg_get_line() that has essentially the semantics of fgets(), except that it allocates a suitable result buffer and hence will never return a truncated line. (pg_get_line has a lot of potential applications to replace randomly-sized fgets buffers elsewhere, but I'll leave that for another patch.) I built pg_get_line() atop stringinfo.c, which requires moving that code to src/common/; but that seems fine since it was a poor fit for src/port/ anyway. This patch is mostly mine, but it owes a good deal to Nathan Bossart who pressed for a solution to the password length problem and created a predecessor patch. Also thanks to Peter Eisentraut and Stephen Frost for ideas and discussion. Discussion: https://postgr.es/m/09512C4F-8CB9-4021-B455-EF4C4F0D55A0@amazon.com
2020-09-04 02:09:18 +02:00
#include "common/string.h"
#include "dumputils.h"
#include "fe_utils/string_utils.h"
#include "parallel.h"
#include "pg_backup_archiver.h"
#include "pg_backup_db.h"
#include "pg_backup_utils.h"
static void _check_database_version(ArchiveHandle *AH);
static void notice_processor(void *arg, const char *message);
They are two different problems; the TOC entry is important for any multiline command or to rerun the command easily later. Whereas displaying the failed SQL command is a matter of fixing the error messages. The latter is complicated by failed COPY commands which, with die-on-errors off, results in the data being processed as a command, so dumping the command will dump all of the data. In the case of long commands, should the whole command be dumped? eg. (eg. several pages of function definition). In the case of the COPY command, I'm not sure what to do. Obviously, it would be best to avoid sending the data, but the data and command are combined (from memory). Also, the 'data' may be in the form of INSERT statements. Attached patch produces the first 125 chars of the command: pg_restore: [archiver (db)] Error while PROCESSING TOC: pg_restore: [archiver (db)] Error from TOC Entry 26; 1255 16449270 FUNCTION plpgsql_call_handler() pjw pg_restore: [archiver (db)] could not execute query: ERROR: function "plpgsql_call_handler" already exists with same argument types Command was: CREATE FUNCTION plpgsql_call_handler() RETURNS language_handler AS '/var/lib/pgsql-8.0b1/lib/plpgsql', 'plpgsql_call_han... pg_restore: [archiver (db)] Error from TOC Entry 27; 1255 16449271 FUNCTION plpgsql_validator(oid) pjw pg_restore: [archiver (db)] could not execute query: ERROR: function "plpgsql_validator" already exists with same argument types Command was: CREATE FUNCTION plpgsql_validator(oid) RETURNS void AS '/var/lib/pgsql-8.0b1/lib/plpgsql', 'plpgsql_validator' LANGU... Philip Warner
2004-08-20 22:00:34 +02:00
static void
_check_database_version(ArchiveHandle *AH)
{
const char *remoteversion_str;
int remoteversion;
PGresult *res;
remoteversion_str = PQparameterStatus(AH->connection, "server_version");
remoteversion = PQserverVersion(AH->connection);
if (remoteversion == 0 || !remoteversion_str)
pg_fatal("could not get server_version from libpq");
AH->public.remoteVersionStr = pg_strdup(remoteversion_str);
AH->public.remoteVersion = remoteversion;
if (!AH->archiveRemoteVersion)
AH->archiveRemoteVersion = AH->public.remoteVersionStr;
if (remoteversion != PG_VERSION_NUM
&& (remoteversion < AH->public.minRemoteVersion ||
remoteversion > AH->public.maxRemoteVersion))
{
pg_log_error("aborting because of server version mismatch");
pg_log_error_detail("server version: %s; %s version: %s",
remoteversion_str, progname, PG_VERSION);
exit(1);
}
/*
* Check if server is in recovery mode, which means we are on a hot
* standby.
*/
res = ExecuteSqlQueryForSingleRow((Archive *) AH,
"SELECT pg_catalog.pg_is_in_recovery()");
AH->public.isStandby = (strcmp(PQgetvalue(res, 0, 0), "t") == 0);
PQclear(res);
}
/*
* Reconnect to the server. If dbname is not NULL, use that database,
* else the one associated with the archive handle.
*/
In pg_dump, force reconnection after issuing ALTER DATABASE SET command(s). The folly of not doing this was exposed by the buildfarm: in some cases, the GUC settings applied through ALTER DATABASE SET may be essential to interpreting the reloaded data correctly. Another argument why we can't really get away with the scheme proposed in commit b3f840120 is that it cannot work for parallel restore: even if the parent process manages to hang onto the previous GUC state, worker processes would see the state post-ALTER-DATABASE. (Perhaps we could have dodged that bullet by delaying DATABASE PROPERTIES restoration to the end of the run, but that does nothing for the data semantics problem.) This leaves us with no solution for the default_transaction_read_only issue that commit 4bd371f6f intended to work around, other than "you gotta remove such settings before dumping/upgrading". However, in view of the fact that parallel restore broke that hack years ago and no one has noticed, it's fair to question how many people care. I'm unexcited about adding a large dollop of new complexity to handle that corner case. This would be a one-liner fix, except it turns out that ReconnectToServer tries to optimize away "redundant" reconnections. While that may have been valuable when coded, a quick survey of current callers shows that there are no cases where that's actually useful, so just remove that check. While at it, remove the function's useless return value. Discussion: https://postgr.es/m/12453.1516655001@sss.pgh.pa.us
2018-01-23 16:55:08 +01:00
void
ReconnectToServer(ArchiveHandle *AH, const char *dbname)
{
PGconn *oldConn = AH->connection;
RestoreOptions *ropt = AH->public.ropt;
Remove arbitrary restrictions on password length. This patch started out with the goal of harmonizing various arbitrary limits on password length, but after awhile a better idea emerged: let's just get rid of those fixed limits. recv_password_packet() has an arbitrary limit on the packet size, which we don't really need, so just drop it. (Note that this doesn't really affect anything for MD5 or SCRAM password verification, since those will hash the user's password to something shorter anyway. It does matter for auth methods that require a cleartext password.) Likewise remove the arbitrary error condition in pg_saslprep(). The remaining limits are mostly in client-side code that prompts for passwords. To improve those, refactor simple_prompt() so that it allocates its own result buffer that can be made as big as necessary. Actually, it proves best to make a separate routine pg_get_line() that has essentially the semantics of fgets(), except that it allocates a suitable result buffer and hence will never return a truncated line. (pg_get_line has a lot of potential applications to replace randomly-sized fgets buffers elsewhere, but I'll leave that for another patch.) I built pg_get_line() atop stringinfo.c, which requires moving that code to src/common/; but that seems fine since it was a poor fit for src/port/ anyway. This patch is mostly mine, but it owes a good deal to Nathan Bossart who pressed for a solution to the password length problem and created a predecessor patch. Also thanks to Peter Eisentraut and Stephen Frost for ideas and discussion. Discussion: https://postgr.es/m/09512C4F-8CB9-4021-B455-EF4C4F0D55A0@amazon.com
2020-09-04 02:09:18 +02:00
/*
* Save the dbname, if given, in override_dbname so that it will also
* affect any later reconnection attempt.
*/
if (dbname)
ropt->cparams.override_dbname = pg_strdup(dbname);
/*
* Note: we want to establish the new connection, and in particular update
* ArchiveHandle's connCancel, before closing old connection. Otherwise
* an ill-timed SIGINT could try to access a dead connection.
*/
AH->connection = NULL; /* dodge error check in ConnectDatabase */
ConnectDatabase((Archive *) AH, &ropt->cparams, true);
PQfinish(oldConn);
}
/*
* Make, or remake, a database connection with the given parameters.
*
* The resulting connection handle is stored in AHX->connection.
*
* An interactive password prompt is automatically issued if required.
* We store the results of that in AHX->savedPassword.
* Note: it's not really all that sensible to use a single-entry password
* cache if the username keeps changing. In current usage, however, the
* username never does change, so one savedPassword is sufficient.
*/
void
ConnectDatabase(Archive *AHX,
const ConnParams *cparams,
bool isReconnect)
{
ArchiveHandle *AH = (ArchiveHandle *) AHX;
trivalue prompt_password;
char *password;
bool new_pass;
if (AH->connection)
pg_fatal("already connected to a database");
/* Never prompt for a password during a reconnection */
prompt_password = isReconnect ? TRI_NO : cparams->promptPassword;
Simplify correct use of simple_prompt(). The previous API for this function had it returning a malloc'd string. That meant that callers had to check for NULL return, which few of them were doing, and it also meant that callers had to remember to free() the string later, which required extra logic in most cases. Instead, make simple_prompt() write into a buffer supplied by the caller. Anywhere that the maximum required input length is reasonably small, which is almost all of the callers, we can just use a local or static array as the buffer instead of dealing with malloc/free. A fair number of callers used "pointer == NULL" as a proxy for "haven't requested the password yet". Maintaining the same behavior requires adding a separate boolean flag for that, which adds back some of the complexity we save by removing free()s. Nonetheless, this nets out at a small reduction in overall code size, and considerably less code than we would have had if we'd added the missing NULL-return checks everywhere they were needed. In passing, clean up the API comment for simple_prompt() and get rid of a very-unnecessary malloc/free in its Windows code path. This is nominally a bug fix, but it does not seem worth back-patching, because the actual risk of an OOM failure in any of these places seems pretty tiny, and all of them are client-side not server-side anyway. This patch is by me, but it owes a great deal to Michael Paquier who identified the problem and drafted a patch for fixing it the other way. Discussion: <CAB7nPqRu07Ot6iht9i9KRfYLpDaF2ZuUv5y_+72uP23ZAGysRg@mail.gmail.com>
2016-08-30 23:02:02 +02:00
password = AH->savedPassword;
if (prompt_password == TRI_YES && password == NULL)
Remove arbitrary restrictions on password length. This patch started out with the goal of harmonizing various arbitrary limits on password length, but after awhile a better idea emerged: let's just get rid of those fixed limits. recv_password_packet() has an arbitrary limit on the packet size, which we don't really need, so just drop it. (Note that this doesn't really affect anything for MD5 or SCRAM password verification, since those will hash the user's password to something shorter anyway. It does matter for auth methods that require a cleartext password.) Likewise remove the arbitrary error condition in pg_saslprep(). The remaining limits are mostly in client-side code that prompts for passwords. To improve those, refactor simple_prompt() so that it allocates its own result buffer that can be made as big as necessary. Actually, it proves best to make a separate routine pg_get_line() that has essentially the semantics of fgets(), except that it allocates a suitable result buffer and hence will never return a truncated line. (pg_get_line has a lot of potential applications to replace randomly-sized fgets buffers elsewhere, but I'll leave that for another patch.) I built pg_get_line() atop stringinfo.c, which requires moving that code to src/common/; but that seems fine since it was a poor fit for src/port/ anyway. This patch is mostly mine, but it owes a good deal to Nathan Bossart who pressed for a solution to the password length problem and created a predecessor patch. Also thanks to Peter Eisentraut and Stephen Frost for ideas and discussion. Discussion: https://postgr.es/m/09512C4F-8CB9-4021-B455-EF4C4F0D55A0@amazon.com
2020-09-04 02:09:18 +02:00
password = simple_prompt("Password: ", false);
/*
* Start the connection. Loop until we have a password if requested by
* backend.
*/
do
{
const char *keywords[8];
const char *values[8];
int i = 0;
/*
* If dbname is a connstring, its entries can override the other
* values obtained from cparams; but in turn, override_dbname can
* override the dbname component of it.
