During Hot Standby we need to check for buffer pin deadlocks when the
Startup process begins to wait, in case it never wakes up again. We
previously made the deadlock check immediately on the basis it was
cheap, though clearer thinking and prima facie evidence shows that
was too simple. Refactor existing code to make it easy to add in
deferral of deadlock check until deadlock_timeout allowing a good
reduction in deadlock checks since far few buffer pins are held for
that duration. It's worth doing anyway, though major goal is to
prevent further reports of context switching with high numbers of
users on occasional tests.
requests for client certs. This lets a client with a keystore select the
appropriate client certificate to send. In particular, this is necessary
to get Java clients to work in all but the most trivial configurations.
Per discussion of bug #5468.
Craig Ringer
1. If we receive a fast shutdown request while in the PM_STARTUP state,
process it just as we would in PM_RECOVERY, PM_HOT_STANDBY, or PM_RUN.
Without this change, an early fast shutdown followed by Hot Standby causes
the database to get stuck in a state where a shutdown is pending (so no new
connections are allowed) but the shutdown request is never processed unless
we end Hot Standby and enter normal running.
2. Avoid removing the backup label file when a smart or fast shutdown occurs
during recovery. It makes sense to do this once we've reached normal running,
since we must be taking a backup which now won't be valid. But during
recovery we must be recovering from a previously taken backup, and any backup
label file is needed to restart recovery from the right place.
Fujii Masao and Robert Haas
If the original IN operator is cross-type, for example int8 = int4,
we need to use int4 < int4 to sort the inner data and int4 = int4
to unique-ify it. We got the first part of that right, but tried to
use the original IN operator for the equality checks. Per bug #5472
from Vlad Romascanu.
Backpatch to 8.4, where the bug was introduced by the patch that unified
SortClause and GroupClause. I was able to take out a whole lot of on-the-fly
calls of get_equality_op_for_ordering_op(), but failed to realize that
I needed to put one back in right here :-(
so simply leads to data waiting in wal_buffers which then causes
later commits to potentially do emergency writes and for all forms
of replication to be potentially delayed without need or benefit.
Issue pointed out exactly by Fujii Masao, following bug report
by Robert Haas on a separate though related topic.
of requirements and documentation on LogStandbySnapshot(). Fixes
two minor bugs reported by Tom Lane that would lead to an incorrect
snapshot after transaction wraparound. Also fix two other problems
discovered that would give incorrect snapshots in certain cases.
ProcArrayApplyRecoveryInfo() substantially rewritten. Some minor
refactoring of xact_redo_apply() and ExpireTreeKnownAssignedTransactionIds().
archive_command) as soon as possible, namely just before issuing a new call
of archive_command, even when there is a backlog of files to be archived.
The original coding would only absorb new settings after clearing the backlog
and returning to the outer loop. Per discussion.
Back-patch to 8.3. The logic in prior versions is a bit different and it
doesn't seem worth taking any risks of breaking it.
Now validators work properly even when the settings contain
parameters that affect behavior of the function, like search_path.
Reported by Erwin Brandstetter.
MIN or MAX, we must take care to insert the added qual in a legal place among
the existing indexquals, if any. The btree index AM requires the quals to
appear in index-column order. We didn't have to worry about this before
because "target IS NOT NULL" was just treated as a plain scan filter condition;
but as of 9.0 it can be an index qual and then it has to follow the rule.
Per report from Ian Barwick.
My initial impression that glibc was measuring the precision in characters
(which is what the Linux man page says it does) was incorrect. It does take
the precision to be in bytes, but it also tries to truncate the string at a
character boundary. The bottom line remains the same: it will mess up
if the string is not in the encoding it expects, so we need to avoid %.*s
anytime there's a significant risk of that. Previous code changes are still
good, but adjust the comments to reflect this knowledge. Per research by
Hernan Gonzalez.
Depending on which spec you read, field widths and precisions in %s may be
counted either in bytes or characters. Our code was assuming bytes, which
is wrong at least for glibc's implementation, and in any case libc might
have a different idea of the prevailing encoding than we do. Hence, for
portable results we must avoid using anything more complex than just "%s"
unless the string to be printed is known to be all-ASCII.
