pg_ctl is using isatty() to verify whether the process is running in a
terminal, and if not it sends its output to Windows' Event Log ... which
does the wrong thing when the output has been redirected to a pipe, as
reported in bug #13592.
To fix, make pg_ctl use the code we already have to detect service-ness:
in the master branch, move src/backend/port/win32/security.c to src/port
(with suitable tweaks so that it runs properly in backend and frontend
environments); pg_ctl already has access to pgport so it Just Works. In
older branches, that's likely to cause trouble, so instead duplicate the
required code in pg_ctl.c.
Author: Michael Paquier
Bug report and diagnosis: Egon Kocjan
Backpatch: all supported branches
In commit 921191912c I claimed that we weren't testing encoding
conversion functions, but further poking around reveals that we did
have an equivalent though hard-wired set of tests in conversion.sql.
AFAICS there is no advantage to doing it like that as compared to letting
the catalog contents drive the test, so let the opr_sanity addition stand
and remove the now-redundant tests in conversion.sql.
Also, remove some infrastructure in src/backend/utils/mb/conversion_procs
for building conversion.sql's list of tests. That was unmaintained, and
had not corresponded to the actual contents of conversion.sql since 2007
or perhaps even further back.
Using cstring as the input type was a poor decision, because that's not
really a full-fledged type. In particular, it lacks implicit coercions
from text or varchar, meaning that usages like to_regproc('foo'||'bar')
wouldn't work; basically the only case that did work without explicit
casting was a simple literal constant argument.
The lack of field complaints about this suggests that hardly anyone
is using these functions, so hopefully fixing it won't cause much of
a compatibility problem. They've only been there since 9.4, anyway.
Petr Korobeinikov
While the in-core authentication mechanism doesn't need to access
pg_shseclabel at all, it's reasonable to think that an authentication
hook will want to look at the label for the role logging in, or for rows
in other catalogs used during the authentication phase of startup.
Catalog version bumped, because this changes the "is nailed" status for
pg_shseclabel.
Author: Adam Brightwell
Some time back we agreed that row_security=off should not be a way to
bypass RLS entirely, but only a way to get an error if it was being
applied. However, the code failed to act that way for table owners.
Per discussion, this is a must-fix bug for 9.5.0.
Adjust the logic in rls.c to behave as expected; also, modify the
error message to be more consistent with the new interpretation.
The regression tests need minor corrections as well. Also update
the comments about row_security in ddl.sgml to be correct. (The
official description of the GUC in config.sgml is already correct.)
I failed to resist the temptation to do some other very minor
cleanup as well, such as getting rid of a duplicate extern declaration.
Aside from any consistency arguments, this is logically necessary because
the I/O functions for these types also handle numeric OID values. Without
a quoting rule it is impossible to distinguish numeric OIDs from role or
namespace names that happen to contain only digits.
Also change the to_regrole and to_regnamespace functions to dequote their
arguments. While not logically essential, this seems like a good idea
since the other to_reg* functions do it. Anyone who really wants raw
lookup of an uninterpreted name can fall back on the time-honored solution
of (SELECT oid FROM pg_namespace WHERE nspname = whatever).
Report and patch by Jim Nasby, reviewed by Michael Paquier
Can't release the AccessExclusiveLock on the target table until commit.
Otherwise there is a race condition whereby other backends might service
our cache invalidation signals before they can actually see the updated
catalog rows.
Just to add insult to injury, RemovePolicyById was closing the rel (with
incorrect lock drop) and then passing the now-dangling rel pointer to
CacheInvalidateRelcache. Probably the reason this doesn't fall over on
CLOBBER_CACHE buildfarm members is that some outer level of the DROP logic
is still holding the rel open ... but it'd have bit us on the arse
eventually, no doubt.
The CHECK_IS_BINARY_UPGRADE macro is not sufficient security protection
if we're going to dereference pass-by-reference arguments before it.
But in any case we really need to explicitly check PG_ARGISNULL for all
the arguments of a non-strict function, not only the ones we expect null
values for.
Oversight in commits 30982be4e5 and
f92fc4c95d. Found by Andreas Seltenreich.
(The other usages in pg_upgrade_support.c seem safe.)
pgwin32_recv() has treated a non-error return of zero bytes from WSARecv()
as being a reason to block ever since the current implementation was
introduced in commit a4c40f140d. However, so far as one can tell
from Microsoft's documentation, that is just wrong: what it means is
graceful connection closure (in stream protocols) or receipt of a
zero-length message (in message protocols), and neither case should result
in blocking here. The only reason the code worked at all was that control
then fell into the retry loop, which did *not* treat zero bytes specially,
so we'd get out after only wasting some cycles. But as of 9.5 we do not
normally reach the retry loop and so the bug is exposed, as reported by
Shay Rojansky and diagnosed by Andres Freund.
Remove the unnecessary test on the byte count, and rearrange the code
in the retry loop so that it looks identical to the initial sequence.
Back-patch to 9.5. The code is wrong all the way back, AFAICS, but
since it's relatively harmless in earlier branches we'll leave it alone.
spg_text_inner_consistent is capable of reconstructing an empty string
to pass down to the next index level; this happens if we have an empty
string coming in, no prefix, and a dummy node label. (In practice, what
is needed to trigger that is insertion of a whole bunch of empty-string
values.) Then, we will arrive at the next level with in->level == 0
and a non-NULL (but zero length) in->reconstructedValue, which is valid
but the Assert tests weren't expecting it.
Per report from Andreas Seltenreich. This has no impact in non-Assert
builds, so should not be a problem in production, but back-patch to
all affected branches anyway.
In passing, remove a couple of useless variable initializations and
shorten the code by not duplicating DatumGetPointer() calls.
flatten_reloptions() supposed that it didn't really need to do anything
beyond inserting commas between reloption array elements. However, in
principle the value of a reloption could be nearly anything, since the
grammar allows a quoted string there. Any restrictions on it would come
from validity checking appropriate to the particular option, if any.
A reloption value that isn't a simple identifier or number could thus lead
to dump/reload failures due to syntax errors in CREATE statements issued
by pg_dump. We've gotten away with not worrying about this so far with
the core-supported reloptions, but extensions might allow reloption values
that cause trouble, as in bug #13840 from Kouhei Sutou.
To fix, split the reloption array elements explicitly, and then convert
any value that doesn't look like a safe identifier to a string literal.
(The details of the quoting rule could be debated, but this way is safe
and requires little code.) While we're at it, also quote reloption names
if they're not safe identifiers; that may not be a likely problem in the
field, but we might as well try to be bulletproof here.
It's been like this for a long time, so back-patch to all supported
branches.
Kouhei Sutou, adjusted some by me
A report from Andy Colson showed that gincostestimate() was not being
nearly paranoid enough about whether to believe the statistics it finds in
the index metapage. The problem is that the metapage stats (other than the
pending-pages count) are only updated by VACUUM, and in the worst case
could still reflect the index's original empty state even when it has grown
to many entries. We attempted to deal with that by scaling up the stats to
match the current index size, but if nEntries is zero then scaling it up
still gives zero. Moreover, the proportion of pages that are entry pages
vs. data pages vs. pending pages is unlikely to be estimated very well by
scaling if the index is now orders of magnitude larger than before.