*/
keywords[i] = "host";
values[i++] = cparams->pghost;
keywords[i] = "port";
values[i++] = cparams->pgport;
keywords[i] = "user";
values[i++] = cparams->username;
keywords[i] = "password";
values[i++] = password;
keywords[i] = "dbname";
values[i++] = cparams->dbname;
if (cparams->override_dbname)
{
keywords[i] = "dbname";
values[i++] = cparams->override_dbname;
}
keywords[i] = "fallback_application_name";
values[i++] = progname;
keywords[i] = NULL;
values[i++] = NULL;
Assert(i <= lengthof(keywords));
new_pass = false;
AH->connection = PQconnectdbParams(keywords, values, true);
if (!AH->connection)
pg_fatal("could not connect to database");
if (PQstatus(AH->connection) == CONNECTION_BAD &&
PQconnectionNeedsPassword(AH->connection) &&
password == NULL &&
prompt_password != TRI_NO)
{
PQfinish(AH->connection);
Remove arbitrary restrictions on password length. This patch started out with the goal of harmonizing various arbitrary limits on password length, but after awhile a better idea emerged: let's just get rid of those fixed limits. recv_password_packet() has an arbitrary limit on the packet size, which we don't really need, so just drop it. (Note that this doesn't really affect anything for MD5 or SCRAM password verification, since those will hash the user's password to something shorter anyway. It does matter for auth methods that require a cleartext password.) Likewise remove the arbitrary error condition in pg_saslprep(). The remaining limits are mostly in client-side code that prompts for passwords. To improve those, refactor simple_prompt() so that it allocates its own result buffer that can be made as big as necessary. Actually, it proves best to make a separate routine pg_get_line() that has essentially the semantics of fgets(), except that it allocates a suitable result buffer and hence will never return a truncated line. (pg_get_line has a lot of potential applications to replace randomly-sized fgets buffers elsewhere, but I'll leave that for another patch.) I built pg_get_line() atop stringinfo.c, which requires moving that code to src/common/; but that seems fine since it was a poor fit for src/port/ anyway. This patch is mostly mine, but it owes a good deal to Nathan Bossart who pressed for a solution to the password length problem and created a predecessor patch. Also thanks to Peter Eisentraut and Stephen Frost for ideas and discussion. Discussion: https://postgr.es/m/09512C4F-8CB9-4021-B455-EF4C4F0D55A0@amazon.com
2020-09-04 02:09:18 +02:00
password = simple_prompt("Password: ", false);
new_pass = true;
}
} while (new_pass);
/* check to see that the backend connection was successfully made */
if (PQstatus(AH->connection) == CONNECTION_BAD)
{
if (isReconnect)
pg_fatal("reconnection failed: %s",
PQerrorMessage(AH->connection));
else
pg_fatal("%s",
PQerrorMessage(AH->connection));
}
/* Start strict; later phases may override this. */
PQclear(ExecuteSqlQueryForSingleRow((Archive *) AH,
ALWAYS_SECURE_SEARCH_PATH_SQL));
Remove arbitrary restrictions on password length. This patch started out with the goal of harmonizing various arbitrary limits on password length, but after awhile a better idea emerged: let's just get rid of those fixed limits. recv_password_packet() has an arbitrary limit on the packet size, which we don't really need, so just drop it. (Note that this doesn't really affect anything for MD5 or SCRAM password verification, since those will hash the user's password to something shorter anyway. It does matter for auth methods that require a cleartext password.) Likewise remove the arbitrary error condition in pg_saslprep(). The remaining limits are mostly in client-side code that prompts for passwords. To improve those, refactor simple_prompt() so that it allocates its own result buffer that can be made as big as necessary. Actually, it proves best to make a separate routine pg_get_line() that has essentially the semantics of fgets(), except that it allocates a suitable result buffer and hence will never return a truncated line. (pg_get_line has a lot of potential applications to replace randomly-sized fgets buffers elsewhere, but I'll leave that for another patch.) I built pg_get_line() atop stringinfo.c, which requires moving that code to src/common/; but that seems fine since it was a poor fit for src/port/ anyway. This patch is mostly mine, but it owes a good deal to Nathan Bossart who pressed for a solution to the password length problem and created a predecessor patch. Also thanks to Peter Eisentraut and Stephen Frost for ideas and discussion. Discussion: https://postgr.es/m/09512C4F-8CB9-4021-B455-EF4C4F0D55A0@amazon.com
2020-09-04 02:09:18 +02:00
if (password && password != AH->savedPassword)
free(password);
/*
* We want to remember connection's actual password, whether or not we got
* it by prompting. So we don't just store the password variable.
*/
if (PQconnectionUsedPassword(AH->connection))
{
free(AH->savedPassword);
AH->savedPassword = pg_strdup(PQpass(AH->connection));
}
/* check for version mismatch */
_check_database_version(AH);
PQsetNoticeProcessor(AH->connection, notice_processor, NULL);
Redesign handling of SIGTERM/control-C in parallel pg_dump/pg_restore. Formerly, Unix builds of pg_dump/pg_restore would trap SIGINT and similar signals and set a flag that was tested in various data-transfer loops. This was prone to errors of omission (cf commit 3c8aa6654); and even if the client-side response was prompt, we did nothing that would cause long-running SQL commands (e.g. CREATE INDEX) to terminate early. Also, the master process would effectively do nothing at all upon receipt of SIGINT; the only reason it seemed to work was that in typical scenarios the signal would also be delivered to the child processes. We should support termination when a signal is delivered only to the master process, though. Windows builds had no console interrupt handler, so they would just fall over immediately at control-C, again leaving long-running SQL commands to finish unmolested. To fix, remove the flag-checking approach altogether. Instead, allow the Unix signal handler to send a cancel request directly and then exit(1). In the master process, also have it forward the signal to the children. On Windows, add a console interrupt handler that behaves approximately the same. The main difference is that a single execution of the Windows handler can send all the cancel requests since all the info is available in one process, whereas on Unix each process sends a cancel only for its own database connection. In passing, fix an old problem that DisconnectDatabase tends to send a cancel request before exiting a parallel worker, even if nothing went wrong. This is at least a waste of cycles, and could lead to unexpected log messages, or maybe even data loss if it happened in pg_restore (though in the current code the problem seems to affect only pg_dump). The cause was that after a COPY step, pg_dump was leaving libpq in PGASYNC_BUSY state, causing PQtransactionStatus() to report PQTRANS_ACTIVE. That's normally harmless because the next PQexec() will silently clear the PGASYNC_BUSY state; but in a parallel worker we might exit without any additional SQL commands after a COPY step. So add an extra PQgetResult() call after a COPY to allow libpq to return to PGASYNC_IDLE state. This is a bug fix, IMO, so back-patch to 9.3 where parallel dump/restore were introduced. Thanks to Kyotaro Horiguchi for Windows testing and code suggestions. Original-Patch: <7005.1464657274@sss.pgh.pa.us> Discussion: <20160602.174941.256342236.horiguchi.kyotaro@lab.ntt.co.jp>
2016-06-02 19:27:53 +02:00
/* arrange for SIGINT to issue a query cancel on this connection */
set_archive_cancel_info(AH, AH->connection);
}
/*
* Close the connection to the database and also cancel off the query if we
* have one running.
*/
void
DisconnectDatabase(Archive *AHX)
{
ArchiveHandle *AH = (ArchiveHandle *) AHX;
char errbuf[1];
if (!AH->connection)
return;
Redesign handling of SIGTERM/control-C in parallel pg_dump/pg_restore. Formerly, Unix builds of pg_dump/pg_restore would trap SIGINT and similar signals and set a flag that was tested in various data-transfer loops. This was prone to errors of omission (cf commit 3c8aa6654); and even if the client-side response was prompt, we did nothing that would cause long-running SQL commands (e.g. CREATE INDEX) to terminate early. Also, the master process would effectively do nothing at all upon receipt of SIGINT; the only reason it seemed to work was that in typical scenarios the signal would also be delivered to the child processes. We should support termination when a signal is delivered only to the master process, though. Windows builds had no console interrupt handler, so they would just fall over immediately at control-C, again leaving long-running SQL commands to finish unmolested. To fix, remove the flag-checking approach altogether. Instead, allow the Unix signal handler to send a cancel request directly and then exit(1). In the master process, also have it forward the signal to the children. On Windows, add a console interrupt handler that behaves approximately the same. The main difference is that a single execution of the Windows handler can send all the cancel requests since all the info is available in one process, whereas on Unix each process sends a cancel only for its own database connection. In passing, fix an old problem that DisconnectDatabase tends to send a cancel request before exiting a parallel worker, even if nothing went wrong. This is at least a waste of cycles, and could lead to unexpected log messages, or maybe even data loss if it happened in pg_restore (though in the current code the problem seems to affect only pg_dump). The cause was that after a COPY step, pg_dump was leaving libpq in PGASYNC_BUSY state, causing PQtransactionStatus() to report PQTRANS_ACTIVE. That's normally harmless because the next PQexec() will silently clear the PGASYNC_BUSY state; but in a parallel worker we might exit without any additional SQL commands after a COPY step. So add an extra PQgetResult() call after a COPY to allow libpq to return to PGASYNC_IDLE state. This is a bug fix, IMO, so back-patch to 9.3 where parallel dump/restore were introduced. Thanks to Kyotaro Horiguchi for Windows testing and code suggestions. Original-Patch: <7005.1464657274@sss.pgh.pa.us> Discussion: <20160602.174941.256342236.horiguchi.kyotaro@lab.ntt.co.jp>
2016-06-02 19:27:53 +02:00
if (AH->connCancel)
{
Redesign handling of SIGTERM/control-C in parallel pg_dump/pg_restore. Formerly, Unix builds of pg_dump/pg_restore would trap SIGINT and similar signals and set a flag that was tested in various data-transfer loops. This was prone to errors of omission (cf commit 3c8aa6654); and even if the client-side response was prompt, we did nothing that would cause long-running SQL commands (e.g. CREATE INDEX) to terminate early. Also, the master process would effectively do nothing at all upon receipt of SIGINT; the only reason it seemed to work was that in typical scenarios the signal would also be delivered to the child processes. We should support termination when a signal is delivered only to the master process, though. Windows builds had no console interrupt handler, so they would just fall over immediately at control-C, again leaving long-running SQL commands to finish unmolested. To fix, remove the flag-checking approach altogether. Instead, allow the Unix signal handler to send a cancel request directly and then exit(1). In the master process, also have it forward the signal to the children. On Windows, add a console interrupt handler that behaves approximately the same. The main difference is that a single execution of the Windows handler can send all the cancel requests since all the info is available in one process, whereas on Unix each process sends a cancel only for its own database connection. In passing, fix an old problem that DisconnectDatabase tends to send a cancel request before exiting a parallel worker, even if nothing went wrong. This is at least a waste of cycles, and could lead to unexpected log messages, or maybe even data loss if it happened in pg_restore (though in the current code the problem seems to affect only pg_dump). The cause was that after a COPY step, pg_dump was leaving libpq in PGASYNC_BUSY state, causing PQtransactionStatus() to report PQTRANS_ACTIVE. That's normally harmless because the next PQexec() will silently clear the PGASYNC_BUSY state; but in a parallel worker we might exit without any additional SQL commands after a COPY step. So add an extra PQgetResult() call after a COPY to allow libpq to return to PGASYNC_IDLE state. This is a bug fix, IMO, so back-patch to 9.3 where parallel dump/restore were introduced. Thanks to Kyotaro Horiguchi for Windows testing and code suggestions. Original-Patch: <7005.1464657274@sss.pgh.pa.us> Discussion: <20160602.174941.256342236.horiguchi.kyotaro@lab.ntt.co.jp>
2016-06-02 19:27:53 +02:00
/*
* If we have an active query, send a cancel before closing, ignoring
* any errors. This is of no use for a normal exit, but might be
* helpful during pg_fatal().