This patch fixes the cases I could find, including the psql formatting
failure reported by Hernan Gonzalez. In HEAD only, I also added comments
to some places where it appears safe to continue using "%.*s".
minRecoveryPoint in control file when replaying a parameter change record,
to ensure that we don't allow hot standby on WAL generated without
wal_level='hot_standby' after a standby restart.
field of the WAL record. The previous coding always wrote to the main fork,
resulting in data corruption if the page was meant to go into a non-default
fork.
At present, the only operation that can produce such WAL records is
ALTER TABLE/INDEX SET TABLESPACE when executed with archive_mode = on.
Data corruption would be observed on standby slaves, and could occur on the
master as well if a database crash and recovery occurred after committing
the ALTER and before the next checkpoint. Per report from Gordon Shannon.
Back-patch to 8.4; the problem doesn't exist in earlier branches because
we didn't have a concept of multiple relation forks then.
MaxStandbyDelay. Use the GUC units mechanism for the value, and choose more
appropriate timestamp functions for performing tests with it. Make the
ps_activity manipulation in ResolveRecoveryConflictWithVirtualXIDs have
behavior similar to ps_activity code elsewhere, notably not updating the
display when update_process_title is off and not truncating the display
contents at an arbitrarily-chosen length. Improve the docs to be explicit
about what MaxStandbyDelay actually measures, viz the difference between
primary and standby servers' clocks, and the possible hazards if their clocks
aren't in sync.
returns EINVAL for an existing shared memory segment. Although it's not
terribly sensible, that behavior does meet the POSIX spec because EINVAL
is the appropriate error code when the existing segment is smaller than the
requested size, and the spec explicitly disclaims any particular ordering of
error checks. Moreover, it does in fact happen on OS X and probably other
BSD-derived kernels. (We were able to talk NetBSD into changing their code,
but purging that behavior from the wild completely seems unlikely to happen.)
We need to distinguish collision with a pre-existing segment from invalid size
request in order to behave sensibly, so it's worth some extra code here to get
it right. Per report from Gavin Kistner and subsequent investigation.
Back-patch to all supported versions, since any of them could get used
with a kernel having the debatable behavior.
to perform a backup without archive_mode being enabled. This gives up some
user-error protection in order to improve usefulness for streaming-replication
scenarios. Per discussion.
confusion with streaming-replication settings. Also, change its default
value to "off", because of concern about executing new and poorly-tested
code during ordinary non-replicating operation. Per discussion.
In passing do some minor editing of related documentation.
contrib/intarray is loaded. Per bug #5417 from Kenaniah Cerny.
Not forcing initdb since backend doesn't directly depend on this,
and few people have run into it.
rather than returning NULL for some-but-not-all failures as they used to.
Remove now-redundant tests for NULL from call sites.
We had to do something about this because many call sites were failing to
check for NULL; and changing it like this seems a lot more useful and
mistake-proof than adding checks to the call sites without them.
archival or hot standby should be WAL-logged, instead of deducing that from
other options like archive_mode. This replaces recovery_connections GUC in
the primary, where it now has no effect, but it's still used in the standby
to enable/disable hot standby.
Remove the WAL-logging of "unlogged operations", like creating an index
without WAL-logging and fsyncing it at the end. Instead, we keep a copy of
the wal_mode setting and the settings that affect how much shared memory a
hot standby server needs to track master transactions (max_connections,
max_prepared_xacts, max_locks_per_xact) in pg_control. Whenever the settings
change, at server restart, write a WAL record noting the new settings and
update pg_control. This allows us to notice the change in those settings in
the standby at the right moment, they used to be included in checkpoint
records, but that meant that a changed value was not reflected in the
standby until the first checkpoint after the change.
Bump PG_CONTROL_VERSION and XLOG_PAGE_MAGIC. Whack XLOG_PAGE_MAGIC back to
the sequence it used to follow, before hot standby and subsequent patches
changed it to 0x9003.
to RFC 3986. In particular, these characters now terminate the path part
of a URL: '"', '<', '>', '\', '^', '`', '{', '|', '}'. The previous behavior
was inconsistent and depended on whether a "?" was present in the path.
