XLOG_GIN_UPDATE_META_PAGE and XLOG_GIN_DELETE_LISTPAGE records were printed
with a list link field labeled as "blkno", which was confusing, especially
when the link was empty (InvalidBlockNumber). Print the metapage block
number instead, since that's what's actually being updated. We could
include the link values too as a separate field, but not clear it's worth
the trouble.
Back-patch to 8.4 where the dubious code was added.
Initialise ckptXidEpoch from starting checkpoint and maintain the correct
value as we roll forwards. This allows GetNextXidAndEpoch() to return the
correct epoch when executed during recovery. Backpatch to 9.0 when the
problem is first observable by a user.
Bug report from Daniel Farina
In commit 4016bdef8a I fixed a bunch of
ginxlog.c bugs having to do with not handling XLogReadBuffer failures
correctly. However, in ginRedoUpdateMetapage and ginRedoDeleteListPages,
I unaccountably thought that failure to read the metapage would be
impossible and just put in an elog(PANIC) call. This is of course wrong:
failure is exactly what will happen if the index got dropped (or rebuilt)
between creation of the WAL record and the crash we're trying to recover
from. I believe this explains Nicholas Wilson's recent report of these
errors getting reached.
Also, fix memory leak in forgetIncompleteSplit. This wasn't of much
concern when the code was written, but in a long-running standby server
page split records could be expected to accumulate indefinitely.
Back-patch to 8.4 --- before that, GIN didn't have a metapage.
Previously we used ReadRecPtr rather than EndRecPtr, which was
not a serious error but caused pg_stat_replication to report
incorrect replay_location until at least one WAL record is replayed.
Fujii Masao
When "vacuuming" a single btree page by removing LP_DEAD tuples, we are not
actually within a vacuum operation, but rather in an ordinary insertion
process that could well be running concurrently with a vacuum. So clearing
the cycleid is incorrect, and could cause the concurrent vacuum to miss
removing tuples that it needs to remove. This is a longstanding bug
introduced by commit e6284649b9 of
2006-07-25. I believe it explains Maxim Boguk's recent report of index
corruption, and probably some other previously unexplained reports.
In 9.0 and up this is a one-line fix; before that we need to introduce a
flag to tell _bt_delitems what to do.
Fix a longstanding thinko in replay of NEXTOID and checkpoint records: we
tried to advance nextOid only if it was behind the value in the WAL record,
but the comparison would draw the wrong conclusion if OID wraparound had
occurred since the previous value. Better to just unconditionally assign
the new value, since OID assignment shouldn't be happening during replay
anyway.
The consequences of a failure to update nextOid would be pretty minimal,
since we have long had the code set up to obtain another OID and try again
if the generated value is already in use. But in the worst case there
could be significant performance glitches while such loops iterate through
many already-used OIDs before finding a free one.
The odds of a wraparound happening during WAL replay would be small in a
crash-recovery scenario, and the length of any ensuing OID-assignment stall
quite limited anyway. But neither of these statements hold true for a
replication slave that follows a WAL stream for a long period; its behavior
upon going live could be almost unboundedly bad. Hence it seems worth
back-patching this fix into all supported branches.
Already fixed in HEAD in commit c6d76d7c82.
RestoreBkpBlocks was in the habit of zeroing and refilling the target
buffer; which was perfectly safe when the code was written, but is unsafe
during Hot Standby operation. The reason is that we have coding rules
that allow backends to continue accessing a tuple in a heap relation while
holding only a pin on its buffer. Such a backend could see transiently
zeroed data, if WAL replay had occasion to change other data on the page.
This has been shown to be the cause of bug #6425 from Duncan Rance (who
deserves kudos for developing a sufficiently-reproducible test case) as
well as Bridget Frey's re-report of bug #6200. It most likely explains the
original report as well, though we don't yet have confirmation of that.
To fix, change the code so that only bytes that are supposed to change will
change, even transiently. This actually saves cycles in RestoreBkpBlocks,
since it's not writing the same bytes twice.
Also fix seq_redo, which has the same disease, though it has to work a bit
harder to meet the requirement.
So far as I can tell, no other WAL replay routines have this type of bug.