Redesign handling of SIGTERM/control-C in parallel pg_dump/pg_restore. Formerly, Unix builds of pg_dump/pg_restore would trap SIGINT and similar signals and set a flag that was tested in various data-transfer loops. This was prone to errors of omission (cf commit 3c8aa6654); and even if the client-side response was prompt, we did nothing that would cause long-running SQL commands (e.g. CREATE INDEX) to terminate early. Also, the master process would effectively do nothing at all upon receipt of SIGINT; the only reason it seemed to work was that in typical scenarios the signal would also be delivered to the child processes. We should support termination when a signal is delivered only to the master process, though. Windows builds had no console interrupt handler, so they would just fall over immediately at control-C, again leaving long-running SQL commands to finish unmolested. To fix, remove the flag-checking approach altogether. Instead, allow the Unix signal handler to send a cancel request directly and then exit(1). In the master process, also have it forward the signal to the children. On Windows, add a console interrupt handler that behaves approximately the same. The main difference is that a single execution of the Windows handler can send all the cancel requests since all the info is available in one process, whereas on Unix each process sends a cancel only for its own database connection. In passing, fix an old problem that DisconnectDatabase tends to send a cancel request before exiting a parallel worker, even if nothing went wrong. This is at least a waste of cycles, and could lead to unexpected log messages, or maybe even data loss if it happened in pg_restore (though in the current code the problem seems to affect only pg_dump). The cause was that after a COPY step, pg_dump was leaving libpq in PGASYNC_BUSY state, causing PQtransactionStatus() to report PQTRANS_ACTIVE. That's normally harmless because the next PQexec() will silently clear the PGASYNC_BUSY state; but in a parallel worker we might exit without any additional SQL commands after a COPY step. So add an extra PQgetResult() call after a COPY to allow libpq to return to PGASYNC_IDLE state. This is a bug fix, IMO, so back-patch to 9.3 where parallel dump/restore were introduced. Thanks to Kyotaro Horiguchi for Windows testing and code suggestions. Original-Patch: <7005.1464657274@sss.pgh.pa.us> Discussion: <20160602.174941.256342236.horiguchi.kyotaro@lab.ntt.co.jp>
2016-06-02 19:27:53 +02:00
*/
if (PQtransactionStatus(AH->connection) == PQTRANS_ACTIVE)
(void) PQcancel(AH->connCancel, errbuf, sizeof(errbuf));
Redesign handling of SIGTERM/control-C in parallel pg_dump/pg_restore. Formerly, Unix builds of pg_dump/pg_restore would trap SIGINT and similar signals and set a flag that was tested in various data-transfer loops. This was prone to errors of omission (cf commit 3c8aa6654); and even if the client-side response was prompt, we did nothing that would cause long-running SQL commands (e.g. CREATE INDEX) to terminate early. Also, the master process would effectively do nothing at all upon receipt of SIGINT; the only reason it seemed to work was that in typical scenarios the signal would also be delivered to the child processes. We should support termination when a signal is delivered only to the master process, though. Windows builds had no console interrupt handler, so they would just fall over immediately at control-C, again leaving long-running SQL commands to finish unmolested. To fix, remove the flag-checking approach altogether. Instead, allow the Unix signal handler to send a cancel request directly and then exit(1). In the master process, also have it forward the signal to the children. On Windows, add a console interrupt handler that behaves approximately the same. The main difference is that a single execution of the Windows handler can send all the cancel requests since all the info is available in one process, whereas on Unix each process sends a cancel only for its own database connection. In passing, fix an old problem that DisconnectDatabase tends to send a cancel request before exiting a parallel worker, even if nothing went wrong. This is at least a waste of cycles, and could lead to unexpected log messages, or maybe even data loss if it happened in pg_restore (though in the current code the problem seems to affect only pg_dump). The cause was that after a COPY step, pg_dump was leaving libpq in PGASYNC_BUSY state, causing PQtransactionStatus() to report PQTRANS_ACTIVE. That's normally harmless because the next PQexec() will silently clear the PGASYNC_BUSY state; but in a parallel worker we might exit without any additional SQL commands after a COPY step. So add an extra PQgetResult() call after a COPY to allow libpq to return to PGASYNC_IDLE state. This is a bug fix, IMO, so back-patch to 9.3 where parallel dump/restore were introduced. Thanks to Kyotaro Horiguchi for Windows testing and code suggestions. Original-Patch: <7005.1464657274@sss.pgh.pa.us> Discussion: <20160602.174941.256342236.horiguchi.kyotaro@lab.ntt.co.jp>
2016-06-02 19:27:53 +02:00
/*
* Prevent signal handler from sending a cancel after this.
*/
set_archive_cancel_info(AH, NULL);
}
PQfinish(AH->connection);
AH->connection = NULL;
}
PGconn *
GetConnection(Archive *AHX)
{
ArchiveHandle *AH = (ArchiveHandle *) AHX;
return AH->connection;
}
static void
notice_processor(void *arg, const char *message)
{
pg_log_info("%s", message);
}
/* Like pg_fatal(), but with a complaint about a particular query. */
static void
Unified logging system for command-line programs This unifies the various ad hoc logging (message printing, error printing) systems used throughout the command-line programs. Features: - Program name is automatically prefixed. - Message string does not end with newline. This removes a common source of inconsistencies and omissions. - Additionally, a final newline is automatically stripped, simplifying use of PQerrorMessage() etc., another common source of mistakes. - I converted error message strings to use %m where possible. - As a result of the above several points, more translatable message strings can be shared between different components and between frontends and backend, without gratuitous punctuation or whitespace differences. - There is support for setting a "log level". This is not meant to be user-facing, but can be used internally to implement debug or verbose modes. - Lazy argument evaluation, so no significant overhead if logging at some level is disabled. - Some color in the messages, similar to gcc and clang. Set PG_COLOR=auto to try it out. Some colors are predefined, but can be customized by setting PG_COLORS. - Common files (common/, fe_utils/, etc.) can handle logging much more simply by just using one API without worrying too much about the context of the calling program, requiring callbacks, or having to pass "progname" around everywhere. - Some programs called setvbuf() to make sure that stderr is unbuffered, even on Windows. But not all programs did that. This is now done centrally. Soft goals: - Reduces vertical space use and visual complexity of error reporting in the source code. - Encourages more deliberate classification of messages. For example, in some cases it wasn't clear without analyzing the surrounding code whether a message was meant as an error or just an info. - Concepts and terms are vaguely aligned with popular logging frameworks such as log4j and Python logging. This is all just about printing stuff out. Nothing affects program flow (e.g., fatal exits). The uses are just too varied to do that. Some existing code had wrappers that do some kind of print-and-exit, and I adapted those. I tried to keep the output mostly the same, but there is a lot of historical baggage to unwind and special cases to consider, and I might not always have succeeded. One significant change is that pg_rewind used to write all error messages to stdout. That is now changed to stderr. Reviewed-by: Donald Dong <xdong@csumb.edu> Reviewed-by: Arthur Zakirov <a.zakirov@postgrespro.ru> Discussion: https://www.postgresql.org/message-id/flat/6a609b43-4f57-7348-6480-bd022f924310@2ndquadrant.com
2019-04-01 14:24:37 +02:00
die_on_query_failure(ArchiveHandle *AH, const char *query)
{
Unified logging system for command-line programs This unifies the various ad hoc logging (message printing, error printing) systems used throughout the command-line programs. Features: - Program name is automatically prefixed. - Message string does not end with newline. This removes a common source of inconsistencies and omissions. - Additionally, a final newline is automatically stripped, simplifying use of PQerrorMessage() etc., another common source of mistakes. - I converted error message strings to use %m where possible. - As a result of the above several points, more translatable message strings can be shared between different components and between frontends and backend, without gratuitous punctuation or whitespace differences. - There is support for setting a "log level". This is not meant to be user-facing, but can be used internally to implement debug or verbose modes. - Lazy argument evaluation, so no significant overhead if logging at some level is disabled. - Some color in the messages, similar to gcc and clang. Set PG_COLOR=auto to try it out. Some colors are predefined, but can be customized by setting PG_COLORS. - Common files (common/, fe_utils/, etc.) can handle logging much more simply by just using one API without worrying too much about the context of the calling program, requiring callbacks, or having to pass "progname" around everywhere. - Some programs called setvbuf() to make sure that stderr is unbuffered, even on Windows. But not all programs did that. This is now done centrally. Soft goals: - Reduces vertical space use and visual complexity of error reporting in the source code. - Encourages more deliberate classification of messages. For example, in some cases it wasn't clear without analyzing the surrounding code whether a message was meant as an error or just an info. - Concepts and terms are vaguely aligned with popular logging frameworks such as log4j and Python logging. This is all just about printing stuff out. Nothing affects program flow (e.g., fatal exits). The uses are just too varied to do that. Some existing code had wrappers that do some kind of print-and-exit, and I adapted those. I tried to keep the output mostly the same, but there is a lot of historical baggage to unwind and special cases to consider, and I might not always have succeeded. One significant change is that pg_rewind used to write all error messages to stdout. That is now changed to stderr. Reviewed-by: Donald Dong <xdong@csumb.edu> Reviewed-by: Arthur Zakirov <a.zakirov@postgrespro.ru> Discussion: https://www.postgresql.org/message-id/flat/6a609b43-4f57-7348-6480-bd022f924310@2ndquadrant.com
2019-04-01 14:24:37 +02:00
pg_log_error("query failed: %s",
PQerrorMessage(AH->connection));
pg_log_error_detail("Query was: %s", query);
exit(1);
}
void
ExecuteSqlStatement(Archive *AHX, const char *query)
{
ArchiveHandle *AH = (ArchiveHandle *) AHX;
PGresult *res;
res = PQexec(AH->connection, query);
if (PQresultStatus(res) != PGRES_COMMAND_OK)