Per gripe from Donald Fraser and spec research by Kevin Grittner.
This is a pre-existing bug, but not back-patching since the risks of
breaking existing applications seem to outweigh the benefits.
and be more tense about the locking requirements for it, to improve performance
in Hot Standby mode. In passing fix a few bugs and improve a number of
comments in the existing HS code.
Simon Riggs, with some editorialization by Tom
in WAL recovery when it sees the shutdown checkpoint record. It's more
user-friendly to find out about it at that point than at the end of
recovery, and you're not left wondering why your hot standby server never
opens up for read-only connections.
Normal superuser processes are allowed to connect even when the database
system is shutting down, or when fewer than superuser_reserved_connection
slots remain. This is intended to make sure an administrator can log in
and troubleshoot, so don't extend these same courtesies to users connecting
for replication.
of parameters. Fix bug report by Robert Haas that error message and
hint was incorrect if wrong mode parameters specified on master.
Internal changes only. Proposals for parameter simplification on
master/primary still under way.
come from the realistion that HEAP2_CLEAN records don't
always remove user visible data, so conflict processing for
them can be skipped. Confirm validity using Assert checks,
clarify circumstances under which we log heap_cleanup_info
records. Tuning arises from bug fixing of earlier safety
check failures.
from lc_ctype, that could happen on Windows. We need to change lc_ctype
together with lc_monetary or lc_numeric, and convert strings in lconv
from lc_ctype encoding to the database encoding.
The bug reported by Mikko, original patch by Hiroshi Inoue,
with changes by Bruce and me.
than during define_custom_variable(). This entails rejecting an ALTER
command if the target variable doesn't have a known (non-placeholder)
definition, unless the calling user is superuser. When the variable *is*
known, we can correctly apply the rule that only superusers can issue ALTER
for SUSET parameters. This allows define_custom_variable to apply ALTER's
values for SUSET parameters at module load time, secure in the knowledge
that only a superuser could have set the ALTER value. This change fixes a
longstanding gotcha in the usage of SUSET-level custom parameters; which
is a good thing to fix now that plpgsql defines such a parameter.
There is no other purpose for this message type than to report
the latestRemovedXid of removed tuples, prior to index scans.
Removes overlooked path for sending invalid latestRemovedXid.
Fixes buildfarm failure on centaur.
to handling of btree delete records mean that all snapshot
conflicts on standby now have a valid, useful latestRemovedXid.
Our earlier approach using LW_EXCLUSIVE was useful when we didnt
always have a valid value, though is no longer useful or necessary.
Asserts added to code path to prove and ensure this is the case.
This will reduce contention and improve performance of larger Hot
Standby servers.
vacuum_log_cleanup_info() now generates log records with a valid
latestRemovedXid set in all cases. Also be careful not to zero the
value when we do a round of vacuuming part-way through lazy_scan_heap().
Incidentally, this reduces frequency of conflicts in Hot Standby.
with database = replication. The previous coding would allow them to match
ordinary records too, but that seems like a recipe for security breaches.
Improve the messages associated with no-such-pg_hba.conf entry to report
replication connections as such, since that's now a critical aspect of
whether the connection matches. Make some cursory improvements in the related
documentation, too.
database to connect to. This is necessary for the walsender code to work
properly (it was previously using an untenable assumption that template1 would
always be available to connect to). This also gets rid of a small security
shortcoming that was introduced in the original patch to eliminate the flat
authentication files: before, you could find out whether or not the requested
database existed even if you couldn't pass the authentication checks.
The changes needed to support this are mainly just to treat pg_authid and
pg_auth_members as nailed relations, so that we can read them without having
to be able to locate real pg_class entries for them. This mechanism was
already debugged for pg_database, but we hadn't recognized the value of
applying it to those catalogs too.
