| Age | Commit message (Collapse) | Author |
|---|
|
There can never be a need to push the indcheckxmin horizon forward, since
any HOT chains that are actually broken with respect to the index must
pre-date its original creation. So we can just avoid changing pg_index
altogether during a REINDEX operation.
This offers a cleaner solution than my previous patch for the problem
found a few days ago that we mustn't try to update pg_index while we are
reindexing it. System catalog indexes will always be created with
indcheckxmin = false during initdb, and with this modified code we should
never try to change their pg_index entries. This avoids special-casing
system catalogs as the former patch did, and should provide a performance
benefit for many cases where REINDEX formerly caused an index to be
considered unusable for a short time.
Back-patch to 8.3 to cover all versions containing HOT. Note that this
patch changes the API for index_build(), but I believe it is unlikely that
any add-on code is calling that directly.
|
|
This reverts commit a03e3e1fd1d4ecfeb1096aeb7854b717061a75d9 of 2011-04-15.
There's a better way to do it, which will follow shortly.
|
|
The places that attempt to change pg_index.indcheckxmin during a reindexing
operation cannot be executed safely if pg_index itself is the subject of
the operation. This is the explanation for a couple of recent reports of
VACUUM FULL failing with
ERROR: duplicate key value violates unique constraint "pg_index_indexrelid_index"
DETAIL: Key (indexrelid)=(2678) already exists.
However, there isn't any real need to update indcheckxmin in such a
situation, if we assume that pg_index can never contain a truly broken HOT
chain. This assumption holds if new indexes are never created on it during
concurrent operations, which is something we don't consider safe for any
system catalog, not just pg_index. Accordingly, modify the code to not
manipulate indcheckxmin when reindexing any system catalog.
Back-patch to 8.3, where HOT was introduced. The known failure scenarios
involve 9.0-style VACUUM FULL, so there might not be any real risk before
9.0, but let's not assume that.
|
|
|
|
When we are doing GEQO join planning, the current memory context is a
short-lived context that will be reset at the end of geqo_eval(). However,
the RelOptInfos for base relations are set up before that and then re-used
across many GEQO cycles. Hence, any code that modifies a baserel during
join planning has to be careful not to put pointers to the short-lived
context into the baserel struct. mark_dummy_rel got this wrong, leading to
easy-to-reproduce-once-you-know-how crashes in 8.4, as reported off-list by
Leo Carson of SDSC. Some improvements made in 9.0 make it difficult to
demonstrate the crash in 9.0 or HEAD; but there's no doubt that there's
still a risk factor here, so patch all branches that have the function.
(Note: 8.3 has a similar function, but it's only applied to joinrels and
thus is not a hazard.)
|
|
than on other platforms, and only IPv6 addresses are returned. Because of
those two issues, fall back to ioctl(SIOCGIFCONF) on HP/UX, so that it at
least compiles and finds IPv4 addresses. This function is currently only
used for interpreting samehost/samenet in pg_hba.conf, which isn't that
critical.
|
|
crash recovery, and throw an error if not. hubert depesz lubaczewski pointed
out that that situation also happens in the crash recovery following a
system crash that happens during an online backup.
We might want to do something smarter in 9.1, like put the check back for
backups taken with pg_basebackup, but that's for another patch.
|
|
to the regular stack. The code to do that is platform and compiler specific,
add support for the HP-UX native compiler.
|
|
pg_hba.conf.
Per gripe from Josh Berkus. Backported from commit ed557a373c406bbb2a1843544ebbd856ca4cac47.
|
|
In particular, if we don't have real ndistinct estimates for both sides,
fall back to assuming that half of the left-hand rows have join partners.
This is what was done in 8.2 and 8.3 (cf nulltestsel() in those versions).
It's pretty stupid but it won't lead us to think that an antijoin produces
no rows out, as seen in recent example from Uwe Schroeder.
|
|
Remove the hard-wired assumption that __mips__ (and only __mips__) lacks
dlopen in FreeBSD and OpenBSD. This assumption is outdated at least for
OpenBSD, as per report from an anonymous 9.1 tester. We can perfectly well
use HAVE_DLOPEN instead to decide which code to use.
Some other cosmetic adjustments to make freebsd.c, netbsd.c, and openbsd.c
exactly alike.
|
|
Instead, write the unconverted output - it will be in the wrong
encoding, but at least we don't crash.
Rushabh Lathia
|
|
archive recovery.