Unified logging system for command-line programs This unifies the various ad hoc logging (message printing, error printing) systems used throughout the command-line programs. Features: - Program name is automatically prefixed. - Message string does not end with newline. This removes a common source of inconsistencies and omissions. - Additionally, a final newline is automatically stripped, simplifying use of PQerrorMessage() etc., another common source of mistakes. - I converted error message strings to use %m where possible. - As a result of the above several points, more translatable message strings can be shared between different components and between frontends and backend, without gratuitous punctuation or whitespace differences. - There is support for setting a "log level". This is not meant to be user-facing, but can be used internally to implement debug or verbose modes. - Lazy argument evaluation, so no significant overhead if logging at some level is disabled. - Some color in the messages, similar to gcc and clang. Set PG_COLOR=auto to try it out. Some colors are predefined, but can be customized by setting PG_COLORS. - Common files (common/, fe_utils/, etc.) can handle logging much more simply by just using one API without worrying too much about the context of the calling program, requiring callbacks, or having to pass "progname" around everywhere. - Some programs called setvbuf() to make sure that stderr is unbuffered, even on Windows. But not all programs did that. This is now done centrally. Soft goals: - Reduces vertical space use and visual complexity of error reporting in the source code. - Encourages more deliberate classification of messages. For example, in some cases it wasn't clear without analyzing the surrounding code whether a message was meant as an error or just an info. - Concepts and terms are vaguely aligned with popular logging frameworks such as log4j and Python logging. This is all just about printing stuff out. Nothing affects program flow (e.g., fatal exits). The uses are just too varied to do that. Some existing code had wrappers that do some kind of print-and-exit, and I adapted those. I tried to keep the output mostly the same, but there is a lot of historical baggage to unwind and special cases to consider, and I might not always have succeeded. One significant change is that pg_rewind used to write all error messages to stdout. That is now changed to stderr. Reviewed-by: Donald Dong <xdong@csumb.edu> Reviewed-by: Arthur Zakirov <a.zakirov@postgrespro.ru> Discussion: https://www.postgresql.org/message-id/flat/6a609b43-4f57-7348-6480-bd022f924310@2ndquadrant.com
2019-04-01 14:24:37 +02:00
die_on_query_failure(AH, query);
PQclear(res);
}
PGresult *
ExecuteSqlQuery(Archive *AHX, const char *query, ExecStatusType status)
{
ArchiveHandle *AH = (ArchiveHandle *) AHX;
PGresult *res;
res = PQexec(AH->connection, query);
if (PQresultStatus(res) != status)
Unified logging system for command-line programs This unifies the various ad hoc logging (message printing, error printing) systems used throughout the command-line programs. Features: - Program name is automatically prefixed. - Message string does not end with newline. This removes a common source of inconsistencies and omissions. - Additionally, a final newline is automatically stripped, simplifying use of PQerrorMessage() etc., another common source of mistakes. - I converted error message strings to use %m where possible. - As a result of the above several points, more translatable message strings can be shared between different components and between frontends and backend, without gratuitous punctuation or whitespace differences. - There is support for setting a "log level". This is not meant to be user-facing, but can be used internally to implement debug or verbose modes. - Lazy argument evaluation, so no significant overhead if logging at some level is disabled. - Some color in the messages, similar to gcc and clang. Set PG_COLOR=auto to try it out. Some colors are predefined, but can be customized by setting PG_COLORS. - Common files (common/, fe_utils/, etc.) can handle logging much more simply by just using one API without worrying too much about the context of the calling program, requiring callbacks, or having to pass "progname" around everywhere. - Some programs called setvbuf() to make sure that stderr is unbuffered, even on Windows. But not all programs did that. This is now done centrally. Soft goals: - Reduces vertical space use and visual complexity of error reporting in the source code. - Encourages more deliberate classification of messages. For example, in some cases it wasn't clear without analyzing the surrounding code whether a message was meant as an error or just an info. - Concepts and terms are vaguely aligned with popular logging frameworks such as log4j and Python logging. This is all just about printing stuff out. Nothing affects program flow (e.g., fatal exits). The uses are just too varied to do that. Some existing code had wrappers that do some kind of print-and-exit, and I adapted those. I tried to keep the output mostly the same, but there is a lot of historical baggage to unwind and special cases to consider, and I might not always have succeeded. One significant change is that pg_rewind used to write all error messages to stdout. That is now changed to stderr. Reviewed-by: Donald Dong <xdong@csumb.edu> Reviewed-by: Arthur Zakirov <a.zakirov@postgrespro.ru> Discussion: https://www.postgresql.org/message-id/flat/6a609b43-4f57-7348-6480-bd022f924310@2ndquadrant.com
2019-04-01 14:24:37 +02:00
die_on_query_failure(AH, query);
return res;
}
/*
* Execute an SQL query and verify that we got exactly one row back.
*/
PGresult *
ExecuteSqlQueryForSingleRow(Archive *fout, const char *query)
{
PGresult *res;
int ntups;
res = ExecuteSqlQuery(fout, query, PGRES_TUPLES_OK);
/* Expecting a single result only */
ntups = PQntuples(res);
if (ntups != 1)
pg_fatal(ngettext("query returned %d row instead of one: %s",
"query returned %d rows instead of one: %s",
ntups),
ntups, query);
return res;
}
/*
* Convenience function to send a query.
* Monitors result to detect COPY statements
*/
static void
ExecuteSqlCommand(ArchiveHandle *AH, const char *qry, const char *desc)
{
PGconn *conn = AH->connection;
PGresult *res;
#ifdef NOT_USED
fprintf(stderr, "Executing: '%s'\n\n", qry);
#endif
res = PQexec(conn, qry);
switch (PQresultStatus(res))
{
case PGRES_COMMAND_OK:
case PGRES_TUPLES_OK:
case PGRES_EMPTY_QUERY:
/* A-OK */
break;
case PGRES_COPY_IN:
/* Assume this is an expected result */
AH->pgCopyIn = true;
break;
default:
/* trouble */
Unified logging system for command-line programs This unifies the various ad hoc logging (message printing, error printing) systems used throughout the command-line programs. Features: - Program name is automatically prefixed. - Message string does not end with newline. This removes a common source of inconsistencies and omissions. - Additionally, a final newline is automatically stripped, simplifying use of PQerrorMessage() etc., another common source of mistakes. - I converted error message strings to use %m where possible. - As a result of the above several points, more translatable message strings can be shared between different components and between frontends and backend, without gratuitous punctuation or whitespace differences. - There is support for setting a "log level". This is not meant to be user-facing, but can be used internally to implement debug or verbose modes. - Lazy argument evaluation, so no significant overhead if logging at some level is disabled. - Some color in the messages, similar to gcc and clang. Set PG_COLOR=auto to try it out. Some colors are predefined, but can be customized by setting PG_COLORS. - Common files (common/, fe_utils/, etc.) can handle logging much more simply by just using one API without worrying too much about the context of the calling program, requiring callbacks, or having to pass "progname" around everywhere. - Some programs called setvbuf() to make sure that stderr is unbuffered, even on Windows. But not all programs did that. This is now done centrally. Soft goals: - Reduces vertical space use and visual complexity of error reporting in the source code. - Encourages more deliberate classification of messages. For example, in some cases it wasn't clear without analyzing the surrounding code whether a message was meant as an error or just an info. - Concepts and terms are vaguely aligned with popular logging frameworks such as log4j and Python logging. This is all just about printing stuff out. Nothing affects program flow (e.g., fatal exits). The uses are just too varied to do that. Some existing code had wrappers that do some kind of print-and-exit, and I adapted those. I tried to keep the output mostly the same, but there is a lot of historical baggage to unwind and special cases to consider, and I might not always have succeeded. One significant change is that pg_rewind used to write all error messages to stdout. That is now changed to stderr. Reviewed-by: Donald Dong <xdong@csumb.edu> Reviewed-by: Arthur Zakirov <a.zakirov@postgrespro.ru> Discussion: https://www.postgresql.org/message-id/flat/6a609b43-4f57-7348-6480-bd022f924310@2ndquadrant.com
2019-04-01 14:24:37 +02:00
warn_or_exit_horribly(AH, "%s: %sCommand was: %s",
desc, PQerrorMessage(conn), qry);
break;
}
PQclear(res);
}
/*
* Process non-COPY table data (that is, INSERT commands).
*
* The commands have been run together as one long string for compressibility,
* and we are receiving them in bufferloads with arbitrary boundaries, so we
* have to locate command boundaries and save partial commands across calls.
* All state must be kept in AH->sqlparse, not in local variables of this
* routine. We assume that AH->sqlparse was filled with zeroes when created.
*
* We have to lex the data to the extent of identifying literals and quoted
* identifiers, so that we can recognize statement-terminating semicolons.
* We assume that INSERT data will not contain SQL comments, E'' literals,
* or dollar-quoted strings, so this is much simpler than a full SQL lexer.
*
* Note: when restoring from a pre-9.0 dump file, this code is also used to
* process BLOB COMMENTS data, which has the same problem of containing
* multiple SQL commands that might be split across bufferloads. Fortunately,
* that data won't contain anything complicated to lex either.
*/
static void
ExecuteSimpleCommands(ArchiveHandle *AH, const char *buf, size_t bufLen)
{
const char *qry = buf;
const char *eos = buf + bufLen;
/* initialize command buffer if first time through */
if (AH->sqlparse.curCmd == NULL)
AH->sqlparse.curCmd = createPQExpBuffer();
for (; qry < eos; qry++)
{
char ch = *qry;
/* For neatness, we skip any newlines between commands */
if (!(ch == '\n' && AH->sqlparse.curCmd->len == 0))
appendPQExpBufferChar(AH->sqlparse.curCmd, ch);
switch (AH->sqlparse.state)
{
case SQL_SCAN: /* Default state == 0, set in _allocAH */
if (ch == ';')
{
/*
* We've found the end of a statement. Send it and reset
* the buffer.
*/
ExecuteSqlCommand(AH, AH->sqlparse.curCmd->data,
"could not execute query");
resetPQExpBuffer(AH->sqlparse.curCmd);
}
else if (ch == '\'')
{
AH->sqlparse.state = SQL_IN_SINGLE_QUOTE;
AH->sqlparse.backSlash = false;
}
else if (ch == '"')
{
AH->sqlparse.state = SQL_IN_DOUBLE_QUOTE;
}
break;
case SQL_IN_SINGLE_QUOTE:
/* We needn't handle '' specially */
if (ch == '\'' && !AH->sqlparse.backSlash)
AH->sqlparse.state = SQL_SCAN;
else if (ch == '\\' && !AH->public.std_strings)
AH->sqlparse.backSlash = !AH->sqlparse.backSlash;
else
AH->sqlparse.backSlash = false;
break;
case SQL_IN_DOUBLE_QUOTE:
/* We needn't handle "" specially */
if (ch == '"')
AH->sqlparse.state = SQL_SCAN;
break;
}
}
}
/*
* Implement ahwrite() for direct-to-DB restore
*/
int
ExecuteSqlCommandBuf(Archive *AHX, const char *buf, size_t bufLen)
{
ArchiveHandle *AH = (ArchiveHandle *) AHX;
if (AH->outputKind == OUTPUT_COPYDATA)
{
/*
* COPY data.
*
* We drop the data on the floor if libpq has failed to enter COPY
* mode; this allows us to behave reasonably when trying to continue
* after an error in a COPY command.
*/
if (AH->pgCopyIn &&
PQputCopyData(AH->connection, buf, bufLen) <= 0)
pg_fatal("error returned by PQputCopyData: %s",
PQerrorMessage(AH->connection));
}
else if (AH->outputKind == OUTPUT_OTHERDATA)
{
/*
* Table data expressed as INSERT commands; or, in old dump files,
* BLOB COMMENTS data (which is expressed as COMMENT ON commands).
*/
ExecuteSimpleCommands(AH, buf, bufLen);
}
else
{
/*
* General SQL commands; we assume that commands will not be split
* across calls.
*
* In most cases the data passed to us will be a null-terminated
* string, but if it's not, we have to add a trailing null.