We can improve matters by expanding the use of the rule-of-thumb estimates
I introduced in commit 7fb008c5ee59b040: if the index has grown by more
than a cutoff amount (here set at 4X growth) since VACUUM, then use the
rule-of-thumb numbers instead of scaling. This might not be exactly right
but it seems much less likely to produce insane estimates.
I also improved both the scaling estimate and the rule-of-thumb estimate
to account for numPendingPages, since it's reasonable to expect that that
is accurate in any case, and certainly pages that are in the pending list
are not either entry or data pages.
As a somewhat separate issue, adjust the estimation equations that are
concerned with extra fetches for partial-match searches. These equations
suppose that a fraction partialEntries / numEntries of the entry and data
pages will be visited as a consequence of a partial-match search. Now,
it's physically impossible for that fraction to exceed one, but our
estimate of partialEntries is mostly bunk, and our estimate of numEntries
isn't exactly gospel either, so we could arrive at a silly value. In the
example presented by Andy we were coming out with a value of 100, leading
to insane cost estimates. Clamp the fraction to one to avoid that.
Like the previous patch, back-patch to all supported branches; this
problem can be demonstrated in one form or another in all of them.
Recovery does not achieve its goal of zeroing all pg_multixact entries
whose accompanying WAL records never reached disk. Remove that claim
and justify its expendability. Detail the need for TrimMultiXact(),
which has little in common with the TrimCLOG() rationale. Merge two
tightly-related comments. Stop presenting pg_multixact as specific to
heap_lock_tuple(); PostgreSQL 9.3 extended its use to heap_update().
Noticed while investigating a report from Andres Freund.
Fix an oversight in commit 321eed5f0f7563a0: replacing an operator's
selectivity functions needs to result in a corresponding update in
pg_depend. We have a function that can handle that, but it was not
called by AlterOperator().
To fix this without enlarging pg_operator.h's #include list beyond
what clients can safely include, split off the function definitions
into a new file pg_operator_fn.h, similarly to what we've done for
some other catalog header files. It's not entirely clear whether
any client-side code needs to include pg_operator.h, but it seems
prudent to assume that there is some such code somewhere.
In the extreme edge case where contended pages are the only ones that
escape being scanned, the previous commit would have allowed us to think
that relfrozenxid could be advanced, which is exactly wrong.
VACUUM can skip heap pages altogether when there's a run of consecutive
pages that are all-visible according to the visibility map. This causes it
to not update its nonempty_pages count, just as if those pages were empty,
which means that at the end we will think they are candidates for deletion.
Thus, we may take the table's AccessExclusive lock only to find that no
pages are really truncatable. This usually causes no real problems on a
master server, thanks to the lock being acquired only conditionally; but on
hot-standby servers, the same lock must be acquired unconditionally which
can result in unnecessary query cancellations.
To improve matters, force examination of the table's last page whenever
we reach there with a nonempty_pages count that would allow a truncation
attempt. If it's not empty, we'll advance nonempty_pages and thereby
prevent the truncation attempt.
If we are unable to acquire cleanup lock on that page, there's no need to
force it, unless we're doing an anti-wraparound vacuum. We can just check
for tuples with a shared buffer lock and then give up. (When we are doing
an anti-wraparound vacuum, and decide it's okay to skip the page because it
contains no freezable tuples, this patch still improves matters because
nonempty_pages is properly updated, which it was not before.)
Since only the last page is special-cased in this way, we might attempt a
truncation that will release many fewer pages than the normal heuristic
would suggest; at worst, only one page would be truncated. But that seems
all right, because the situation won't repeat during the next vacuum.
The real problem with the old logic is that the useless truncation attempt
happens every time we vacuum, so long as the state of the last few dozen
pages doesn't change.
This is a longstanding deficiency, but since the consequences aren't very
severe in most scenarios, I'm not going to risk a back-patch.
Jeff Janes and Tom Lane
The rationale for the way targetlist processing is done wasn't clearly
stated anywhere, and I for one had forgotten some of the details. Having
just painfully re-learned them, add some breadcrumbs for the next person.
Commit 6f8cb1e234 tried to centralize rewriteTargetView's copying
of a target view's Query struct. However, it ignored the fact that the
jointree->quals field was used twice. This only accidentally failed to
fail immediately because the same ChangeVarNodes mutation is applied in
both cases, so that we end up with logically identical expression trees
for both uses (and, as the code stands, the second ChangeVarNodes call
actually does nothing). However, we end up linking *physically*
identical expression trees into both an RTE's securityQuals list and
the WithCheckOption list. That's pretty dangerous, mainly because
prepsecurity.c is utterly cavalier about further munging such structures
without copying them first.
There may be no live bug in HEAD as a consequence of the fact that we apply
preprocess_expression in between here and prepsecurity.c, and that will
make a copy of the tree anyway. Or it may just be that the regression
tests happen to not trip over it. (I noticed this only because things
fell over pretty badly when I tried to relocate the planner's call of
expand_security_quals to before expression preprocessing.) In any case
it's very fragile because if anyone tried to make the securityQuals and
WithCheckOption trees diverge before we reach preprocess_expression, it
would not work. The fact that the current code will preprocess
securityQuals and WithCheckOptions lists at completely different times in
different query levels does nothing to increase my trust that that can't
happen.
In view of the fact that 9.5.0 is almost upon us and the aforesaid commit
has seen exactly zero field testing, the prudent course is to make an extra
copy of the quals so that the behavior is not different from what has been
in the field during beta.
The variables newestCommitTs and oldestCommitTs sound as if they are
timestamps, but in fact they are the transaction Ids that correspond
to the newest and oldest timestamps rather than the actual timestamps.
Rename these variables to reflect that they are actually xids: to wit
newestCommitTsXid and oldestCommitTsXid respectively. Also modify
related code in a similar fashion, particularly the user facing output
emitted by pg_controldata and pg_resetxlog.
Complaint and patch by me, review by Tom Lane and Alvaro Herrera.
Backpatch to 9.5 where these variables were first introduced.
MergeAttributes() rejects cases where columns to be merged have the same
type but different typmod, which is correct; but the error message it
printed didn't show either typmod, which is unhelpful. Changing this
requires using format_type_with_typemod() in place of TypeNameToString(),
which will have some minor side effects on the way some type names are
printed, but on balance this is an improvement: the old code sometimes
printed one type according to one set of rules and the other type according
to the other set, which could be confusing in its own way.
Oddly, there were no regression test cases covering any of this behavior,
so add some.
Complaint and fix by Amit Langote
brin_summarize_new_values() did not check that the passed OID was for
an index at all, much less that it was a BRIN index, and would fail in
obscure ways if it wasn't (possibly damaging data first?). It also
lacked any permissions test; by analogy to VACUUM, we should only allow
the table's owner to summarize.