It's possible to restore an online backup without recovery.conf, by simply
copying all the necessary WAL files to pg_xlog. "pg_basebackup -x" does that
too. That's the use case where this cross-check is useful.
Backpatch to 9.0. We used to do this in earlier versins, but in 9.0 the code
was inadvertently changed so that the check is only performed after archive
recovery.
Fujii Masao.
|
|
Eventually we might be able to allow that, but it's not clear how many
places need to be fixed to prevent infinite recursion when there's a direct
or indirect inclusion of a rowtype in itself. One such place is
CheckAttributeType(), which will recurse to stack overflow in cases such as
those exhibited in bug #5950 from Alex Perepelica. If we were sure it was
the only such place, we could easily modify the code added by this patch to
stop the recursion without a complaint ... but it probably isn't the only
such place. Hence, throw error until such time as someone is excited
enough about this type of usage to put work into making it safe.
Back-patch as far as 8.3. 8.2 doesn't have the recursive call in
CheckAttributeType in the first place, so I see no need to add code there
in the absence of clear evidence of a problem elsewhere.
|
|
Startup process waited for cleanup lock but when hot_standby = off
the pid was not registered, so that the bgwriter would not wake
the waiting process as intended.
|
|
Opening a catcache's index could require reading from that cache's own
catalog, which of course would acquire AccessShareLock on the catalog.
So the original coding here risks locking index before heap, which could
deadlock against another backend trying to get exclusive locks in the
normal order. Because InitCatCachePhase2 is only called when a backend
has to start up without a relcache init file, the deadlock was seldom seen
in the field. (And by the same token, there's no need to worry about any
performance disadvantage; so not much point in trying to distinguish
exactly which catalogs have the risk.)
Bug report, diagnosis, and patch by Nikhil Sontakke. Additional commentary
by me. Back-patch to all supported branches.
|
|
We previously heard of the same problem in int24div(), so there's not a
good reason to suppose the problem is confined to cases involving int8.
|
|
Add dummy returns before every potential division-by-zero in int8.c,
because apparently further "improvements" in gcc's optimizer have
enabled it to break functions that weren't broken before.
Aurelien Jarno, via Martin Pitt
|
|
contains newly-inserted tuples that according to our OldestXmin are not
yet visible to everyone. The value returned by GetOldestXmin() is conservative,
and it can move backwards on repeated calls, so if we see that contradiction
between the PD_ALL_VISIBLE flag and status of tuples on the page, we have to
assume it's because an earlier vacuum calculated a higher OldestXmin value,
and all the tuples really are visible to everyone.
We have received several reports of this bug, with the "PD_ALL_VISIBLE flag
was incorrectly set in relation ..." warning appearing in logs. We were
finally able to hunt it down with David Gould's help to run extra diagnostics
in an environment where this happened frequently.
Also reword the warning, per Robert Haas' suggestion, to not imply that the
PD_ALL_VISIBLE flag is necessarily at fault, as it might also be a symptom
of corruption on a tuple header.
Backpatch to 8.4, where the PD_ALL_VISIBLE flag was introduced.
|
|
ExecUpdate checked for whether ExecBRUpdateTriggers had returned a new
tuple value by seeing if the returned tuple was pointer-equal to the old
one. But the "old one" was in estate->es_junkFilter's result slot, which
would be scribbled on if we had done an EvalPlanQual update in response to
a concurrent update of the target tuple; therefore we were comparing a
dangling pointer to a live one. Given the right set of circumstances we
could get a false match, resulting in not forcing the tuple to be stored in
the slot we thought it was stored in. In the case reported by Maxim Boguk
in bug #5798, this led to "cannot extract system attribute from virtual
tuple" failures when trying to do "RETURNING ctid". I believe there is a
very-low-probability chance of more serious errors, such as generating
incorrect index entries based on the original rather than the
trigger-modified version of the row.
In HEAD, change all of ExecBRInsertTriggers, ExecIRInsertTriggers,
ExecBRUpdateTriggers, and ExecIRUpdateTriggers so that they continue to
have similar APIs. In the back branches I just changed
ExecBRUpdateTriggers, since there is no bug in the ExecBRInsertTriggers
case.
|
|
ts_typanalyze.c computes MCE statistics as fractions of the non-null rows,
which seems fairly reasonable, and anyway changing it in released versions
wouldn't be a good idea. But then ts_selfuncs.c has to account for that.