*/
if (buf[bufLen] == '\0')
ExecuteSqlCommand(AH, buf, "could not execute query");
else
{
char *str = (char *) pg_malloc(bufLen + 1);
memcpy(str, buf, bufLen);
str[bufLen] = '\0';
ExecuteSqlCommand(AH, str, "could not execute query");
free(str);
}
}
return bufLen;
}
/*
* Terminate a COPY operation during direct-to-DB restore
*/
void
EndDBCopyMode(Archive *AHX, const char *tocEntryTag)
{
ArchiveHandle *AH = (ArchiveHandle *) AHX;
if (AH->pgCopyIn)
{
PGresult *res;
if (PQputCopyEnd(AH->connection, NULL) <= 0)
pg_fatal("error returned by PQputCopyEnd: %s",
PQerrorMessage(AH->connection));
/* Check command status and return to normal libpq state */
res = PQgetResult(AH->connection);
if (PQresultStatus(res) != PGRES_COMMAND_OK)
Unified logging system for command-line programs This unifies the various ad hoc logging (message printing, error printing) systems used throughout the command-line programs. Features: - Program name is automatically prefixed. - Message string does not end with newline. This removes a common source of inconsistencies and omissions. - Additionally, a final newline is automatically stripped, simplifying use of PQerrorMessage() etc., another common source of mistakes. - I converted error message strings to use %m where possible. - As a result of the above several points, more translatable message strings can be shared between different components and between frontends and backend, without gratuitous punctuation or whitespace differences. - There is support for setting a "log level". This is not meant to be user-facing, but can be used internally to implement debug or verbose modes. - Lazy argument evaluation, so no significant overhead if logging at some level is disabled. - Some color in the messages, similar to gcc and clang. Set PG_COLOR=auto to try it out. Some colors are predefined, but can be customized by setting PG_COLORS. - Common files (common/, fe_utils/, etc.) can handle logging much more simply by just using one API without worrying too much about the context of the calling program, requiring callbacks, or having to pass "progname" around everywhere. - Some programs called setvbuf() to make sure that stderr is unbuffered, even on Windows. But not all programs did that. This is now done centrally. Soft goals: - Reduces vertical space use and visual complexity of error reporting in the source code. - Encourages more deliberate classification of messages. For example, in some cases it wasn't clear without analyzing the surrounding code whether a message was meant as an error or just an info. - Concepts and terms are vaguely aligned with popular logging frameworks such as log4j and Python logging. This is all just about printing stuff out. Nothing affects program flow (e.g., fatal exits). The uses are just too varied to do that. Some existing code had wrappers that do some kind of print-and-exit, and I adapted those. I tried to keep the output mostly the same, but there is a lot of historical baggage to unwind and special cases to consider, and I might not always have succeeded. One significant change is that pg_rewind used to write all error messages to stdout. That is now changed to stderr. Reviewed-by: Donald Dong <xdong@csumb.edu> Reviewed-by: Arthur Zakirov <a.zakirov@postgrespro.ru> Discussion: https://www.postgresql.org/message-id/flat/6a609b43-4f57-7348-6480-bd022f924310@2ndquadrant.com
2019-04-01 14:24:37 +02:00
warn_or_exit_horribly(AH, "COPY failed for table \"%s\": %s",
tocEntryTag, PQerrorMessage(AH->connection));
PQclear(res);
Redesign handling of SIGTERM/control-C in parallel pg_dump/pg_restore. Formerly, Unix builds of pg_dump/pg_restore would trap SIGINT and similar signals and set a flag that was tested in various data-transfer loops. This was prone to errors of omission (cf commit 3c8aa6654); and even if the client-side response was prompt, we did nothing that would cause long-running SQL commands (e.g. CREATE INDEX) to terminate early. Also, the master process would effectively do nothing at all upon receipt of SIGINT; the only reason it seemed to work was that in typical scenarios the signal would also be delivered to the child processes. We should support termination when a signal is delivered only to the master process, though. Windows builds had no console interrupt handler, so they would just fall over immediately at control-C, again leaving long-running SQL commands to finish unmolested. To fix, remove the flag-checking approach altogether. Instead, allow the Unix signal handler to send a cancel request directly and then exit(1). In the master process, also have it forward the signal to the children. On Windows, add a console interrupt handler that behaves approximately the same. The main difference is that a single execution of the Windows handler can send all the cancel requests since all the info is available in one process, whereas on Unix each process sends a cancel only for its own database connection. In passing, fix an old problem that DisconnectDatabase tends to send a cancel request before exiting a parallel worker, even if nothing went wrong. This is at least a waste of cycles, and could lead to unexpected log messages, or maybe even data loss if it happened in pg_restore (though in the current code the problem seems to affect only pg_dump). The cause was that after a COPY step, pg_dump was leaving libpq in PGASYNC_BUSY state, causing PQtransactionStatus() to report PQTRANS_ACTIVE. That's normally harmless because the next PQexec() will silently clear the PGASYNC_BUSY state; but in a parallel worker we might exit without any additional SQL commands after a COPY step. So add an extra PQgetResult() call after a COPY to allow libpq to return to PGASYNC_IDLE state. This is a bug fix, IMO, so back-patch to 9.3 where parallel dump/restore were introduced. Thanks to Kyotaro Horiguchi for Windows testing and code suggestions. Original-Patch: <7005.1464657274@sss.pgh.pa.us> Discussion: <20160602.174941.256342236.horiguchi.kyotaro@lab.ntt.co.jp>
2016-06-02 19:27:53 +02:00
/* Do this to ensure we've pumped libpq back to idle state */
if (PQgetResult(AH->connection) != NULL)
Unified logging system for command-line programs This unifies the various ad hoc logging (message printing, error printing) systems used throughout the command-line programs. Features: - Program name is automatically prefixed. - Message string does not end with newline. This removes a common source of inconsistencies and omissions. - Additionally, a final newline is automatically stripped, simplifying use of PQerrorMessage() etc., another common source of mistakes. - I converted error message strings to use %m where possible. - As a result of the above several points, more translatable message strings can be shared between different components and between frontends and backend, without gratuitous punctuation or whitespace differences. - There is support for setting a "log level". This is not meant to be user-facing, but can be used internally to implement debug or verbose modes. - Lazy argument evaluation, so no significant overhead if logging at some level is disabled. - Some color in the messages, similar to gcc and clang. Set PG_COLOR=auto to try it out. Some colors are predefined, but can be customized by setting PG_COLORS. - Common files (common/, fe_utils/, etc.) can handle logging much more simply by just using one API without worrying too much about the context of the calling program, requiring callbacks, or having to pass "progname" around everywhere. - Some programs called setvbuf() to make sure that stderr is unbuffered, even on Windows. But not all programs did that. This is now done centrally. Soft goals: - Reduces vertical space use and visual complexity of error reporting in the source code. - Encourages more deliberate classification of messages. For example, in some cases it wasn't clear without analyzing the surrounding code whether a message was meant as an error or just an info. - Concepts and terms are vaguely aligned with popular logging frameworks such as log4j and Python logging. This is all just about printing stuff out. Nothing affects program flow (e.g., fatal exits). The uses are just too varied to do that. Some existing code had wrappers that do some kind of print-and-exit, and I adapted those. I tried to keep the output mostly the same, but there is a lot of historical baggage to unwind and special cases to consider, and I might not always have succeeded. One significant change is that pg_rewind used to write all error messages to stdout. That is now changed to stderr. Reviewed-by: Donald Dong <xdong@csumb.edu> Reviewed-by: Arthur Zakirov <a.zakirov@postgrespro.ru> Discussion: https://www.postgresql.org/message-id/flat/6a609b43-4f57-7348-6480-bd022f924310@2ndquadrant.com
2019-04-01 14:24:37 +02:00
pg_log_warning("unexpected extra results during COPY of table \"%s\"",
Redesign handling of SIGTERM/control-C in parallel pg_dump/pg_restore. Formerly, Unix builds of pg_dump/pg_restore would trap SIGINT and similar signals and set a flag that was tested in various data-transfer loops. This was prone to errors of omission (cf commit 3c8aa6654); and even if the client-side response was prompt, we did nothing that would cause long-running SQL commands (e.g. CREATE INDEX) to terminate early. Also, the master process would effectively do nothing at all upon receipt of SIGINT; the only reason it seemed to work was that in typical scenarios the signal would also be delivered to the child processes. We should support termination when a signal is delivered only to the master process, though. Windows builds had no console interrupt handler, so they would just fall over immediately at control-C, again leaving long-running SQL commands to finish unmolested. To fix, remove the flag-checking approach altogether. Instead, allow the Unix signal handler to send a cancel request directly and then exit(1). In the master process, also have it forward the signal to the children. On Windows, add a console interrupt handler that behaves approximately the same. The main difference is that a single execution of the Windows handler can send all the cancel requests since all the info is available in one process, whereas on Unix each process sends a cancel only for its own database connection. In passing, fix an old problem that DisconnectDatabase tends to send a cancel request before exiting a parallel worker, even if nothing went wrong. This is at least a waste of cycles, and could lead to unexpected log messages, or maybe even data loss if it happened in pg_restore (though in the current code the problem seems to affect only pg_dump). The cause was that after a COPY step, pg_dump was leaving libpq in PGASYNC_BUSY state, causing PQtransactionStatus() to report PQTRANS_ACTIVE. That's normally harmless because the next PQexec() will silently clear the PGASYNC_BUSY state; but in a parallel worker we might exit without any additional SQL commands after a COPY step. So add an extra PQgetResult() call after a COPY to allow libpq to return to PGASYNC_IDLE state. This is a bug fix, IMO, so back-patch to 9.3 where parallel dump/restore were introduced. Thanks to Kyotaro Horiguchi for Windows testing and code suggestions. Original-Patch: <7005.1464657274@sss.pgh.pa.us> Discussion: <20160602.174941.256342236.horiguchi.kyotaro@lab.ntt.co.jp>
2016-06-02 19:27:53 +02:00
tocEntryTag);
AH->pgCopyIn = false;
}
}
void
StartTransaction(Archive *AHX)
{
ArchiveHandle *AH = (ArchiveHandle *) AHX;
ExecuteSqlCommand(AH, "BEGIN", "could not start database transaction");
}
void
CommitTransaction(Archive *AHX)
{
ArchiveHandle *AH = (ArchiveHandle *) AHX;
ExecuteSqlCommand(AH, "COMMIT", "could not commit database transaction");
}
Rearrange pg_dump's handling of large objects for better efficiency. Commit c0d5be5d6 caused pg_dump to create a separate BLOB metadata TOC entry for each large object (blob), but it did not touch the ancient decision to put all the blobs' data into a single "BLOBS" TOC entry. This is bad for a few reasons: for databases with millions of blobs, the TOC becomes unreasonably large, causing performance issues; selective restore of just some blobs is quite impossible; and we cannot parallelize either dump or restore of the blob data, since our architecture for that relies on farming out whole TOC entries to worker processes. To improve matters, let's group multiple blobs into each blob metadata TOC entry, and then make corresponding per-group blob data TOC entries. Selective restore using pg_restore's -l/-L switches is then possible, though only at the group level. (Perhaps we should provide a switch to allow forcing one-blob-per-group for users who need precise selective restore and don't have huge numbers of blobs. This patch doesn't do that, instead just hard-wiring the maximum number of blobs per entry at 1000.) The blobs in a group must all have the same owner, since the TOC entry format only allows one owner to be named. In this implementation we also require them to all share the same ACL (grants); the archive format wouldn't require that, but pg_dump's representation of DumpableObjects does. It seems unlikely that either restriction will be problematic for databases with huge numbers of blobs. The metadata TOC entries now have a "desc" string of "BLOB METADATA", and their "defn" string is just a newline-separated list of blob OIDs. The restore code has to generate creation commands, ALTER OWNER commands, and drop commands (for --clean mode) from that. We would need special-case code for ALTER OWNER and drop in any case, so the alternative of keeping the "defn" as directly executable SQL code for creation wouldn't buy much, and it seems like it'd bloat the archive to little purpose. Since we require the blobs of a metadata group to share the same ACL, we can furthermore store only one copy of that ACL, and then make pg_restore regenerate the appropriate commands for each blob. This saves space in the dump file not only by removing duplicative SQL command strings, but by not needing a separate TOC entry for each blob's ACL. In turn, that reduces client-side memory requirements for handling many blobs. ACL TOC entries that need this special processing are labeled as "ACL"/"LARGE OBJECTS nnn..nnn". If we have a blob with a unique ACL, continue to label it as "ACL"/"LARGE OBJECT nnn". We don't actually have to make such a distinction, but it saves a few cycles during restore for the easy case, and it seems like a good idea to not change the TOC contents unnecessarily. The data TOC entries ("BLOBS") are exactly the same as before, except that now there can be more than one, so we'd better give them identifying tag strings. Also, commit c0d5be5d6 put the new BLOB metadata TOC entries into SECTION_PRE_DATA, which perhaps is defensible in some ways, but it's a rather odd choice considering that we go out of our way to treat blobs as data. Moreover, because parallel restore handles the PRE_DATA section serially, this means we'd only get part of the parallelism speedup we could hope for. Move these entries into SECTION_DATA, letting us parallelize the lo_create calls not just the data loading when there are many blobs. Add dependencies to ensure that we won't try to load data for a blob we've not yet created. As this stands, we still generate a separate TOC entry for any comment or security label attached to a blob. I feel comfortable in believing that comments and security labels on blobs are rare, so this patch should be enough to get most of the useful TOC compression for blobs. We have to bump the archive file format version number, since existing versions of pg_restore wouldn't know they need to do something special for BLOB METADATA, plus they aren't going to work correctly with multiple BLOBS entries or multiple-large-object ACL entries. The directory and tar-file format handlers need some work for multiple BLOBS entries: they used to hard-wire the file name as "blobs.toc", which is replaced here with "blobs_<dumpid>.toc". The 002_pg_dump.pl test script also knows about that and requires minor updates. (I had to drop the test for manually-compressed blobs.toc files with LZ4, because lz4's obtuse command line design requires explicit specification of the output file name which seems impractical here. I don't think we're losing any useful test coverage thereby; that test stanza seems completely duplicative with the gzip and zstd cases anyway.) In passing, centralize management of the lo_buf used to hold data while restoring blobs. The code previously had each format handler create lo_buf, which seems rather pointless given that the format handlers all make it the same way. Moreover, the format handlers never use lo_buf directly, making this setup a failure from a separation-of-concerns standpoint. Let's move the responsibility into pg_backup_archiver.c, which is the only module concerned with lo_buf. The reason to do this in this patch is that it allows a centralized fix for the now-false assumption that we never restore blobs in parallel. Also, get rid of dead code in DropLOIfExists: it's been a long time since we had any need to be able to restore to a pre-9.0 server. Discussion: https://postgr.es/m/a9f9376f1c3343a6bb319dce294e20ac@EX13D05UWC001.ant.amazon.com
2024-04-01 22:25:56 +02:00
/*
* Issue per-blob commands for the large object(s) listed in the TocEntry
*
* The TocEntry's defn string is assumed to consist of large object OIDs,
* one per line. Wrap these in the given SQL command fragments and issue
* the commands. (cmdEnd need not include a semicolon.)