Noted by Jeff Janes, fix by Michael Paquier and me
The original coding read tuples from workers in round-robin fashion,
but performance testing shows that it works much better to read enough
to empty one queue before moving on to the next. I believe the
reason for this is that, with the old approach, we could easily wake
up a worker repeatedly to write only one new tuple into the shm_mq
each time. With this approach, by the time the process gets scheduled,
it has a decent chance of being able to fill the entire buffer in
one go.
Patch by me. Dilip Kumar helped with performance testing.
This should have been part of the original commit, but was missed.
Pushing data between processes is expensive, so we definitely want
to project away unneeded columns here, just as we do for other nodes
like Sort and Hash that care about the volume of data.
Omitted boundaries represent the upper or lower limit of the corresponding
array subscript. This allows simpler specification of many common
use-cases.
(Revised version of commit 9246af6799)
YUriy Zhuravlev
When use_remote_estimate is enabled, consider adding ORDER BY to the
query we sending to the remote server so that we can use that ordered
data for a merge join. Commit f18c944b61
arranges to push down the query pathkeys, which seems like the case
mostly likely to be a win, but testing shows this can sometimes win,
too.
For a regular table, we know which indexes are present and therefore
test whether the ordering provided by each such index is useful. Here,
we take the opposite approach: guess what orderings would be useful if
they could be generated cheaply, and then ask the remote side what those
will cost.
Ashutosh Bapat, with very substantial cosmetic revisions by me. Also
reviewed by Rushabh Lathia.
Rather than expect the Query returned by get_view_query() to be
read-only and then copy bits and pieces of it out, simply copy the
entire structure when we get it. This addresses an issue where
AcquireRewriteLocks, which is called by acquireLocksOnSubLinks(),
scribbles on the parsetree passed in, which was actually an entry
in relcache, leading to segfaults with certain view definitions.
This also future-proofs us a bit for anyone adding more code to this
path.
The acquireLocksOnSubLinks() was added in commit c3e0ddd40.
Back-patch to 9.3 as that commit was.
Previously, each subroutine in initdb fired up its own standalone backend
session. Over time we'd grown as many as fifteen of these sessions,
and the cumulative startup and shutdown work for them was getting pretty
noticeable. Combining things so that all these steps share a single
backend session cuts a good 10% off the total runtime of initdb, more
if you're not fsync'ing.
The main stumbling block to doing this before was that some of the sessions
were run with -j and some not. The improved definition of -j mode
implemented by my previous commit makes it possible to fix that by running
all the post-bootstrap steps with -j; we just have to use double instead of
single newlines to end command strings. (This is only absolutely necessary
around the VACUUM and CREATE DATABASE steps, since those can't be run in a
transaction block. But it seems best to make them all use double newlines
so that the commands remain separate for error-reporting purposes.)
A minor disadvantage is that since initdb can't tell how much of its
output the backend has executed, we can no longer have the per-step
progress reporting initdb used to print. But things are fast enough
nowadays that that's not really all that useful anyway.
In passing, add more const decoration to some of the static arrays in
initdb.c.
Previously, -j caused the entire input file to be read in and executed as
a single command string. That's undesirable, not least because any error
causes the entire file to be regurgitated as the "failing query". Some
experimentation suggests a better rule: end the command string when we see
a semicolon immediately followed by two newlines, ie, an empty line after
a query. This serves nicely to break up the existing examples such as
information_schema.sql and system_views.sql. A limitation is that it's
no longer possible to write such a sequence within a string literal or
multiline comment in a file meant to be read with -j; but there are no
instances of such a problem within the data currently used by initdb.
(If someone does make such a mistake in future, it'll be obvious because
they'll get an unterminated-literal or unterminated-comment syntax error.)
Other than that, there shouldn't be any negative consequences; you're not
forced to end statements that way, it's just a better idea in most cases.
In passing, remove src/include/tcop/tcopdebug.h, which is dead code
because it's not included anywhere, and hasn't been for more than
ten years. One of the debug-support symbols it purported to describe
has been unreferenced for at least the same amount of time, and the
other is removed by this commit on the grounds that it was useless:
forcing -j mode all the time would have broken initdb. The lack of
complaints about that, or about the missing inclusion, shows that
no one has tried to use TCOP_DONTUSENEWLINE in many years.
This is necessary so that REASSIGN OWNED does the right thing with
composite types, to wit, that it also alters ownership of the type's
pg_class entry -- previously, the pg_class entry remained owned by the
original user, which caused later other failures such as the new owner's
inability to use ALTER TYPE to rename an attribute of the affected
composite. Also, if the original owner is later dropped, the pg_class
entry becomes owned by a non-existant user which is bogus.
To fix, create a new routine AlterTypeOwner_oid which knows whether to
pass the request to ATExecChangeOwner or deal with it directly, and use
that in shdepReassignOwner rather than calling AlterTypeOwnerInternal
directly. AlterTypeOwnerInternal is now simpler in that it only
modifies the pg_type entry and recurses to handle a possible array type;
higher-level tasks are handled by either AlterTypeOwner directly or
AlterTypeOwner_oid.
I took the opportunity to add a few more objects to the test rig for
REASSIGN OWNED, so that more cases are exercised. Additional ones could
be added for superuser-only-ownable objects (such as FDWs and event
triggers) but I didn't want to push my luck by adding a new superuser to
the tests on a backpatchable bug fix.
Per bug #13666 reported by Chris Pacejo.
Backpatch to 9.5.
(I would back-patch this all the way back, except that it doesn't apply
cleanly in 9.4 and earlier because 59367fdf9 wasn't backpatched. If we
decide that we need this in earlier branches too, we should backpatch
both.)
Encode TIDs as 64-bit integers to speed up comparisons. This seems to
speed things up on all platforms, but is even more beneficial when
8-byte integers are passed by value.
Peter Geoghegan. Design suggestions and review by Tom Lane. Review
also by Simon Riggs and by me.
FOREIGN KEY constraints have behaved this way for a long time, but for
some reason the behavior of CHECK constraints has been inconsistent up
until now.
Amit Langote and Amul Sul, with assorted tweaks by me.
It's entirely surprising that mdnblocks() has the side effect of
creating new files on disk, so let's make it not do that. One
consequence of the old behavior is that, if running on a damaged
cluster that is missing a file, mdnblocks() can recreate the file
and allow a subsequent _mdfd_getseg() for a higher segment to succeed.
This happens because, while mdnblocks() stops when it finds a segment
that is shorter than 1GB, _mdfd_getseg() has no such check, and thus
the empty file created by mdnblocks() can allow it to continue its
traversal and find higher-numbered segments which remain.
It might be a good idea for _mdfd_getseg() to actually verify that
each segment it finds is exactly 1GB before proceeding to the next
one, but that would involve some additional system calls, so for
now I'm just doing this much.
Patch by me, per off-list analysis by Kevin Grittner and Rahila Syed.