Since the current code doesn't have support for accessing toast tables before
we've brought up all of the relcache, remove pg_authid's toast table to ensure
that no one can store an out-of-line toasted value of rolpassword. The case
seems quite unlikely to occur in practice, and was effectively unsupported
anyway in the old "flatfiles" implementation.
Update genbki.pl to actually implement the same rules as bootstrap.c does for
not-nullability of catalog columns. The previous coding was a bit cheesy but
worked all right for the previous set of bootstrap catalogs. It does not work
for pg_authid, where rolvaliduntil needs to be nullable.
Initdb forced due to minor catalog changes (mainly the toast table removal).
Also, make the name of the GUC and the name of the backing variable match.
Alnong the way, clean up a couple of slight typographical errors in the
related docs.
those process types that go through InitPostgres; in particular, bootstrap
and standalone-backend cases. This ensures that we have set up a PGPROC
and done some other basic initialization steps (corresponding to the
if (IsUnderPostmaster) block in AuxiliaryProcessMain) before we attempt to
run WAL recovery in a standalone backend. As was discovered last September,
this is necessary for some corner-case code paths during WAL recovery,
particularly end-of-WAL cleanup.
Moving the bootstrap case here too is not necessary for correctness, but it
seems like a good idea since it reduces the number of distinct code paths.
libpq to send queries, making the waiting for responses interruptible on
platforms where PQexec() can't normally be interrupted by signals, such
as win32.
Fujii Masao and Magnus Hagander
The logic for determining whether to materialize has been significantly
overhauled for 9.0. In case there should be any doubt about whether
materialization is a win in any particular case, this should provide a
convenient way of seeing what happens without it; but even with enable_material
turned off, we still materialize in cases where it is required for
correctness.
Thanks to Tom Lane for the review.
Now doesn't report it is waiting until it actually is waiting,
plus message doesn't appear until at least 5 seconds wait, so
we avoid reporting the wait before we've given the archiver
a reasonable time to wake up and archive the file we just
created earlier in the function.
Also add new unconditional message to confirm safe completion.
Now a normal, healthy execution does not report waiting at
all, just safe completion.
through normal backends. Makes code clearer also, since we
avoid various Assert()s. Performance of snapshots taken
during recovery no longer depends upon number of read-only
backends.
reload and rotation signals, and a helper thread reads messages from the
pipe and writes them to the log file. However, server code isn't generally
thread-safe, so if both try to do e.g palloc()/pfree() at the same time,
bad things will happen. To fix that, use a critical section (which is like
a mutex) to enforce that only one the threads are active at a time.
relcache reload works. In the patched code, a relcache entry in process of
being rebuilt doesn't get unhooked from the relcache hash table; which means
that if a cache flush occurs due to sinval queue overrun while we're
rebuilding it, the entry could get blown away by RelationCacheInvalidate,
resulting in crash or misbehavior. Fix by ensuring that an entry being
rebuilt has positive refcount, so it won't be seen as a target for removal
if a cache flush occurs. (This will mean that the entry gets rebuilt twice
in such a scenario, but that's okay.) It appears that the problem can only
arise within a transaction that has previously reassigned the relfilenode of
a pre-existing table, via TRUNCATE or a similar operation. Per bug #5412
from Rusty Conover.
Back-patch to 8.2, same as the patch that introduced the problem.
I think that the failure can't actually occur in 8.2, since it lacks the
rd_newRelfilenodeSubid optimization, but let's make it work like the later
branches anyway.
Patch by Heikki, slightly editorialized on by me.
after actually removing one, so that if we can't remove segments because
WAL archiving is lagging behind, we don't unnecessarily forbid streaming
the old not-yet-archived segments that are still perfectly valid. Per
suggestion from Fujii Masao.
doesn't take into account how far the WAL senders are. This way a hung
WAL sender doesn't prevent old WAL segments from being recycled/removed
in the primary, ultimately causing the disk to fill up. Instead add
standby_keep_segments setting to control how many old WAL segments are
kept in the primary. This also makes it more reliable to use streaming
replication without WAL archiving, assuming that you set
standby_keep_segments high enough.