*/
void
Rearrange pg_dump's handling of large objects for better efficiency. Commit c0d5be5d6 caused pg_dump to create a separate BLOB metadata TOC entry for each large object (blob), but it did not touch the ancient decision to put all the blobs' data into a single "BLOBS" TOC entry. This is bad for a few reasons: for databases with millions of blobs, the TOC becomes unreasonably large, causing performance issues; selective restore of just some blobs is quite impossible; and we cannot parallelize either dump or restore of the blob data, since our architecture for that relies on farming out whole TOC entries to worker processes. To improve matters, let's group multiple blobs into each blob metadata TOC entry, and then make corresponding per-group blob data TOC entries. Selective restore using pg_restore's -l/-L switches is then possible, though only at the group level. (Perhaps we should provide a switch to allow forcing one-blob-per-group for users who need precise selective restore and don't have huge numbers of blobs. This patch doesn't do that, instead just hard-wiring the maximum number of blobs per entry at 1000.) The blobs in a group must all have the same owner, since the TOC entry format only allows one owner to be named. In this implementation we also require them to all share the same ACL (grants); the archive format wouldn't require that, but pg_dump's representation of DumpableObjects does. It seems unlikely that either restriction will be problematic for databases with huge numbers of blobs. The metadata TOC entries now have a "desc" string of "BLOB METADATA", and their "defn" string is just a newline-separated list of blob OIDs. The restore code has to generate creation commands, ALTER OWNER commands, and drop commands (for --clean mode) from that. We would need special-case code for ALTER OWNER and drop in any case, so the alternative of keeping the "defn" as directly executable SQL code for creation wouldn't buy much, and it seems like it'd bloat the archive to little purpose. Since we require the blobs of a metadata group to share the same ACL, we can furthermore store only one copy of that ACL, and then make pg_restore regenerate the appropriate commands for each blob. This saves space in the dump file not only by removing duplicative SQL command strings, but by not needing a separate TOC entry for each blob's ACL. In turn, that reduces client-side memory requirements for handling many blobs. ACL TOC entries that need this special processing are labeled as "ACL"/"LARGE OBJECTS nnn..nnn". If we have a blob with a unique ACL, continue to label it as "ACL"/"LARGE OBJECT nnn". We don't actually have to make such a distinction, but it saves a few cycles during restore for the easy case, and it seems like a good idea to not change the TOC contents unnecessarily. The data TOC entries ("BLOBS") are exactly the same as before, except that now there can be more than one, so we'd better give them identifying tag strings. Also, commit c0d5be5d6 put the new BLOB metadata TOC entries into SECTION_PRE_DATA, which perhaps is defensible in some ways, but it's a rather odd choice considering that we go out of our way to treat blobs as data. Moreover, because parallel restore handles the PRE_DATA section serially, this means we'd only get part of the parallelism speedup we could hope for. Move these entries into SECTION_DATA, letting us parallelize the lo_create calls not just the data loading when there are many blobs. Add dependencies to ensure that we won't try to load data for a blob we've not yet created. As this stands, we still generate a separate TOC entry for any comment or security label attached to a blob. I feel comfortable in believing that comments and security labels on blobs are rare, so this patch should be enough to get most of the useful TOC compression for blobs. We have to bump the archive file format version number, since existing versions of pg_restore wouldn't know they need to do something special for BLOB METADATA, plus they aren't going to work correctly with multiple BLOBS entries or multiple-large-object ACL entries. The directory and tar-file format handlers need some work for multiple BLOBS entries: they used to hard-wire the file name as "blobs.toc", which is replaced here with "blobs_<dumpid>.toc". The 002_pg_dump.pl test script also knows about that and requires minor updates. (I had to drop the test for manually-compressed blobs.toc files with LZ4, because lz4's obtuse command line design requires explicit specification of the output file name which seems impractical here. I don't think we're losing any useful test coverage thereby; that test stanza seems completely duplicative with the gzip and zstd cases anyway.) In passing, centralize management of the lo_buf used to hold data while restoring blobs. The code previously had each format handler create lo_buf, which seems rather pointless given that the format handlers all make it the same way. Moreover, the format handlers never use lo_buf directly, making this setup a failure from a separation-of-concerns standpoint. Let's move the responsibility into pg_backup_archiver.c, which is the only module concerned with lo_buf. The reason to do this in this patch is that it allows a centralized fix for the now-false assumption that we never restore blobs in parallel. Also, get rid of dead code in DropLOIfExists: it's been a long time since we had any need to be able to restore to a pre-9.0 server. Discussion: https://postgr.es/m/a9f9376f1c3343a6bb319dce294e20ac@EX13D05UWC001.ant.amazon.com
2024-04-01 22:25:56 +02:00
IssueCommandPerBlob(ArchiveHandle *AH, TocEntry *te,
const char *cmdBegin, const char *cmdEnd)
{
Rearrange pg_dump's handling of large objects for better efficiency. Commit c0d5be5d6 caused pg_dump to create a separate BLOB metadata TOC entry for each large object (blob), but it did not touch the ancient decision to put all the blobs' data into a single "BLOBS" TOC entry. This is bad for a few reasons: for databases with millions of blobs, the TOC becomes unreasonably large, causing performance issues; selective restore of just some blobs is quite impossible; and we cannot parallelize either dump or restore of the blob data, since our architecture for that relies on farming out whole TOC entries to worker processes. To improve matters, let's group multiple blobs into each blob metadata TOC entry, and then make corresponding per-group blob data TOC entries. Selective restore using pg_restore's -l/-L switches is then possible, though only at the group level. (Perhaps we should provide a switch to allow forcing one-blob-per-group for users who need precise selective restore and don't have huge numbers of blobs. This patch doesn't do that, instead just hard-wiring the maximum number of blobs per entry at 1000.) The blobs in a group must all have the same owner, since the TOC entry format only allows one owner to be named. In this implementation we also require them to all share the same ACL (grants); the archive format wouldn't require that, but pg_dump's representation of DumpableObjects does. It seems unlikely that either restriction will be problematic for databases with huge numbers of blobs. The metadata TOC entries now have a "desc" string of "BLOB METADATA", and their "defn" string is just a newline-separated list of blob OIDs. The restore code has to generate creation commands, ALTER OWNER commands, and drop commands (for --clean mode) from that. We would need special-case code for ALTER OWNER and drop in any case, so the alternative of keeping the "defn" as directly executable SQL code for creation wouldn't buy much, and it seems like it'd bloat the archive to little purpose. Since we require the blobs of a metadata group to share the same ACL, we can furthermore store only one copy of that ACL, and then make pg_restore regenerate the appropriate commands for each blob. This saves space in the dump file not only by removing duplicative SQL command strings, but by not needing a separate TOC entry for each blob's ACL. In turn, that reduces client-side memory requirements for handling many blobs. ACL TOC entries that need this special processing are labeled as "ACL"/"LARGE OBJECTS nnn..nnn". If we have a blob with a unique ACL, continue to label it as "ACL"/"LARGE OBJECT nnn". We don't actually have to make such a distinction, but it saves a few cycles during restore for the easy case, and it seems like a good idea to not change the TOC contents unnecessarily. The data TOC entries ("BLOBS") are exactly the same as before, except that now there can be more than one, so we'd better give them identifying tag strings. Also, commit c0d5be5d6 put the new BLOB metadata TOC entries into SECTION_PRE_DATA, which perhaps is defensible in some ways, but it's a rather odd choice considering that we go out of our way to treat blobs as data. Moreover, because parallel restore handles the PRE_DATA section serially, this means we'd only get part of the parallelism speedup we could hope for. Move these entries into SECTION_DATA, letting us parallelize the lo_create calls not just the data loading when there are many blobs. Add dependencies to ensure that we won't try to load data for a blob we've not yet created. As this stands, we still generate a separate TOC entry for any comment or security label attached to a blob. I feel comfortable in believing that comments and security labels on blobs are rare, so this patch should be enough to get most of the useful TOC compression for blobs. We have to bump the archive file format version number, since existing versions of pg_restore wouldn't know they need to do something special for BLOB METADATA, plus they aren't going to work correctly with multiple BLOBS entries or multiple-large-object ACL entries. The directory and tar-file format handlers need some work for multiple BLOBS entries: they used to hard-wire the file name as "blobs.toc", which is replaced here with "blobs_<dumpid>.toc". The 002_pg_dump.pl test script also knows about that and requires minor updates. (I had to drop the test for manually-compressed blobs.toc files with LZ4, because lz4's obtuse command line design requires explicit specification of the output file name which seems impractical here. I don't think we're losing any useful test coverage thereby; that test stanza seems completely duplicative with the gzip and zstd cases anyway.) In passing, centralize management of the lo_buf used to hold data while restoring blobs. The code previously had each format handler create lo_buf, which seems rather pointless given that the format handlers all make it the same way. Moreover, the format handlers never use lo_buf directly, making this setup a failure from a separation-of-concerns standpoint. Let's move the responsibility into pg_backup_archiver.c, which is the only module concerned with lo_buf. The reason to do this in this patch is that it allows a centralized fix for the now-false assumption that we never restore blobs in parallel. Also, get rid of dead code in DropLOIfExists: it's been a long time since we had any need to be able to restore to a pre-9.0 server. Discussion: https://postgr.es/m/a9f9376f1c3343a6bb319dce294e20ac@EX13D05UWC001.ant.amazon.com
2024-04-01 22:25:56 +02:00
/* Make a writable copy of the command string */
char *buf = pg_strdup(te->defn);
char *st;
char *en;
st = buf;
while ((en = strchr(st, '\n')) != NULL)
{