Review by Andres Freund.
Move the content lock directly into the BufferDesc, so that locking and
pinning a buffer touches only one cache line rather than two. Adjust
the definition of BufferDesc slightly so that this doesn't make the
BufferDesc any larger than one cache line (at least on platforms where
a spinlock is only 1 or 2 bytes).
We can't fit the I/O locks into the BufferDesc and stay within one
cache line, so move those to a completely separate tranche. This
leaves a relatively limited number of LWLocks in the main tranche, so
increase the padding of those remaining locks to a full cache line,
rather than allowing adjacent locks to share a cache line, hopefully
reducing false sharing.
Performance testing shows that these changes make little difference
on laptop-class machines, but help significantly on larger servers,
especially those with more than 2 sockets.
Andres Freund, originally based on an earlier patch by Simon Riggs.
Review and cosmetic adjustments (including heavy rewriting of the
comments) by me.
It's a bit cumbersome to use LWLockNewTrancheId(), because the returned
value needs to be shared between backends so that each backend can call
LWLockRegisterTranche() with the correct ID. So, for built-in tranches,
use a hard-coded value instead.
This is motivated by an upcoming patch adding further built-in tranches.
Andres Freund and Robert Haas
We carry around information about if a given query has row security or
not to allow the plancache to use that information to invalidate a
planned query in the event that the environment changes.
Previously, the flag of one of the subqueries was simply being copied
into place to indicate if the query overall included RLS components.
That's wrong as we need the global OR of all subqueries. Fix by
changing the code to match how fireRIRules works, which is results
in OR'ing all of the flags.
Noted by Tom.
Back-patch to 9.5 where RLS was introduced.
Apparently, there are bugs in this code that cause it to loop endlessly.
That bug still needs more research, but in the meantime it's clear that
the loop is missing a check for interrupts so that it can be cancelled
timely.
Backpatch to 9.1 -- this has been missing since 49475aab8d.
Previously, if find_multixact_start() failed, SetOffsetVacuumLimit() would
install 0 into MultiXactState->offsetStopLimit if it previously succeeded.
Luckily, there are no known cases where find_multixact_start() will return
an error in 9.5 and above. But if it were to happen, for example due to
filesystem permission issues, it'd be somewhat bad: GetNewMultiXactId()
could continue allocating mxids even if close to a wraparound, or it could
erroneously stop allocating mxids, even if no wraparound is looming. The
wrong value would be corrected the next time SetOffsetVacuumLimit() is
called, or by a restart.
Reported-By: Noah Misch, although this is not his preferred fix
Discussion: 20151210140450.GA22278@alap3.anarazel.de
Backpatch: 9.5, where the bug was introduced as part of 4f627f
e3f4cfc7 introduced a LWLockHeldByMe() call, without the corresponding
Assert() surrounding it.
Spotted by Coverity.
Backpatch: 9.1+, like the previous commit
Previously the "sent" field would be set to 0 and all other xlog
pointers be set to NULL if there were no valid values (such as when
in a backup sending walsender).
These would leak random xlog positions if a walsender used for backup would
a walsender slot previously used by a replication walsender.
In passing also fix a couple of cases where the xlog pointer is directly
compared to zero instead of using XLogRecPtrIsInvalid, noted by
Michael Paquier.
Changing the tablespace of an unlogged relation did not WAL log the
creation and content of the init fork. Thus, after a standby is
promoted, unlogged relation cannot be accessed anymore, with errors
like:
ERROR: 58P01: could not open file "pg_tblspc/...": No such file or directory
Additionally the init fork was not synced to disk, independent of the
configured wal_level, a relatively small durability risk.
Investigation of that problem also brought to light that, even for
permanent relations, the creation of !main forks was not WAL logged,
i.e. no XLOG_SMGR_CREATE record were emitted. That mostly turns out not
to be a problem, because these files were created when the actual
relation data is copied; nonexistent files are not treated as an error
condition during replay. But that doesn't work for empty files, and
generally feels a bit haphazard. Luckily, outside init and main forks,
empty forks don't occur often or are not a problem.
Add the required WAL logging and syncing to disk.
Reported-By: Michael Paquier
Author: Michael Paquier and Andres Freund
Discussion: 20151210163230.GA11331@alap3.anarazel.de
Backpatch: 9.1, where unlogged relations were introduced
As reported in bug #13809 by Alexander Ashurkov, the code for REASSIGN
OWNED hadn't gotten word about user mappings. Deal with them in the
same way default ACLs do, which is to ignore them altogether; they are
handled just fine by DROP OWNED. The other foreign object cases are
already handled correctly by both commands.
Also add a REASSIGN OWNED statement to foreign_data test to exercise the
foreign data objects. (The changes are just before the "cleanup" phase,
so it shouldn't remove any existing live test.)
Reported by Alexander Ashurkov, then independently by Jaime Casanova.
DROP OWNED BY handled GRANT-based ACLs but was not removing roles from
policies. Fix that by having DROP OWNED BY remove the role specified
from the list of roles the policy (or policies) apply to, or the entire
policy (or policies) if it only applied to the role specified.
As with ACLs, the DROP OWNED BY caller must have permission to modify
the policy or a WARNING is thrown and no change is made to the policy.
I originally modeled this data structure on SpecialJoinInfo, but after
commit acfcd45cac that looks like a pretty poor decision.
All we really need is relid sets identifying laterally-referenced rels;
and most of the time, what we want to know about includes indirect lateral
references, a case the LateralJoinInfo data was unsuited to compute with
any efficiency. The previous commit redefined RelOptInfo.lateral_relids
as the transitive closure of lateral references, so that it easily supports
checking indirect references. For the places where we really do want just
direct references, add a new RelOptInfo field direct_lateral_relids, which
is easily set up as a copy of lateral_relids before we perform the
transitive closure calculation. Then we can just drop lateral_info_list
and LateralJoinInfo and the supporting code. This makes the planner's
handling of lateral references noticeably more efficient, and shorter too.
Such a change can't be back-patched into stable branches for fear of
breaking extensions that might be looking at the planner's data structures;
but it seems not too late to push it into 9.5, so I've done so.
ALTER POLICY hadn't fully considered partial policy alternation
(eg: change just the roles on the policy, or just change one of
the expressions) when rebuilding the dependencies. Instead, it
would happily remove all dependencies which existed for the
policy and then only recreate the dependencies for the objects
referred to in the specific ALTER POLICY command.
Correct that by extracting and building the dependencies for all
objects referenced by the policy, regardless of if they were
provided as part of the ALTER POLICY command or were already in
place as part of the pre-existing policy.
More fuzz testing by Andreas Seltenreich exposed that the planner did not
cope well with chains of lateral references. If relation X references Y
laterally, and Y references Z laterally, then we will have to scan X on the
inside of a nestloop with Z, so for all intents and purposes X is laterally
dependent on Z too. The planner did not understand this and would generate
intermediate joins that could not be used. While that was usually harmless
except for wasting some planning cycles, under the right circumstances it
would lead to "failed to build any N-way joins" or "could not devise a
query plan" planner failures.