In particular, the index-related replay routines, which would certainly be
broken if they had to meet the same standard, are not at risk because we
do not have coding rules that allow access to an index page when not
holding a buffer lock on it.
Back-patch to 9.0 where Hot Standby was added.
In commit 7b0d0e9356, I made CLUSTER and
VACUUM FULL try to preserve toast value OIDs from the original toast table
to the new one. However, if we have to copy both live and recently-dead
versions of a row that has a toasted column, those versions may well
reference the same toast value with the same OID. The patch then led to
duplicate-key failures as we tried to insert the toast value twice with the
same OID. (The previous behavior was not very desirable either, since it
would have silently inserted the same value twice with different OIDs.
That wastes space, but what's worse is that the toast values inserted for
already-dead heap rows would not be reclaimed by subsequent ordinary
VACUUMs, since they go into the new toast table marked live not deleted.)
To fix, check if the copied OID already exists in the new toast table, and
if so, assume that it stores the desired value. This is reasonably safe
since the only case where we will copy an OID from a previous toast pointer
is when toast_insert_or_update was given that toast pointer and so we just
pulled the data from the old table; if we got two different values that way
then we have big problems anyway. We do have to assume that no other
backend is inserting items into the new toast table concurrently, but
that's surely safe for CLUSTER and VACUUM FULL.
Per bug #6393 from Maxim Boguk. Back-patch to 9.0, same as the previous
patch.
smgrdounlink takes care to not throw an ERROR if it fails to unlink
something, but that caution was rendered useless by commit
3396000684, which put an smgrexists call in
front of it; smgrexists *does* throw error if anything looks funny, such
as getting a permissions error from trying to open the file. If that
happens post-commit, you get a PANIC, and what's worse the same logic
appears in the WAL replay code, so the database even fails to restart.
Restore the intended behavior by removing the smgrexists call --- it isn't
accomplishing anything that we can't do better by adjusting mdunlink's
ideas of whether it ought to warn about ENOENT or not.
Per report from Joseph Shraibman of unrecoverable crash after trying to
drop a table whose FSM fork had somehow gotten chmod'd to 000 permissions.
Backpatch to 8.4, where the bogus coding was introduced.
we don't reach consistency before replaying all of the WAL. Rename the
variable to reachedConsistency, to make its intention clearer.
In master, that was an active bug because of the recent patch to
immediately PANIC if a reference to a missing page is found in WAL after
reaching consistency, as Tom Lane's test case demonstrated. In 9.1 and 9.0,
the only consequence was a misleading "consistent recovery state reached at
%X/%X" message in the log at the beginning of crash recovery (the database
is not consistent at that point yet). In 8.4, the log message was not
printed in crash recovery, even though there was a similar
reachedMinRecoveryPoint local variable that was also set early. So,
backpatch to 9.1 and 9.0.
A simple thinko in ginRedoUpdateMetapage, namely failing to increment a
loop counter, led to inserting records into the last pending-list page in
the wrong order (the opposite of that intended). So far as I can tell,
this would not upset the code that eventually flushes pending items into
the main part of the GIN index. But it did break the code that searched
the pending list for matches, resulting in transient failure to find
matching entries during index lookups, as illustrated in bug #6307 from
Maksym Boguk.
Back-patch to 8.4 where the incorrect code was introduced.
This assumption can be wrong when the toaster is passed a raw on-disk
tuple, because the tuple might pre-date an ALTER TABLE ADD COLUMN operation
that added columns without rewriting the table. In such a case the tuple's
natts value is smaller than what we expect from the tuple descriptor, and
so its t_hoff value could be smaller too. In fact, the tuple might not
have a null bitmap at all, and yet our current opinion of it is that it
contains some trailing nulls.
In such a situation, toast_insert_or_update did the wrong thing, because
to save a few lines of code it would use the old t_hoff value as the offset
where heap_fill_tuple should start filling data. This did not leave enough
room for the new nulls bitmap, with the result that the first few bytes of
data could be overwritten with null flag bits, as in a recent report from
Hubert Depesz Lubaczewski.
The particular case reported requires ALTER TABLE ADD COLUMN followed by
CREATE TABLE AS SELECT * FROM ... or INSERT ... SELECT * FROM ..., and
further requires that there be some out-of-line toasted fields in one of
the tuples to be copied; else we'll not reach the troublesome code.