Rearrange pg_dump's handling of large objects for better efficiency. Commit c0d5be5d6 caused pg_dump to create a separate BLOB metadata TOC entry for each large object (blob), but it did not touch the ancient decision to put all the blobs' data into a single "BLOBS" TOC entry. This is bad for a few reasons: for databases with millions of blobs, the TOC becomes unreasonably large, causing performance issues; selective restore of just some blobs is quite impossible; and we cannot parallelize either dump or restore of the blob data, since our architecture for that relies on farming out whole TOC entries to worker processes. To improve matters, let's group multiple blobs into each blob metadata TOC entry, and then make corresponding per-group blob data TOC entries. Selective restore using pg_restore's -l/-L switches is then possible, though only at the group level. (Perhaps we should provide a switch to allow forcing one-blob-per-group for users who need precise selective restore and don't have huge numbers of blobs. This patch doesn't do that, instead just hard-wiring the maximum number of blobs per entry at 1000.) The blobs in a group must all have the same owner, since the TOC entry format only allows one owner to be named. In this implementation we also require them to all share the same ACL (grants); the archive format wouldn't require that, but pg_dump's representation of DumpableObjects does. It seems unlikely that either restriction will be problematic for databases with huge numbers of blobs. The metadata TOC entries now have a "desc" string of "BLOB METADATA", and their "defn" string is just a newline-separated list of blob OIDs. The restore code has to generate creation commands, ALTER OWNER commands, and drop commands (for --clean mode) from that. We would need special-case code for ALTER OWNER and drop in any case, so the alternative of keeping the "defn" as directly executable SQL code for creation wouldn't buy much, and it seems like it'd bloat the archive to little purpose. Since we require the blobs of a metadata group to share the same ACL, we can furthermore store only one copy of that ACL, and then make pg_restore regenerate the appropriate commands for each blob. This saves space in the dump file not only by removing duplicative SQL command strings, but by not needing a separate TOC entry for each blob's ACL. In turn, that reduces client-side memory requirements for handling many blobs. ACL TOC entries that need this special processing are labeled as "ACL"/"LARGE OBJECTS nnn..nnn". If we have a blob with a unique ACL, continue to label it as "ACL"/"LARGE OBJECT nnn". We don't actually have to make such a distinction, but it saves a few cycles during restore for the easy case, and it seems like a good idea to not change the TOC contents unnecessarily. The data TOC entries ("BLOBS") are exactly the same as before, except that now there can be more than one, so we'd better give them identifying tag strings. Also, commit c0d5be5d6 put the new BLOB metadata TOC entries into SECTION_PRE_DATA, which perhaps is defensible in some ways, but it's a rather odd choice considering that we go out of our way to treat blobs as data. Moreover, because parallel restore handles the PRE_DATA section serially, this means we'd only get part of the parallelism speedup we could hope for. Move these entries into SECTION_DATA, letting us parallelize the lo_create calls not just the data loading when there are many blobs. Add dependencies to ensure that we won't try to load data for a blob we've not yet created. As this stands, we still generate a separate TOC entry for any comment or security label attached to a blob. I feel comfortable in believing that comments and security labels on blobs are rare, so this patch should be enough to get most of the useful TOC compression for blobs. We have to bump the archive file format version number, since existing versions of pg_restore wouldn't know they need to do something special for BLOB METADATA, plus they aren't going to work correctly with multiple BLOBS entries or multiple-large-object ACL entries. The directory and tar-file format handlers need some work for multiple BLOBS entries: they used to hard-wire the file name as "blobs.toc", which is replaced here with "blobs_<dumpid>.toc". The 002_pg_dump.pl test script also knows about that and requires minor updates. (I had to drop the test for manually-compressed blobs.toc files with LZ4, because lz4's obtuse command line design requires explicit specification of the output file name which seems impractical here. I don't think we're losing any useful test coverage thereby; that test stanza seems completely duplicative with the gzip and zstd cases anyway.) In passing, centralize management of the lo_buf used to hold data while restoring blobs. The code previously had each format handler create lo_buf, which seems rather pointless given that the format handlers all make it the same way. Moreover, the format handlers never use lo_buf directly, making this setup a failure from a separation-of-concerns standpoint. Let's move the responsibility into pg_backup_archiver.c, which is the only module concerned with lo_buf. The reason to do this in this patch is that it allows a centralized fix for the now-false assumption that we never restore blobs in parallel. Also, get rid of dead code in DropLOIfExists: it's been a long time since we had any need to be able to restore to a pre-9.0 server. Discussion: https://postgr.es/m/a9f9376f1c3343a6bb319dce294e20ac@EX13D05UWC001.ant.amazon.com
2024-04-01 22:25:56 +02:00
*en++ = '\0';
ahprintf(AH, "%s%s%s;\n", cmdBegin, st, cmdEnd);
st = en;
}
Rearrange pg_dump's handling of large objects for better efficiency. Commit c0d5be5d6 caused pg_dump to create a separate BLOB metadata TOC entry for each large object (blob), but it did not touch the ancient decision to put all the blobs' data into a single "BLOBS" TOC entry. This is bad for a few reasons: for databases with millions of blobs, the TOC becomes unreasonably large, causing performance issues; selective restore of just some blobs is quite impossible; and we cannot parallelize either dump or restore of the blob data, since our architecture for that relies on farming out whole TOC entries to worker processes. To improve matters, let's group multiple blobs into each blob metadata TOC entry, and then make corresponding per-group blob data TOC entries. Selective restore using pg_restore's -l/-L switches is then possible, though only at the group level. (Perhaps we should provide a switch to allow forcing one-blob-per-group for users who need precise selective restore and don't have huge numbers of blobs. This patch doesn't do that, instead just hard-wiring the maximum number of blobs per entry at 1000.) The blobs in a group must all have the same owner, since the TOC entry format only allows one owner to be named. In this implementation we also require them to all share the same ACL (grants); the archive format wouldn't require that, but pg_dump's representation of DumpableObjects does. It seems unlikely that either restriction will be problematic for databases with huge numbers of blobs. The metadata TOC entries now have a "desc" string of "BLOB METADATA", and their "defn" string is just a newline-separated list of blob OIDs. The restore code has to generate creation commands, ALTER OWNER commands, and drop commands (for --clean mode) from that. We would need special-case code for ALTER OWNER and drop in any case, so the alternative of keeping the "defn" as directly executable SQL code for creation wouldn't buy much, and it seems like it'd bloat the archive to little purpose. Since we require the blobs of a metadata group to share the same ACL, we can furthermore store only one copy of that ACL, and then make pg_restore regenerate the appropriate commands for each blob. This saves space in the dump file not only by removing duplicative SQL command strings, but by not needing a separate TOC entry for each blob's ACL. In turn, that reduces client-side memory requirements for handling many blobs. ACL TOC entries that need this special processing are labeled as "ACL"/"LARGE OBJECTS nnn..nnn". If we have a blob with a unique ACL, continue to label it as "ACL"/"LARGE OBJECT nnn". We don't actually have to make such a distinction, but it saves a few cycles during restore for the easy case, and it seems like a good idea to not change the TOC contents unnecessarily. The data TOC entries ("BLOBS") are exactly the same as before, except that now there can be more than one, so we'd better give them identifying tag strings. Also, commit c0d5be5d6 put the new BLOB metadata TOC entries into SECTION_PRE_DATA, which perhaps is defensible in some ways, but it's a rather odd choice considering that we go out of our way to treat blobs as data. Moreover, because parallel restore handles the PRE_DATA section serially, this means we'd only get part of the parallelism speedup we could hope for. Move these entries into SECTION_DATA, letting us parallelize the lo_create calls not just the data loading when there are many blobs. Add dependencies to ensure that we won't try to load data for a blob we've not yet created. As this stands, we still generate a separate TOC entry for any comment or security label attached to a blob. I feel comfortable in believing that comments and security labels on blobs are rare, so this patch should be enough to get most of the useful TOC compression for blobs. We have to bump the archive file format version number, since existing versions of pg_restore wouldn't know they need to do something special for BLOB METADATA, plus they aren't going to work correctly with multiple BLOBS entries or multiple-large-object ACL entries. The directory and tar-file format handlers need some work for multiple BLOBS entries: they used to hard-wire the file name as "blobs.toc", which is replaced here with "blobs_<dumpid>.toc". The 002_pg_dump.pl test script also knows about that and requires minor updates. (I had to drop the test for manually-compressed blobs.toc files with LZ4, because lz4's obtuse command line design requires explicit specification of the output file name which seems impractical here. I don't think we're losing any useful test coverage thereby; that test stanza seems completely duplicative with the gzip and zstd cases anyway.) In passing, centralize management of the lo_buf used to hold data while restoring blobs. The code previously had each format handler create lo_buf, which seems rather pointless given that the format handlers all make it the same way. Moreover, the format handlers never use lo_buf directly, making this setup a failure from a separation-of-concerns standpoint. Let's move the responsibility into pg_backup_archiver.c, which is the only module concerned with lo_buf. The reason to do this in this patch is that it allows a centralized fix for the now-false assumption that we never restore blobs in parallel. Also, get rid of dead code in DropLOIfExists: it's been a long time since we had any need to be able to restore to a pre-9.0 server. Discussion: https://postgr.es/m/a9f9376f1c3343a6bb319dce294e20ac@EX13D05UWC001.ant.amazon.com
2024-04-01 22:25:56 +02:00
ahprintf(AH, "\n");
pg_free(buf);
}
/*
* Process a "LARGE OBJECTS" ACL TocEntry.
*
* To save space in the dump file, the TocEntry contains only one copy
* of the required GRANT/REVOKE commands, written to apply to the first
* blob in the group (although we do not depend on that detail here).
* We must expand the text to generate commands for all the blobs listed
* in the associated BLOB METADATA entry.