To fix that, convert the existing per-relation lateral_relids and
lateral_referencers relid sets into their transitive closures; that is,
they now show all relations on which a rel is directly or indirectly
laterally dependent. This not only fixes the chained-reference problem
but allows some of the relevant tests to be made substantially simpler
and faster, since they can be reduced to simple bitmap manipulations
instead of searches of the LateralJoinInfo list.
Also, when a PlaceHolderVar that is due to be evaluated at a join contains
lateral references, we should treat those references as indirect lateral
dependencies of each of the join's base relations. This prevents us from
trying to join any individual base relations to the lateral reference
source before the join is formed, which again cannot work.
Andreas' testing also exposed another oversight in the "dangerous
PlaceHolderVar" test added in commit 85e5e222b1. Simply rejecting
unsafe join paths in joinpath.c is insufficient, because in some cases
we will end up rejecting *all* possible paths for a particular join, again
leading to "could not devise a query plan" failures. The restriction has
to be known also to join_is_legal and its cohort functions, so that they
will not select a join for which that will happen. I chose to move the
supporting logic into joinrels.c where the latter functions are.
Back-patch to 9.3 where LATERAL support was introduced.
This module needs explicit initialization in order to replay WAL records
in recovery, but we had broken this recently following changes to make
other (stranger) scenarios work correctly. To fix, rework the
initialization sequence so that it always takes place before WAL replay
commences for both master and standby.
I could have gone for a more localized fix that just added a "startup"
call for the master server, but it seemed better to restructure the
existing callers as well so that the whole thing made more sense. As a
drawback, there is more control logic in xlog.c now than previously, but
doing otherwise meant passing down the ControlFile flag, which seemed
uglier as a whole.
This also meant adding a check to not re-execute ActivateCommitTs if it
had already been called.
Reported by Fujii Masao.
Backpatch to 9.5.
ExecOnConflictUpdate() passed t_ctid of the to-be-updated tuple to
ExecUpdate(). That's problematic primarily because of two reason: First
and foremost t_ctid could point to a different tuple. Secondly, and
that's what triggered the complaint by Stanislav, t_ctid is changed by
heap_update() to point to the new tuple version. The behavior of AFTER
UPDATE triggers was therefore broken, with NEW.* and OLD.* tuples
spuriously identical within AFTER UPDATE triggers.
To fix both issues, pass a pointer to t_self of a on-stack HeapTuple
instead.
Fixing this bug lead to one change in regression tests, which previously
failed due to the first issue mentioned above. There's a reasonable
expectation that test fails, as it updates one row repeatedly within one
INSERT ... ON CONFLICT statement. That is only possible if the second
update is triggered via ON CONFLICT ... SET, ON CONFLICT ... WHERE, or
by a WITH CHECK expression, as those are executed after
ExecOnConflictUpdate() does a visibility check. That could easily be
prohibited, but given it's allowed for plain UPDATEs and a rare corner
case, it doesn't seem worthwhile.
Reported-By: Stanislav Grozev
Author: Andres Freund and Peter Geoghegan
Discussion: CAA78GVqy1+LisN-8DygekD_Ldfy=BJLarSpjGhytOsgkpMavfQ@mail.gmail.com
Backpatch: 9.5, where ON CONFLICT was introduced
At the end of crash recovery, unlogged relations are reset to the empty
state, using their init fork as the template. The init fork is copied to
the main fork without going through shared buffers. Unfortunately WAL
replay so far has not necessarily flushed writes from shared buffers to
disk at that point. In normal crash recovery, and before the
introduction of 'fast promotions' in fd4ced523 / 9.3, the
END_OF_RECOVERY checkpoint flushes the buffers out in time. But with
fast promotions that's not the case anymore.
To fix, force WAL writes targeting the init fork to be flushed
immediately (using the new FlushOneBuffer() function). In 9.5+ that
flush can centrally be triggered from the code dealing with restoring
full page writes (XLogReadBufferForRedoExtended), in earlier releases
that responsibility is in the hands of XLOG_HEAP_NEWPAGE's replay
function.
Backpatch to 9.1, even if this currently is only known to trigger in
9.3+. Flushing earlier is more robust, and it is advantageous to keep
the branches similar.
Typical symptoms of this bug are errors like
'ERROR: index "..." contains unexpected zero page at block 0'
shortly after promoting a node.
Reported-By: Thom Brown
Author: Andres Freund and Michael Paquier
Discussion: 20150326175024.GJ451@alap3.anarazel.de
Backpatch: 9.1-
The original parallel sequential scan commit included only very limited
changes to the EXPLAIN output. Aggregated totals from all workers were
displayed, but there was no way to see what each individual worker did
or to distinguish the effort made by the workers from the effort made by
the leader.
Per a gripe by Thom Brown (and maybe others). Patch by me, reviewed
by Amit Kapila.
The single linked list of memory contexts could result in O(N^2)
performance to free a set of contexts if they were not freed in
reverse order of creation. In many cases the reverse order was
used, but there were some significant exceptions that caused real-
world performance problems. Rather than requiring all callers to
care about the order in which contexts were freed, and hunting down
and changing all existing cases where the wrong order was used, we
add one pointer per memory context so that the implementation
details are not so visible.
Jan Wieck
Commit e7cb7ee145 provided basic
infrastructure for allowing a foreign data wrapper or custom scan
provider to replace a join of one or more tables with a scan.
However, this infrastructure failed to take into account the need
for possible EvalPlanQual rechecks, and ExecScanFetch would fail
an assertion (or just overwrite memory) if such a check was attempted
for a plan containing a pushed-down join. To fix, adjust the EPQ
machinery to skip some processing steps when scanrelid == 0, making
those the responsibility of scan's recheck method, which also has
the responsibility in this case of correctly populating the relevant
slot.
To allow foreign scans to gain control in the right place to make
use of this new facility, add a new, optional RecheckForeignScan
method. Also, allow a foreign scan to have a child plan, which can
be used to correctly populate the slot (or perhaps for something
else, but this is the only use currently envisioned).
KaiGai Kohei, reviewed by Robert Haas, Etsuro Fujita, and Kyotaro
Horiguchi.
While convincing myself that commit 7e19db0c09 would solve both of
the problems recently reported by Andreas Seltenreich, I realized that
add_paths_to_joinrel's handling of LATERAL restrictions could be made
noticeably simpler and faster if we were to retain the minimum possible
parameterization for each joinrel (that is, the set of relids supplying
unsatisfied lateral references in it). We already retain that for
baserels, in RelOptInfo.lateral_relids, so we can use that field for
joinrels too.
I re-pgindent'd the files touched here, which affects some unrelated
comments.
This is, I believe, just a minor optimization not a bug fix, so no
back-patch.
It was possible for the planner to decide to join a LATERAL subquery to
the outer side of an outer join before the outer join itself is completed.