The problem can only manifest in this form in 8.4 and later, because
before commit a77eaa6a95, CREATE TABLE AS or
INSERT/SELECT wouldn't result in raw disk tuples getting passed directly
to heap_insert --- there would always have been at least a junkfilter in
between, and that would reconstitute the tuple header with an up-to-date
t_natts and hence t_hoff. But I'm backpatching the tuptoaster change all
the way anyway, because I'm not convinced there are no older code paths
that present a similar risk.
This reverts commit 048fffed55.
As pointed out by Naoya Anzai, we need to do more work to make that
idea handle end-of-index cases, and it is looking like too much risk
for a back-patch. So bug #6278 is only going to be fixed in HEAD.
There was a timing window between when oldestActiveXid was derived
and when it should have been derived that only shows itself under
heavy load. Move code around to ensure correct timing of derivation.
No change to StartupSUBTRANS() code, which is where this failed.
Bug report by Chris Redekop
If a tuple in a syscache contains an out-of-line toasted field, and we
try to fetch that field shortly after some other transaction has committed
an update or deletion of the tuple, there is a race condition: vacuum
could come along and remove the toast tuples before we can fetch them.
This leads to transient failures like "missing chunk number 0 for toast
value NNNNN in pg_toast_2619", as seen in recent reports from Andrew
Hammond and Tim Uckun.
The design idea of syscache is that access to stale syscache entries
should be prevented by relation-level locks, but that fails for at least
two cases where toasted fields are possible: ANALYZE updates pg_statistic
rows without locking out sessions that might want to plan queries on the
same table, and CREATE OR REPLACE FUNCTION updates pg_proc rows without
any meaningful lock at all.
The least risky fix seems to be an idea that Heikki suggested when we
were dealing with a related problem back in August: forcibly detoast any
out-of-line fields before putting a tuple into syscache in the first place.
This avoids the problem because at the time we fetch the parent tuple from
the catalog, we should be holding an MVCC snapshot that will prevent
removal of the toast tuples, even if the parent tuple is outdated
immediately after we fetch it. (Note: I'm not convinced that this
statement holds true at every instant where we could be fetching a syscache
entry at all, but it does appear to hold true at the times where we could
fetch an entry that could have a toasted field. We will need to be a bit
wary of adding toast tables to low-level catalogs that don't have them
already.) An additional benefit is that subsequent uses of the syscache
entry should be faster, since they won't have to detoast the field.
Back-patch to all supported versions. The problem is significantly harder
to reproduce in pre-9.0 releases, because of their willingness to flush
every entry in a syscache whenever the underlying catalog is vacuumed
(cf CatalogCacheFlushRelation); but there is still a window for trouble.
The existing scan-direction-sensitive tests were overly complex, and
failed to stop the scan in cases where it's perfectly legitimate to do so.
Per bug #6278 from Maksym Boguk.
Back-patch to 8.3, which is as far back as the patch applies easily.
Doesn't seem worth sweating over a relatively minor performance issue in
8.2 at this late date. (But note that this was a performance regression
from 8.1 and before, so 8.2 is being left as an outlier.)
As observed by Heikki, we need not conflict on heap page locks during an
insert; heap page locks are only aggregated tuple locks, they don't imply
locking "gaps" as index page locks do. So we can avoid some unnecessary
conflicts, and also do the SSI check while not holding exclusive lock on
the target buffer.
Kevin Grittner, reviewed by Jeff Davis. Back-patch to 9.1.
This oversight led to a massive memory leak --- upwards of 10KB per tuple
--- during creation-time verification of an exclusion constraint based on a
GIST index. In most other scenarios it'd just be a leak of 10KB that would
be recovered at end of query, so not too significant; though perhaps the
leak would be noticeable in a situation where a GIST index was being used
in a nestloop inner indexscan. In any case, it's a real leak of long
standing, so patch all supported branches. Per report from Harald Fuchs.
streamed backup, throw an error and refuse to start up. The restore has not
finished correctly in that case and the data directory is possibly corrupt.