At present, killing the startup process does not release any locks it holds,
so we must wait to stop the startup and walreceiver processes until all
read-only backends have exited. Without this patch, the startup and
walreceiver processes never exit, so the server gets permanently stuck in
a half-shutdown state.
Fujii Masao, with review, docs, and comment adjustments by me.
rather than only sort-of working as the previous attempt had left it.
Clean up some unnecessary differences between the way these were coded and
the way the YYYY case was coded. Update the regression test cases that
proved that it wasn't working.
recovery. We might want to relax this in the future, but ThisTimeLineID
isn't currently correct in backends during recovery, so the filename
returned was wrong.
is changed to match the hard-wired default. This avoids accumulating useless
catalog entries, and also provides a path for dropping the owning role without
using DROP OWNED BY. Per yesterday's complaint from Jaime Casanova, the
need to use DROP OWNED BY for that is less than obvious, so providing this
alternative method might save some user frustration.
be added during GRANT and can only be removed during REVOKE; and fix its
callers to not lie to it about the existing set of dependencies when
instantiating a formerly-default ACL. The previous coding accidentally failed
to malfunction so long as default ACLs contain only references to the object's
owning role, because that role is ignored by updateAclDependencies. However
this is obviously pretty fragile, as well as being an undocumented assumption.
The new coding is a few lines longer but IMO much clearer.
This allows us to see what mode the server is in before it starts to
perform actions that can block or hang. Otherwise server messages
may not appear until after messages that say FATAL the database
server is starting up.
Windows, thanks to a feature in CRT called Parameter Validation.
Backpatch to 8.2, which is the oldest version supported on Windows. In
8.2 and 8.3 also backpatch the earlier change to use DEVNULL instead of
NULL_DEV #define for a /dev/null-like device. NULL_DEV was hard-coded to
"/dev/null" regardless of platform, which didn't work on Windows, while
DEVNULL works on all platforms. Restarting syslogger didn't work on
Windows on versions 8.3 and below because of that.
The error message now makes explicit reference to the GUC that must be changed
to fix the problem, using wording suggested by Tom Lane. Along the way,
rename the GUC from MaxWalSenders to max_wal_senders for consistency and
grep-ability.
constraint exclusion on an inheritance set that is the target of an UPDATE
or DELETE query. Per gripe from Marc Cousin. Back-patch to 8.4 where
the feature was introduced.
pg_xlog directory. This is essential for replaying WAL records that
were streamed from the master, after a standby server restart.
If a corrupt record is seen in a file restored from the archive or
streamed from the master, log it as a WARNING and keep retrying. If the
corruption is permanent, and not just a glitch in the whatever copies the
files to the archive or a network error not caught by CRC checks in TCP
for example, we will keep retrying and logging the WARNING indefinitely.
But that's better than shutting down completely, the standby is still
useful for running read-only queries. In PITR the recovery ends at such a
corrupt record, which is a bit questionable, but that's the behavior we
had in previous releases and we don't feel like chaning it now. It does
make sense for tools like pg_standby.
example to 'on or 'off' rather than 'true' or 'false', as shown
in docs. Add restartpoint_command. Add section header for recovery
target parameters, matching docs.
WAL record for btree delete contains a list of tids, even when backup
blocks are present. We follow the tids to their heap tuples, taking
care to follow LP_REDIRECT tuples. We ignore LP_DEAD tuples on the
understanding that they will always have xmin/xmax earlier than any
LP_NORMAL tuples referred to by killed index tuples. Iff all tuples
are LP_DEAD we return InvalidTransactionId. The heap relfilenode is
added to the WAL record, requiring API changes to pass down the heap
Relation. XLOG_PAGE_MAGIC updated.
by a superuser -- "ALTER USER f RESET setting" already disallows removing such a
setting.
Apply the same treatment to ALTER DATABASE d RESET ALL when run by a database
owner that's not superuser.
doing nothing, caused by naptime specified in milliseconds yet units of
pg_usleep() parameter is microseconds. Correctly specifying units
reduces call frequency by 1000. Reduction in CPU consumption verified.
so that we won't try to attach any context printouts to messages that get
emitted while exiting. Per report from Dennis Koegel, the context functions
won't necessarily work after we've started shutting down the backend, and it
seems possible that debug_query_string could be pointing at freed storage
as well. The context information doesn't seem particularly relevant to
such messages anyway, so there's little lost by suppressing it.