Normally that's fine because of the associativity rules, but it doesn't
work if the subquery contains a lateral reference to the inner side of the
outer join. In such a situation the outer join *must* be done first.
join_is_legal() missed this consideration and would allow the join to be
attempted, but the actual path-building code correctly decided that no
valid join path could be made, sometimes leading to planner errors such as
"failed to build any N-way joins".
Per report from Andreas Seltenreich. Back-patch to 9.3 where LATERAL
support was added.
As pointed out by Fujii Masao, we weren't quite there on a standby
behaving sanely: first because we were failing to acquire the correct
state in the case where no XLOG_PARAMETER_CHANGE message was sent
(because a checkpoint had already happened after the setting was changed
in the master, and then the standby was restarted); and second because
promoting the standby with the feature enabled failed to activate it if
the master had the feature disabled.
This patch fixes both those misbehaviors hopefully without
re-introducing any old problems.
Also change the hint emitted in a standby together with the error
message about the feature being disabled, to make it point out that the
place to chance the setting is the master. Otherwise, if the setting is
already enabled in the standby, it is very confusing to have it say that
the setting must be enabled ...
Authors: Álvaro Herrera, Petr Jelínek.
Backpatch to 9.5.
We tried to fetch statistics data from the index metapage, which does not
work if the index isn't actually present. If the index is hypothetical,
instead extrapolate some plausible internal statistics based on the index
page count provided by the index-advisor plugin.
There was already some code in gincostestimate() to invent internal stats
in this way, but since it was only meant as a stopgap for pre-9.1 GIN
indexes that hadn't been vacuumed since upgrading, it was pretty crude.
If we want it to support index advisors, we should try a little harder.
A small amount of testing says that it's better to estimate the entry pages
as 90% of the index, not 100%. Also, estimating the number of entries
(keys) as equal to the heap tuple count could be wildly wrong in either
direction. Instead, let's estimate 100 entries per entry page.
Perhaps someday somebody will want the index advisor to be able to provide
these numbers more directly, but for the moment this should serve.
Problem report and initial patch by Julien Rouhaud; modified by me to
invent less-bogus internal statistics. Back-patch to all supported
branches, since we've supported index advisors since 9.0.
In commit 8abb3cda0d I attempted to cache
the expression state trees constructed for domain CHECK constraints for
the life of the backend (assuming the domain's constraints don't get
redefined). However, this turns out not to work very well, because
execQual.c will run those state trees with ecxt_per_query_memory pointing
to a query-lifespan context, and in some situations we'll end up with
pointers into that context getting stored into the state trees. This
happens in particular with SQL-language functions, as reported by
Emre Hasegeli, but there are many other cases.
To fix, keep only the expression plan trees for domain CHECK constraints
in the typcache's data structure, and revert to performing ExecInitExpr
(at least) once per query to set up expression state trees in the query's
context.
Eventually it'd be nice to undo this, but that will require some careful
thought about memory management for expression state trees, and it seems
far too late for any such redesign in 9.5. This way is still much more
efficient than what happened before 8abb3cda0.
Previously, we did many conversions for Cyrillic and Central European
single-byte encodings by converting to a related MULE_INTERNAL coding
scheme before converting to the destination. This seems unnecessarily
inefficient. Moreover, if the conversion encounters an untranslatable
character, the error message will confusingly complain about failure
to convert to or from MULE_INTERNAL, rather than the user-visible
encodings. Worse still, this approach results in some completely
unnecessary conversion failures; there are cases where the chosen
MULE subset lacks characters that exist in both of the user-visible
encodings, causing a conversion failure that need not occur.
This patch fixes the first two of those deficiencies by introducing
a new local2local() conversion support subroutine for direct conversion
between any two single-byte character sets, and adding new conversion
tables where needed. However, I generated the new conversion tables by
testing PG 9.5's behavior, so that the actual conversion behavior is
bug-compatible with previous releases; the only user-visible behavior
change is that the error messages for conversion failures are saner.
Changes in the conversion behavior will probably ensue after discussion.
Interestingly, although this approach requires more tables, the .so files
actually end up smaller (at least on my x86_64 machine); the tables are
smaller than the management code needed for double conversion.
Per a complaint from Albe Laurenz.
Some of the Unicode/*.map files had identification comments added to them,
evidently by hand. Others did not. Modify the generating scripts to
produce these comments automatically, and update the generated files that
lacked them.
This is just minor cleanup as a by-product of trying to verify that the
*.map files can indeed be reproduced from authoritative data. There are a
depressingly large number that fail to reproduce from the claimed sources.
I have not touched those in this commit, except for the JIS 2004-related
files which required only a single comment update to match.
Since this only affects comments, no need to consider a back-patch.
Failure to initially palloc the comboCids array, or to realloc it bigger
when needed, left combocid's data structures in an inconsistent state that
would cause trouble if the top transaction continues to execute. Noted
while examining a user complaint about the amount of memory used for this.
(There's not much we can do about that, but it does point up that repalloc
failure has a non-negligible chance of occurring here.)
In HEAD/9.5, also avoid possible invocation of memcpy() with a null pointer
in SerializeComboCIDState; cf commit 13bba0227.
The integer overflow situation in div_var_fast() is a great deal more
complicated than the pre-existing comments would suggest. Moreover, the
comments were also flat out incorrect as to the precise statement of the
maxdiv loop invariant. Upon clarifying that, it becomes apparent that the
way in which we updated maxdiv after a carry propagation pass was overly
slow, complex, and conservative: we can just reset it to one, which is much
easier and also reduces the number of times carry propagation occurs.
Fix that and improve the relevant comments.
Since this is mostly a comment fix, with only a rather marginal performance
boost, no need for back-patch.
Tom Lane and Dean Rasheed
The POSIX standard for tar headers requires archive member sizes to be
printed in octal with at most 11 digits, limiting the representable file
size to 8GB. However, GNU tar and apparently most other modern tars
support a convention in which oversized values can be stored in base-256,
allowing any practical file to be a tar member. Adopt this convention
to remove two limitations:
* pg_dump with -Ft output format failed if the contents of any one table
exceeded 8GB.
* pg_basebackup failed if the data directory contained any file exceeding
8GB. (This would be a fatal problem for installations configured with a
table segment size of 8GB or more, and it has also been seen to fail when
large core dump files exist in the data directory.)
File sizes under 8GB are still printed in octal, so that no compatibility
issues are created except in cases that would have failed entirely before.
In addition, this patch fixes several bugs in the same area:
* In 9.3 and later, we'd defined tarCreateHeader's file-size argument as
size_t, which meant that on 32-bit machines it would write a corrupt tar
header for file sizes between 4GB and 8GB, even though no error was raised.
This broke both "pg_dump -Ft" and pg_basebackup for such cases.
* pg_restore from a tar archive would fail on tables of size between 4GB
and 8GB, on machines where either "size_t" or "unsigned long" is 32 bits.
This happened even with an archive file not affected by the previous bug.
* pg_basebackup would fail if there were files of size between 4GB and 8GB,
even on 64-bit machines.