Failure to do so results in overestimates in columns with a significant
fraction of null documents. Back-patch to 8.4 where this stuff was
introduced.
Jesper Krogh
|
|
That function was supposing that indexoid == 0 for a hypothetical index,
but that is not likely to be true in any non-toy implementation of an index
adviser, since assigning a fake OID is the only way to know at EXPLAIN time
which hypothetical index got selected. Fix by adding a flag to
IndexOptInfo to mark hypothetical indexes. Back-patch to 9.0 where
get_actual_variable_range() was added.
Gurjeet Singh
|
|
Recent releases had a check on rel->rd_refcnt in heap_drop_with_catalog,
but failed to cover the possibility of pending trigger events at DROP time.
(Before 8.4 we didn't even check the refcnt.) When the trigger events were
eventually fired, you'd get "could not open relation with OID nnn" errors,
as in recent report from strk. Better to throw a suitable error when the
DROP is attempted.
Also add a similar check in DROP INDEX.
Back-patch to all supported branches.
|
|
Flattening of subquery range tables during setrefs.c could lead to the
rangetable indexes in PlanRowMark nodes not matching up with the column
names previously assigned to the corresponding resjunk ctid (resp. tableoid
or wholerow) columns. Typical symptom would be either a "cannot extract
system attribute from virtual tuple" error or an Assert failure. This
wasn't a problem before 9.0 because we didn't support FOR UPDATE below the
top query level, and so the final flattening could never renumber an RTE
that was relevant to FOR UPDATE. Fix by using a plan-tree-wide unique
number for each PlanRowMark to label the associated resjunk columns, so
that the number need not change during flattening.
Per report from David Johnston (though I'm darned if I can see how this got
past initial testing of the relevant code). Back-patch to 9.0.
|
|
Previously reported as ERRCODE_ADMIN_SHUTDOWN, this case is now
reported as ERRCODE_DATABASE_DROPPED. No message text change.
Unlikely to happen on most servers, so low impact change to allow
session poolers to correctly handle this situation.
Tatsuo Ishii and Simon Riggs
|
|
maximum allowed' messages, that have reported one-less dimensions.
Alexey Klyukin
|
|
All retryable conflict errors now have an error code that indicates that
a retry is possible, correcting my incomplete fix of 2010/05/12
Tatsuo Ishii and Simon Riggs, input from Robert Haas and Florian Pflug
|
|
reduce_outer_joins() mistakenly treated a semijoin like a left join for
purposes of deciding whether not-null constraints created by the join's
quals could be passed down into the join's left-hand side (possibly
resulting in outer-join simplification there). Actually, semijoin works
like inner join for this purpose, ie, we do not need to see any rows that
can't possibly satisfy the quals. Hence, two-line fix to treat semi and
inner joins alike. Per observation by Andres Freund about a performance
gripe from Yazan Suleiman.
Back-patch to 8.4, since this oversight has been there since the current
handling of semijoins was implemented.
|
|
We only need that header when compiling with icc, since the gcc variant of
ia64_get_bsp() uses in-line assembly code. Per report from Frank Brendel,
the header doesn't exist on all IA64 platforms; so don't include it unless
we need it.
|
|
|
|
Failure to do so can lead to constraint violations. This was broken by
commit 1ddc2703a936d03953657f43345460b9242bbed1 on 2010-02-07, so
back-patch to 9.0.
Noah Misch. Regression test by me.
|
|
If the slice to be assigned to was before the existing array lower bound
(requiring at least one null element to spring into existence to fill the
gap), the code miscalculated how many entries needed to be copied from
the old array's null bitmap. This could result in trashing the array's
data area (as seen in bug #5840 from Karsten Loesing), or worse.
This has been broken since we first allowed the behavior of assigning to
non-adjacent slices, in 8.2. Back-patch to all affected versions.
|
|
Otherwise WAL recovery will replay the un-flushed WAL after walreceiver has
exited, which can lead to a non-recoverable standby if the system crashes hard
at that point.
|
|
that can be read without blocking. It used to conclude that there isn't, even
though there was data in the socket receive buffer. That lead walreceiver to
flush the WAL after every received chunk, potentially causing big performance
issues.