We already errored out in case of archive recovery, but could not during
crash recovery because we couldn't distinguish between the case that
pg_start_backup() was called and the database then crashed (must not error,
data is OK), and the case that we're restoring from a backup and not all
the needed WAL was replayed (data can be corrupt).
To distinguish those cases, add a line to backup_label to indicate
whether the backup was taken with pg_start/stop_backup(), or by streaming
(ie. pg_basebackup).
This is a different implementation than what I committed to 9.2 a week ago.
That implementation was not back-patchable because it required re-initdb.
Fujii Masao
This works around the problem that a catalog cache entry might contain a
toast pointer that we try to dereference just as a VACUUM FULL completes
on that catalog. We will see the sinval message on the cache entry when
we acquire lock on the toast table, but by that point we've already told
tuptoaster.c "here's the pointer to fetch", so it's difficult from a code
structural standpoint to update the pointer before we use it. Much less
painful to ensure that toast pointers are not invalidated in the first
place. We have to add a bit of code to deal with the case that a value
that previously wasn't toasted becomes so; but that should be a
seldom-exercised corner case, so the inefficiency shouldn't be significant.
Back-patch to 9.0. In prior versions, we didn't allow CLUSTER on system
catalogs, and VACUUM FULL didn't result in reassignment of toast OIDs, so
there was no problem.
The previous code tried to synchronize by unlinking the init file twice,
but that doesn't actually work: it leaves a window wherein a third process
could read the already-stale init file but miss the SI messages that would
tell it the data is stale. The result would be bizarre failures in catalog
accesses, typically "could not read block 0 in file ..." later during
startup.
Instead, hold RelCacheInitLock across both the unlink and the sending of
the SI messages. This is more straightforward, and might even be a bit
faster since only one unlink call is needed.
This has been wrong since it was put in (in 2002!), so back-patch to all
supported releases.
Fix a whole bunch of signal handlers that had been hacked to do things that
might change errno, without adding the necessary save/restore logic for
errno. Also make some minor fixes in unix_latch.c, and clean up bizarre
and unsafe scheme for disowning the process's latch. While at it, rename
the PGPROC latch field to procLatch for consistency with 9.2.
Issues noted while reviewing a patch by Peter Geoghegan.
The original definition had the problem that timeouts exceeding about 2100
seconds couldn't be specified on 32-bit machines. Milliseconds seem like
sufficient resolution, and finer grain than that would be fantasy anyway
on many platforms.
Back-patch to 9.1 so that this aspect of the latch API won't change between
9.1 and later releases.
Peter Geoghegan
First, when following a right-link, we incorrectly marked the current page
as the parent of the right sibling. In reality, the parent of the right page
is the same as the parent of the current page (or some page to the right of
it, gistFindCorrectParent() will sort that out).
Secondly, when we follow a right-link, we must prepend, not append, the right
page to our list of pages to visit. That's because we assume that once we
hit a leaf page in the list, all the rest are leaf pages too, and give up.
To hit these bugs, you need concurrent actions and several unlucky accidents.
Another backend must split the root page, while you're in process of
splitting a lower-level page. Furthermore, while you scan the internal nodes
to re-find the parent, another backend needs to again split some more internal
pages. Even then, the bugs don't necessarily manifest as user-visible errors
or index corruption.
While we're at it, make the error reporting a bit better if gistFindPath()
fails to re-find the parent. It used to be an assertion, but an elog() seems
more appropriate.
Backpatch to all supported branches.
Such a condition is unsatisfiable in combination with any other type of
btree-indexable condition (since we assume btree operators are always
strict). 8.3 and 8.4 had an explicit test for this, which I removed in
commit 29c4ad9829, mistakenly thinking that
the case would be subsumed by the more general handling of IS (NOT) NULL
added in that patch. Put it back, and improve the comments about it, and
add a regression test case.
Per bug #6079 from Renat Nasyrov, and analysis by Dean Rasheed.
more consistent that way, since all the other PredicateLock* calls are
made in various heapam.c and index AM functions. The call in nodeSeqscan.c
was unnecessarily aggressive anyway, there's no need to try to lock the
relation every time a tuple is fetched, it's enough to do it once.