*/
void
IssueACLPerBlob(ArchiveHandle *AH, TocEntry *te)
{
TocEntry *blobte = getTocEntryByDumpId(AH, te->dependencies[0]);
char *buf;
char *st;
char *st2;
char *en;
bool inquotes;
if (!blobte)
pg_fatal("could not find entry for ID %d", te->dependencies[0]);
Assert(strcmp(blobte->desc, "BLOB METADATA") == 0);
/* Make a writable copy of the ACL commands string */
buf = pg_strdup(te->defn);
/*
* We have to parse out the commands sufficiently to locate the blob OIDs
* and find the command-ending semicolons. The commands should not
* contain anything hard to parse except for double-quoted role names,
* which are easy to ignore. Once we've split apart the first and second
* halves of a command, apply IssueCommandPerBlob. (This means the
* updates on the blobs are interleaved if there's multiple commands, but
* that should cause no trouble.)
*/
inquotes = false;
st = en = buf;
st2 = NULL;
while (*en)
{
Rearrange pg_dump's handling of large objects for better efficiency. Commit c0d5be5d6 caused pg_dump to create a separate BLOB metadata TOC entry for each large object (blob), but it did not touch the ancient decision to put all the blobs' data into a single "BLOBS" TOC entry. This is bad for a few reasons: for databases with millions of blobs, the TOC becomes unreasonably large, causing performance issues; selective restore of just some blobs is quite impossible; and we cannot parallelize either dump or restore of the blob data, since our architecture for that relies on farming out whole TOC entries to worker processes. To improve matters, let's group multiple blobs into each blob metadata TOC entry, and then make corresponding per-group blob data TOC entries. Selective restore using pg_restore's -l/-L switches is then possible, though only at the group level. (Perhaps we should provide a switch to allow forcing one-blob-per-group for users who need precise selective restore and don't have huge numbers of blobs. This patch doesn't do that, instead just hard-wiring the maximum number of blobs per entry at 1000.) The blobs in a group must all have the same owner, since the TOC entry format only allows one owner to be named. In this implementation we also require them to all share the same ACL (grants); the archive format wouldn't require that, but pg_dump's representation of DumpableObjects does. It seems unlikely that either restriction will be problematic for databases with huge numbers of blobs. The metadata TOC entries now have a "desc" string of "BLOB METADATA", and their "defn" string is just a newline-separated list of blob OIDs. The restore code has to generate creation commands, ALTER OWNER commands, and drop commands (for --clean mode) from that. We would need special-case code for ALTER OWNER and drop in any case, so the alternative of keeping the "defn" as directly executable SQL code for creation wouldn't buy much, and it seems like it'd bloat the archive to little purpose. Since we require the blobs of a metadata group to share the same ACL, we can furthermore store only one copy of that ACL, and then make pg_restore regenerate the appropriate commands for each blob. This saves space in the dump file not only by removing duplicative SQL command strings, but by not needing a separate TOC entry for each blob's ACL. In turn, that reduces client-side memory requirements for handling many blobs. ACL TOC entries that need this special processing are labeled as "ACL"/"LARGE OBJECTS nnn..nnn". If we have a blob with a unique ACL, continue to label it as "ACL"/"LARGE OBJECT nnn". We don't actually have to make such a distinction, but it saves a few cycles during restore for the easy case, and it seems like a good idea to not change the TOC contents unnecessarily. The data TOC entries ("BLOBS") are exactly the same as before, except that now there can be more than one, so we'd better give them identifying tag strings. Also, commit c0d5be5d6 put the new BLOB metadata TOC entries into SECTION_PRE_DATA, which perhaps is defensible in some ways, but it's a rather odd choice considering that we go out of our way to treat blobs as data. Moreover, because parallel restore handles the PRE_DATA section serially, this means we'd only get part of the parallelism speedup we could hope for. Move these entries into SECTION_DATA, letting us parallelize the lo_create calls not just the data loading when there are many blobs. Add dependencies to ensure that we won't try to load data for a blob we've not yet created. As this stands, we still generate a separate TOC entry for any comment or security label attached to a blob. I feel comfortable in believing that comments and security labels on blobs are rare, so this patch should be enough to get most of the useful TOC compression for blobs. We have to bump the archive file format version number, since existing versions of pg_restore wouldn't know they need to do something special for BLOB METADATA, plus they aren't going to work correctly with multiple BLOBS entries or multiple-large-object ACL entries. The directory and tar-file format handlers need some work for multiple BLOBS entries: they used to hard-wire the file name as "blobs.toc", which is replaced here with "blobs_<dumpid>.toc". The 002_pg_dump.pl test script also knows about that and requires minor updates. (I had to drop the test for manually-compressed blobs.toc files with LZ4, because lz4's obtuse command line design requires explicit specification of the output file name which seems impractical here. I don't think we're losing any useful test coverage thereby; that test stanza seems completely duplicative with the gzip and zstd cases anyway.) In passing, centralize management of the lo_buf used to hold data while restoring blobs. The code previously had each format handler create lo_buf, which seems rather pointless given that the format handlers all make it the same way. Moreover, the format handlers never use lo_buf directly, making this setup a failure from a separation-of-concerns standpoint. Let's move the responsibility into pg_backup_archiver.c, which is the only module concerned with lo_buf. The reason to do this in this patch is that it allows a centralized fix for the now-false assumption that we never restore blobs in parallel. Also, get rid of dead code in DropLOIfExists: it's been a long time since we had any need to be able to restore to a pre-9.0 server. Discussion: https://postgr.es/m/a9f9376f1c3343a6bb319dce294e20ac@EX13D05UWC001.ant.amazon.com
2024-04-01 22:25:56 +02:00
/* Ignore double-quoted material */
if (*en == '"')
inquotes = !inquotes;
if (inquotes)
{
en++;
continue;
}
/* If we found "LARGE OBJECT", that's the end of the first half */
if (strncmp(en, "LARGE OBJECT ", 13) == 0)
{
/* Terminate the first-half string */
en += 13;
Assert(isdigit((unsigned char) *en));
*en++ = '\0';
/* Skip the rest of the blob OID */
while (isdigit((unsigned char) *en))
en++;
/* Second half starts here */
Assert(st2 == NULL);
st2 = en;
}
/* If we found semicolon, that's the end of the second half */
else if (*en == ';')
{
/* Terminate the second-half string */
*en++ = '\0';
Assert(st2 != NULL);
/* Issue this command for each blob */
IssueCommandPerBlob(AH, blobte, st, st2);
/* For neatness, skip whitespace before the next command */
while (isspace((unsigned char) *en))
en++;
/* Reset for new command */
st = en;
st2 = NULL;
}
else
en++;
}
Rearrange pg_dump's handling of large objects for better efficiency. Commit c0d5be5d6 caused pg_dump to create a separate BLOB metadata TOC entry for each large object (blob), but it did not touch the ancient decision to put all the blobs' data into a single "BLOBS" TOC entry. This is bad for a few reasons: for databases with millions of blobs, the TOC becomes unreasonably large, causing performance issues; selective restore of just some blobs is quite impossible; and we cannot parallelize either dump or restore of the blob data, since our architecture for that relies on farming out whole TOC entries to worker processes. To improve matters, let's group multiple blobs into each blob metadata TOC entry, and then make corresponding per-group blob data TOC entries. Selective restore using pg_restore's -l/-L switches is then possible, though only at the group level. (Perhaps we should provide a switch to allow forcing one-blob-per-group for users who need precise selective restore and don't have huge numbers of blobs. This patch doesn't do that, instead just hard-wiring the maximum number of blobs per entry at 1000.) The blobs in a group must all have the same owner, since the TOC entry format only allows one owner to be named. In this implementation we also require them to all share the same ACL (grants); the archive format wouldn't require that, but pg_dump's representation of DumpableObjects does. It seems unlikely that either restriction will be problematic for databases with huge numbers of blobs. The metadata TOC entries now have a "desc" string of "BLOB METADATA", and their "defn" string is just a newline-separated list of blob OIDs. The restore code has to generate creation commands, ALTER OWNER commands, and drop commands (for --clean mode) from that. We would need special-case code for ALTER OWNER and drop in any case, so the alternative of keeping the "defn" as directly executable SQL code for creation wouldn't buy much, and it seems like it'd bloat the archive to little purpose. Since we require the blobs of a metadata group to share the same ACL, we can furthermore store only one copy of that ACL, and then make pg_restore regenerate the appropriate commands for each blob. This saves space in the dump file not only by removing duplicative SQL command strings, but by not needing a separate TOC entry for each blob's ACL. In turn, that reduces client-side memory requirements for handling many blobs. ACL TOC entries that need this special processing are labeled as "ACL"/"LARGE OBJECTS nnn..nnn". If we have a blob with a unique ACL, continue to label it as "ACL"/"LARGE OBJECT nnn". We don't actually have to make such a distinction, but it saves a few cycles during restore for the easy case, and it seems like a good idea to not change the TOC contents unnecessarily. The data TOC entries ("BLOBS") are exactly the same as before, except that now there can be more than one, so we'd better give them identifying tag strings. Also, commit c0d5be5d6 put the new BLOB metadata TOC entries into SECTION_PRE_DATA, which perhaps is defensible in some ways, but it's a rather odd choice considering that we go out of our way to treat blobs as data. Moreover, because parallel restore handles the PRE_DATA section serially, this means we'd only get part of the parallelism speedup we could hope for. Move these entries into SECTION_DATA, letting us parallelize the lo_create calls not just the data loading when there are many blobs. Add dependencies to ensure that we won't try to load data for a blob we've not yet created. As this stands, we still generate a separate TOC entry for any comment or security label attached to a blob. I feel comfortable in believing that comments and security labels on blobs are rare, so this patch should be enough to get most of the useful TOC compression for blobs. We have to bump the archive file format version number, since existing versions of pg_restore wouldn't know they need to do something special for BLOB METADATA, plus they aren't going to work correctly with multiple BLOBS entries or multiple-large-object ACL entries. The directory and tar-file format handlers need some work for multiple BLOBS entries: they used to hard-wire the file name as "blobs.toc", which is replaced here with "blobs_<dumpid>.toc". The 002_pg_dump.pl test script also knows about that and requires minor updates. (I had to drop the test for manually-compressed blobs.toc files with LZ4, because lz4's obtuse command line design requires explicit specification of the output file name which seems impractical here. I don't think we're losing any useful test coverage thereby; that test stanza seems completely duplicative with the gzip and zstd cases anyway.) In passing, centralize management of the lo_buf used to hold data while restoring blobs. The code previously had each format handler create lo_buf, which seems rather pointless given that the format handlers all make it the same way. Moreover, the format handlers never use lo_buf directly, making this setup a failure from a separation-of-concerns standpoint. Let's move the responsibility into pg_backup_archiver.c, which is the only module concerned with lo_buf. The reason to do this in this patch is that it allows a centralized fix for the now-false assumption that we never restore blobs in parallel. Also, get rid of dead code in DropLOIfExists: it's been a long time since we had any need to be able to restore to a pre-9.0 server. Discussion: https://postgr.es/m/a9f9376f1c3343a6bb319dce294e20ac@EX13D05UWC001.ant.amazon.com
2024-04-01 22:25:56 +02:00
pg_free(buf);
}
void
DropLOIfExists(ArchiveHandle *AH, Oid oid)
{
ahprintf(AH,
"SELECT pg_catalog.lo_unlink(oid) "
"FROM pg_catalog.pg_largeobject_metadata "
"WHERE oid = '%u';\n",
oid);
}