Backpatch to 9.0, because the performance impact can be very significant.
|
|
In an inherited UPDATE/DELETE, each target table has its own subplan,
because it might have a column set different from other targets. This
means that the resjunk columns we add to support EvalPlanQual might be
at different physical column numbers in each subplan. The EvalPlanQual
rewrite I did for 9.0 failed to account for this, resulting in possible
misbehavior or even crashes during concurrent updates to the same row,
as seen in a recent report from Gordon Shannon. Revise the data structure
so that we track resjunk column numbers separately for each subplan.
I also chose to move responsibility for identifying the physical column
numbers back to executor startup, instead of assuming that numbers derived
during preprocess_targetlist would stay valid throughout subsequent
massaging of the plan. That's a bit slower, so we might want to consider
undoing it someday; but it would complicate the patch considerably and
didn't seem justifiable in a bug fix that has to be back-patched to 9.0.
|
|
The "date" type supports a wider range of dates than int64 timestamps do.
However, there is pre-int64-timestamp code in the planner that assumes that
all date values can be converted to timestamp with impunity. Fortunately,
what we really need out of the conversion is always a double (float8)
value; so even when the date is out of timestamp's range it's possible to
produce a sane answer. All we need is a code path that doesn't try to
force the result into int64. Per trouble report from David Rericha.
Back-patch to all supported versions. Although this is surely a corner
case, there's not much point in advertising a date range wider than
timestamp's if we will choke on such values in unexpected places.
|
|
eval_const_expressions() can replace CaseTestExprs with constants when
the surrounding CASE's test expression is a constant. This confuses
ruleutils.c's heuristic for deparsing simple-form CASEs, leading to
Assert failures or "unexpected CASE WHEN clause" errors. I had put in
a hack solution for that years ago (see commit
514ce7a331c5bea8e55b106d624e55732a002295 of 2006-10-01), but bug #5794
from Peter Speck shows that that solution failed to cover all cases.
Fortunately, there's a much better way, which came to me upon reflecting
that Peter's "CASE TRUE WHEN" seemed pretty redundant: we can "simplify"
the simple-form CASE to the general form of CASE, by simply omitting the
constant test expression from the rebuilt CASE construct. This is
intuitively valid because there is no need for the executor to evaluate
the test expression at runtime; it will never be referenced, because any
CaseTestExprs that would have referenced it are now replaced by constants.
This won't save a whole lot of cycles, since evaluating a Const is pretty
cheap, but a cycle saved is a cycle earned. In any case it beats kluging
ruleutils.c still further. So this patch improves const-simplification
and reverts the previous change in ruleutils.c.
Back-patch to all supported branches. The bug exists in 8.1 too, but it's
out of warranty.
|
|
After parsing a parenthesized subexpression, we must pop all pending
ANDs and NOTs off the stack, just like the case for a simple operand.
Per bug #5793.
Also fix clones of this routine in contrib/intarray and contrib/ltree,
where input of types query_int and ltxtquery had the same problem.
Back-patch to all supported versions.
|
|
We don't actually need optreset, because we can easily fix the code to
ensure that it's cleanly restartable after having completed a scan over the
argv array; which is the only case we need to restart in. Getting rid of
it avoids a class of interactions with the system libraries and allows
reversion of my change of yesterday in postmaster.c and postgres.c.
Back-patch to 8.4. Before that the getopt code was a bit different anyway.
|
|
The mingw people don't appear to care about compatibility with non-GNU
versions of getopt, so force use of our own copy of getopt on Windows.
Also, ensure that we make use of optreset when using our own copy.
Per report from Andrew Dunstan. Back-patch to all versions supported
on Windows.
|
|
|
|
The original coding in tuplestore_trim() was only meant to work efficiently
in cases where each trim call deleted most of the tuples in the store.
Which, in fact, was the pattern of the original usage with a Material node
supporting mark/restore operations underneath a MergeJoin. However,
WindowAgg now uses tuplestores and it has considerably less friendly
trimming behavior. In particular it can attempt to trim one tuple at a
time off a large tuplestore. tuplestore_trim() had O(N^2) runtime in this
situation because of repeatedly shifting its tuple pointer array. Fix by
avoiding shifting the array until a reasonably large number of tuples have
been deleted. This can waste some pointer space, but we do still reclaim
the tuples themselves, so the percentage wastage should be pretty small.
Per Jie Li's report of slow percent_rank() evaluation. cume_dist() and
ntile() would certainly be affected as well, along with any other window
function that has a moving frame start and requires reading substantially
ahead of the current row.