Back-patch to all supported branches. I can only demonstrate a crash with
log_disconnections messages back to 8.1, but the risk seems real in 8.0 and
before anyway.
catalog entries via SearchSysCache and related operations. Although, at the
time that these callbacks are called by elog.c, we have not officially aborted
the current transaction, it still seems rather risky to initiate any new
catalog fetches. In all these cases the needed information is readily
available in the caller and so it's just a matter of a bit of extra notation
to pass it to the callback.
Per crash report from Dennis Koegel. I've concluded that the real fix for
his problem is to clear the error context stack at entry to proc_exit, but
it still seems like a good idea to make the callbacks a bit less fragile
for other cases.
Backpatch to 8.4. We could go further back, but the patch doesn't apply
cleanly. In the absence of proof that this fixes something and isn't just
paranoia, I'm not going to expend the effort.
was broken for a replication connection and no messages were
displayed on either standby or primary, at any debug level.
Connection messages needed to diagnose session drop/reconnect
events. Use LOG mode for now, discuss lowering in later releases.
present since 8.0 was never fully meaningful, since two recovery targets
cannot be specified. Refactor recovery target type to make this change
and associated code easier to understand. No change in function.
Bug report arising from internal support question.
field into WAL record and reset it from there, rather than using
FrozenTransactionId which can lead to some corner case bugs.
Problem report and suggested route to a fix from Heikki, details by me.
in recovery_end_command, it always came out as 0 because InRedo was
cleared before recovery_end_command was executed. Also, always take
ControlFileLock when reading checkpoint location for %r.
The recovery_end_command bug and the missing locking was present in 8.4
as well, that part of this patch will be backported separately.
to transformAggregateCall, instead of abusing fields in Aggref to carry them
temporarily. No change in functionality but hopefully the code is a bit
clearer now. Per gripe from Gokulakannan Somasundaram.
unable to read a stats file for reasons other than ENOENT, and having to reset
last_statrequest because it's later than current time in the collector.
Not clear if this will shed any light on the "pgstat wait timeout" business,
but it seems like a good idea in general.
In passing, do some message-style-police work on recently-added
pgstat_reset_shared_counters code.
instead of an exclusive lock.
The change is almost for code cleanup. Since there seems to be no
performance benefits from it, backports should not be needed.
Fujii Masao
when warning about column-level privileges. This is more useful than before
and makes the apparent duplication complained of by Piyush Newe not so
duplicate. Also fix lack of quote marks in a related message text.
Back-patch to 8.4, where column-level privileges were introduced.
Stephen Frost
unless (1) the @ isn't quoted and (2) the filename isn't empty. This guards
against unexpectedly treating usernames or other strings in "flat files"
as inclusion requests, as seen in a recent trouble report from Ed L.
The empty-filename case would be guaranteed to misbehave anyway, because our
subsequent path-munging behavior results in trying to read the directory
containing the current input file.
I think this might finally explain the report at
http://archives.postgresql.org/pgsql-bugs/2004-05/msg00132.php
of a crash after printing "authentication file token too long, skipping",
since I was able to duplicate that message (though not a crash) on a
platform where stdio doesn't refuse to read directories. We never got
far in investigating that problem, but now I'm suspicious that the trigger
condition was an @ in the flat password file.
Back-patch to all active branches since the problem can be demonstrated in all
branches except HEAD. The test case, creating a user named "@", doesn't cause
a problem in HEAD since we got rid of the flat password file. Nonetheless it
seems like a good idea to not consider quoted @ as a file inclusion spec,
so I changed HEAD too.
Tom Lane
Simon Riggs
Robert Haas
Itagaki Takahiro
Peter Eisentraut
Bruce Momjian
Heikki Linnakangas
Magnus Hagander
Alvaro Herrera