* In 9.3 and later, "pg_basebackup -Ft" failed entirely, for any file size,
on 64-bit big-endian machines.
In view of these potential data-loss bugs, back-patch to all supported
branches, even though removal of the documented 8GB limit might otherwise
be considered a new feature rather than a bug fix.
The previous way of reconstructing check constraints was to do a separate
"ALTER TABLE ONLY tab ADD CONSTRAINT" for each table in an inheritance
hierarchy. However, that way has no hope of reconstructing the check
constraints' own inheritance properties correctly, as pointed out in
bug #13779 from Jan Dirk Zijlstra. What we should do instead is to do
a regular "ALTER TABLE", allowing recursion, at the topmost table that
has a particular constraint, and then suppress the work queue entries
for inherited instances of the constraint.
Annoyingly, we'd tried to fix this behavior before, in commit 5ed6546cf,
but we failed to notice that it wasn't reconstructing the pg_constraint
field values correctly.
As long as I'm touching pg_get_constraintdef_worker anyway, tweak it to
always schema-qualify the target table name; this seems like useful backup
to the protections installed by commit 5f173040.
In HEAD/9.5, get rid of get_constraint_relation_oids, which is now unused.
(I could alternatively have modified it to also return conislocal, but that
seemed like a pretty single-purpose API, so let's not pretend it has some
other use.) It's unused in the back branches as well, but I left it in
place just in case some third-party code has decided to use it.
In HEAD/9.5, also rename pg_get_constraintdef_string to
pg_get_constraintdef_command, as the previous name did nothing to explain
what that entry point did differently from others (and its comment was
equally useless). Again, that change doesn't seem like material for
back-patching.
I did a bit of re-pgindenting in tablecmds.c in HEAD/9.5, as well.
Otherwise, back-patch to all supported branches.
The previous coding attempts to destroy the DSM in this case, but
child nodes might have stored data there and still be holding onto
pointers in this case. So don't do that.
Also, free the reader array instead of leaking it.
Extracted from two different patch versions both by Amit Kapila.
This was already true for CREATE EXTENSION, but historically has not
been true for other object types. Therefore, this is a backward
incompatibility. Per discussion on pgsql-hackers, everyone seems to
agree that the new behavior is better.
Marti Raudsepp, reviewed by Haribabu Kommi and myself
At least one of the names was, due to a function renaming late in the
development of ON CONFLICT, wrong. Since including function names in
error messages is against the message style guide anyway, remove them
from the messages.
Discussion: CAM3SWZT8paz=usgMVHm0XOETkQvzjRtAUthATnmaHQQY0obnGw@mail.gmail.com
Backpatch: 9.5, where ON CONFLICT was introduced
When I wrote this code originally, the intention was to recompute the
remapinfo only when the tupledesc changes. This presumably only
happens once per query, but I copied the design pattern from other
DestReceivers. However, due to a silly oversight on my part,
tqueue->tupledesc never got set, leading to recomputation for every
tuple.
This should improve the performance of parallel scans that return a
significant number of tuples.
Report by Amit Kapila; patch by me, reviewed by him.
div_var_fast() postpones propagating carries in the same way as mul_var(),
so it has the same corner-case overflow risk we fixed in 246693e5ae,
namely that the size of the carries has to be accounted for when setting
the threshold for executing a carry propagation step. We've not devised
a test case illustrating the brokenness, but the required fix seems clear
enough. Like the previous fix, back-patch to all active branches.
Dean Rasheed
Prior to commit 0709b7ee72, access to
variables within a spinlock-protected critical section had to be done
through a volatile pointer, but that should no longer be necessary.
Review by Andres Freund
Since commit 11e131854f, ruleutils.c has
attempted to ensure that each RTE in a query or plan tree has a unique
alias name. However, the code that was added for this could be quite slow,
even as bad as O(N^3) if N identical RTE names must be replaced, as noted
by Jeff Janes. Improve matters by building a transient hash table within
set_rtable_names. The hash table in itself reduces the cost of detecting a
duplicate from O(N) to O(1), and we can save another factor of N by storing
the number of de-duplicated names already created for each entry, so that
we don't have to re-try names already created. This way is probably a bit
slower overall for small range tables, but almost by definition, such cases
should not be a performance problem.
In principle the same problem applies to the column-name-de-duplication
code; but in practice that seems to be less of a problem, first because
N is limited since we don't support extremely wide tables, and second
because duplicate column names within an RTE are fairly rare, so that in
practice the cost is more like O(N^2) not O(N^3). It would be very much
messier to fix the column-name code, so for now I've left that alone.
An independent problem in the same area was that the de-duplication code
paid no attention to the identifier length limit, and would happily produce
identifiers that were longer than NAMEDATALEN and wouldn't be unique after
truncation to NAMEDATALEN. This could result in dump/reload failures, or
perhaps even views that silently behaved differently than before. We can
fix that by shortening the base name as needed. Fix it for both the
relation and column name cases.
In passing, check for interrupts in set_rtable_names, just in case it's
still slow enough to be an issue.
Back-patch to 9.3 where this code was introduced.
Normally ruleutils prints a whole-row Var as "foo.*". We already knew that
that doesn't work at top level of a SELECT list, because the parser would
treat the "*" as a directive to expand the reference into separate columns,
not a whole-row Var. However, Joshua Yanovski points out in bug #13776
that the same thing happens at top level of a ROW() construct; and some
nosing around in the parser shows that the same is true in VALUES().
Hence, apply the same workaround already devised for the SELECT-list case,
namely to add a forced cast to the appropriate rowtype in these cases.
(The alternative of just printing "foo" was rejected because it is
difficult to avoid ambiguity against plain columns named "foo".)
Back-patch to all supported branches.
Set the "rscales" for intermediate-result calculations to ensure that
suitable numbers of significant digits are maintained throughout. The
previous coding hadn't thought this through in any detail, and as a result
could deliver results with many inaccurate digits, or in the worst cases
even fail with divide-by-zero errors as a result of losing all nonzero
digits of intermediate results.
In exp_var(), get rid entirely of the logic that separated the calculation
into integer and fractional parts: that was neither accurate nor
particularly fast. The existing range-reduction method of dividing by 2^n
can be applied across the full input range instead of only 0..1, as long as
we are careful to set an appropriate rscale for each step.
Also fix the logic in mul_var() for shortening the calculation when the
caller asks for fewer output digits than an exact calculation would
require. This bug doesn't affect simple multiplications since that code
path asks for an exact result, but it does contribute to accuracy issues
in the transcendental math functions.
In passing, improve performance of mul_var() a bit by forcing the shorter
input to be on the left, thus reducing the number of iterations of the
outer loop and probably also reducing the number of carry-propagation
steps needed.
This is arguably a bug fix, but in view of the lack of field complaints,
it does not seem worth the risk of back-patching.
Dean Rasheed
This makes it significantly easier to identify these lwlocks in
LWLOCK_STATS or Trace_lwlocks output. It's also arguably better
from a modularity standpoint, since lwlock.c no longer needs to
know anything about the LWLock needs of the higher-level SLRU
facility.