Back-patch to 8.4, where window functions were introduced. There's no
need to tweak it before that.
|
|
Hot Standby conflicts only with tuples that were visible at
some point. So ignore tuples from aborted transactions or for
tuples updated/deleted during the inserting transaction when
generating the conflict transaction ids.
Following detailed analysis and test case by Noah Misch.
Original report covered btree delete records, correctly observed
by Heikki Linnakangas that this applies to other cases also.
Fix covers all sources of cleanup records via common code.
Includes additional fix compared to commit on HEAD
|
|
Recent versions of the Linux system header files cause xlogdefs.h to
believe that open_datasync should be the default sync method, whereas
formerly fdatasync was the default on Linux. open_datasync is a bad
choice, first because it doesn't actually outperform fdatasync (in fact
the reverse), and second because we try to use O_DIRECT with it, causing
failures on certain filesystems (e.g., ext4 with data=journal option).
This part of the patch is largely per a proposal from Marti Raudsepp.
More extensive changes are likely to follow in HEAD, but this is as much
change as we want to back-patch.
Also clean up confusing code and incorrect documentation surrounding the
fsync_writethrough option. Those changes shouldn't result in any actual
behavioral change, but I chose to back-patch them anyway to keep the
branches looking similar in this area.
In 9.0 and HEAD, also do some copy-editing on the WAL Reliability
documentation section.
Back-patch to all supported branches, since any of them might get used
on modern Linux versions.
|
|
an old transaction running in the master, and a lot of transactions have
started and finished since, and a WAL-record is written in the gap between
the creating the running-xacts snapshot and WAL-logging it, recovery will fail
with "too many KnownAssignedXids" error. This bug was reported by
Joachim Wieland on Nov 19th.
In the same scenario, when fewer transactions have started so that all the
xids fit in KnownAssignedXids despite the first bug, a more serious bug
arises. We incorrectly initialize the clog code with the oldest still running
transaction, and when we see the WAL record belonging to a transaction with
an XID larger than one that committed already before the checkpoint we're
recovering from, we zero the clog page containing the already committed
transaction, leading to data loss.
In hindsight, trying to track xids in the known-assigned-xids array before
seeing the running-xacts record was too complicated. To fix that, hold
XidGenLock while the running-xacts snapshot is taken and WAL-logged. That
ensures that no transaction can begin or end in that gap, so that in recvoery
we know that the snapshot contains all transactions running at that point in
WAL.
|
|
There are some code paths, such as SPI_execute(), where we invoke
copyObject() on raw parse trees before doing parse analysis on them. Since
the bison grammar is capable of building heavily nested parsetrees while
itself using only minimal stack depth, this means that copyObject() can be
the front-line function that hits stack overflow before anything else does.
Accordingly, it had better have a check_stack_depth() call. I did a bit of
performance testing and found that this slows down copyObject() by only a
few percent, so the hit ought to be negligible in the context of complete
processing of a query.
Per off-list report from Toshihide Katayama. Back-patch to all supported
branches.
|
|
|
|
There were corner cases in which the planner would attempt to inline such
a function, which would result in a failure at runtime due to loss of
information about exactly what the result record type is. Fix by disabling
inlining when the function's recorded result type is RECORD. There might
be some sub-cases where inlining could still be allowed, but this is a
simple and backpatchable fix, so leave refinements for another day.
Per bug #5777 from Nate Carson.
Back-patch to all supported branches. 8.1 happens to avoid a core-dump
here, but it still does the wrong thing.
|
|
removing an infrequently occurring race condition in Hot Standby.
An xid must be assigned before a lock appears in shared memory,
rather than immediately after, else GetRunningTransactionLocks()
may see InvalidTransactionId, causing assertion failures during
lock processing on standby.
Bug report and diagnosis by Fujii Masao, fix by me.
|
|
When using default autovacuum_vac_cost_limit, autovac_balance_cost relied
on VacuumCostLimit to contain the correct global value ... but after the
first time through in a particular worker process, it didn't, because we'd
trashed it in previous iterations. Depending on the state of other autovac
workers, this could result in a steady reduction of the effective
cost_limit setting as a particular worker processed more and more tables,
causing it to go slower and slower. Spotted by Simon Poole (bug #5759).
Fix by saving and restoring the GUC variables in the loop in do_autovacuum.
In passing, improve a few comments.
Back-patch to 8.3 ... the cost rebalancing code has been buggy since it was
put in.
|