This has the user-visible effect that if a seq scan is initialized in the
executor, but never executed, we now acquire the predicate lock on the heap
relation anyway. We could avoid that by taking the lock on the first
heap_getnext() call instead, but it doesn't seem worth the trouble given
that it feels more natural to do it in heap_beginscan().
Also, remove the retail PredicateLockTuple() calls from heap_getnext(). In
a seqscan, started with heap_begin(), we're holding a whole-relation
predicate lock on the heap so there's no need to lock the tuples
individually.
Kevin Grittner and me
WAL records of type XLOG_BTREE_REUSE_PAGE were generated using a
latestRemovedXid one higher than actually needed because xid used was
page opaque->btpo.xact rather than an actually removed xid.
Noticed on an otherwise quiet system by Noah Misch.
Noah Misch and Simon Riggs
Btree pages were recycled after VACUUM deletes all records on a
page and then a subsequent VACUUM occurs after the RecentXmin
horizon is reached. Using RecentXmin meant that we did not respond
correctly to the user controls provide to avoid Hot Standby
conflicts and so spurious conflicts could be generated in some
workload combinations. We now reuse pages only when we reach
RecentGlobalXmin, which can be much later in the presence of long
running queries and is also controlled by vacuum_defer_cleanup_age
and hot_standby_feedback.
Noah Misch and Simon Riggs
snapshots, like in REINDEX, are basically non-transactional operations. The
DDL operation itself might participate in SSI, but there's separate
functions for that.
Kevin Grittner and Dan Ports, with some changes by me.
ReadRecord's habit of using both direct references to tmpRecPtr and
references to *RecPtr (which is pointing at tmpRecPtr) triggers an
optimization bug in gcc 4.6.0, which apparently has forgotten about
aliasing rules. Avoid the compiler bug, and make the code more readable
to boot, by getting rid of the direct references. Improve the comments
while at it.
Back-patch to all supported versions, in case they get built with 4.6.0.
Tom Lane, with some cosmetic suggestions from Alex Hunsaker
Apparently sane-looking penalty code might return small negative values,
for example because of roundoff error. This will confuse places like
gistchoose(). Prevent problems by clamping negative penalty values to
zero. (Just to be really sure, I also made it force NaNs to zero.)
Back-patch to all supported branches.
Alexander Korotkov
On further analysis, it turns out that it is not needed to duplicate predicate
locks to the new row version at update, the lock on the version that the
transaction saw as visible is enough. However, there was a different bug in
the code that checks for dangerous structures when a new rw-conflict happens.
Fix that bug, and remove all the row-version chaining related code.
Kevin Grittner & Dan Ports, with some comment editorialization by me.
The SSI patch inserted a call of RegisterPredicateLockingXid into
GetNewTransactionId, which was a bad idea on a couple of grounds. First,
it's not necessary to hold XidGenLock while manipulating that shared
memory, and doing so is bad because XidGenLock is a high-contention lock
that should be held for as short a time as possible. (Not to mention that
it adds an entirely unnecessary deadlock hazard, since we must take
SerializableXactHashLock as well.) Second, the specific place where it was
put was between extending CLOG and advancing nextXid, which could result in
unpleasant behavior in case of a failure there. Pull the call out to
AssignTransactionId, which is much safer and arguably better from a
modularity standpoint too.
There is more work to do to clean up the failure-before-advancing-nextXid
issue, but that is a separate change that will need to be back-patched.
So for the moment I just want to make GetNewTransactionId look the same as
it did in prior versions.
Experimentation with contrib/btree_gist shows that the majority of the GIST
support functions potentially need collation information. Safest policy
seems to be to pass it to all of them, instead of making assumptions about
which ones could possibly need it.
This patch is almost entirely cosmetic --- mostly cleaning up a lot of
neglected comments, and fixing code layout problems in places where the
patch made lines too long and then pgindent did weird things with that.
I did find a bug-of-omission in equalTupleDescs().
Before commit c016ce7281, this wasn't
needed, but now that multiple resource manager IDs can percolate down
through here, we have to make sure we know which one we've got.
Otherwise, we can confuse (for example) an XLOG_XACT_COMMIT record
with an XLOG_CHECKPOINT_SHUTDOWN record.
Review by Jaime Casanova
Tom Lane