Ildus Kurbangaliev, reviewd by Álvaro Herrera and by me.
Letting backends continue to run if the postmaster has exited prevents
PostgreSQL from being restarted, which in many environments is
catastrophic. Worse, if some other backend crashes, we no longer have
any protection against shared memory corruption. So, arrange for them
to exit instead. We don't want to expend many cycles on this, but
including postmaster death in the set of things that we wait for when
a backend is idle seems cheap enough.
Rajeev Rastogi and Robert Haas
Commit a0d9f6e434 added this support for
all other plan node types; this fills in the gap.
Since TextOutCustomScan complicates this and is pretty well useless,
remove it.
KaiGai Kohei, with some modifications by me.
In commit 5d1ff6bd55 I added some logic to
relcache.c to try to ensure that the regression tests would fail if we
made a mistake about which relations belong in the relcache init files.
I'm quite sure I tested that, but I must have done so only for the
non-shared-catalog case, because a report from Adam Brightwell showed that
the regression tests still pass just fine if we bollix the shared-catalog
init file in the way this code was supposed to catch. The reason is that
that file gets loaded before we do client authentication, so the WARNING
is not sent to the client, only to the postmaster log, where it's far too
easily missed.
The least Rube Goldbergian answer to this is to put an Assert(false)
after the elog(WARNING). That will certainly get developers' attention,
while not breaking production builds' ability to recover from corner
cases with similar symptoms.
Since this is only of interest to developers, there seems no need for
a back-patch, even though the previous commit went into all branches.
Add a new flag, consider_parallel, to each RelOptInfo, indicating
whether a plan for that relation could conceivably be run inside of
a parallel worker. Right now, we're pretty conservative: for example,
it might be possible to defer applying a parallel-restricted qual
in a worker, and later do it in the leader, but right now we just
don't try to parallelize access to that relation. That's probably
the right decision in most cases, anyway.
Using the new flag, generate parallel sequential scan plans for plain
baserels, meaning that we now have parallel sequential scan in
PostgreSQL. The logic here is pretty unsophisticated right now: the
costing model probably isn't right in detail, and we can't push joins
beneath Gather nodes, so the number of plans that can actually benefit
from this is pretty limited right now. Lots more work is needed.
Nevertheless, it seems time to enable this functionality so that all
this code can actually be tested easily by users and developers.
Note that, if you wish to test this functionality, it will be
necessary to set max_parallel_degree to a value greater than the
default of 0. Once a few more loose ends have been tidied up here, we
might want to consider changing the default value of this GUC, but
I'm leaving it alone for now.
Along the way, fix a bug in cost_gather: the previous coding thought
that a Gather node's transfer overhead should be costed on the basis of
the relation size rather than the number of tuples that actually need
to be passed off to the leader.
Patch by me, reviewed in earlier versions by Amit Kapila.
In addition, this path fills in a number of missing bits and pieces in
the parallel infrastructure. Paths and plans now have a parallel_aware
flag indicating whether whatever parallel-aware logic they have should
be engaged. It is believed that we will need this flag for a number of
path/plan types, not just sequential scans, which is why the flag is
generic rather than part of the SeqScan structures specifically.
Also, execParallel.c now gives parallel nodes a chance to initialize
their PlanState nodes from the DSM during parallel worker startup.
Amit Kapila, with a fair amount of adjustment by me. Review of previous
patch versions by Haribabu Kommi and others.
The patches committed as 6e71dd7ce9
and 3a1f8611f2 were developed in
parallel but dependent on each other in a way that I failed to
notice.
This patch to fix the problem was prepared by Amit Kapila.
By accident the replication origin was not set properly in
DecodeCommit(). That's bad because the origin is passed to the output
plugins origin filter, and accessible from the output plugin via
ReorderBufferTXN->origin_id. Accessing the origin of individual changes
worked before the fix, which is why this wasn't notices earlier.
Reported-By: Craig Ringer
Author: Craig Ringer
Discussion: CAMsr+YFhBJLp=qfSz3-J+0P1zLkE8zNXM2otycn20QRMx380gw@mail.gmail.com
Backpatch: 9.5, where replication origins where introduced
Lookahead and lookbehind constraints aren't allowed to contain backrefs,
and parentheses within them are always considered non-capturing. Or so
says the manual. But the regexp parser forgot about these rules once
inside a parenthesized subexpression, so that constructs like (\w)(?=(\1))
were accepted (but then not correctly executed --- a case like this acted
like (\w)(?=\w), without any enforcement that the two \w's match the same
text). And in (?=((foo))) the innermost parentheses would be counted as
capturing parentheses, though no text would ever be captured for them.
To fix, properly pass down the "type" argument to the recursive invocation
of parse().
Back-patch to all supported branches; it was agreed that silent
misexecution of such patterns is worse than throwing an error, even though
new errors in minor releases are generally not desirable.
Commit 4a4e6893aa, which introduced this
mechanism, failed to account for the fact that the RECORD pseudo-type
uses transient typmods that are only meaningful within a single
backend. Transferring such tuples without modification between two
cooperating backends does not work. This commit installs a system
for passing the tuple descriptors over the same shm_mq being used to
send the tuples themselves. The two sides might not assign the same
transient typmod to any given tuple descriptor, so we must also
substitute the appropriate receiver-side typmod for the one used by
the sender. That adds some CPU overhead, but still seems better than
being unable to pass records between cooperating parallel processes.
Along the way, move the logic for handling multiple tuple queues from
tqueue.c to nodeGather.c; tqueue.c now provides a TupleQueueReader,
which reads from a single queue, rather than a TupleQueueFunnel, which
potentially reads from multiple queues. This change was suggested
previously as a way to make sure that nodeGather.c rather than tqueue.c
had policy control over the order in which to read from queues, but
it wasn't clear to me until now how good an idea it was. typmod
mapping needs to be performed separately for each queue, and it is
much simpler if the tqueue.c code handles that and leaves multiplexing
multiple queues to higher layers of the stack.
This introduces a simple encoding scheme to produce abbreviated keys:
pack as many bytes of each UUID as will fit into a Datum. On
little-endian machines, a byteswap is also performed; the abbreviated
comparator can therefore just consist of a simple 3-way unsigned integer
comparison.
The purpose of this change is to speed up sorting data on a column
of type UUID.
Peter Geoghegan
With include_realm=1 being set down in parse_hba_auth_opt, if multiple
options are passed on the pg_hba line, such as:
host all all 0.0.0.0/0 gss include_realm=0 krb_realm=XYZ.COM
We would mistakenly reset include_realm back to 1. Instead, we need to
set include_realm=1 up in parse_hba_line, prior to parsing any of the
additional options.
Discovered by Jeff McCormick during testing.
Bug introduced by 9a08841.
Back-patch to 9.5
Previously, negative values were always displayed in bytes, regardless
of how large they were.
Adrian Vondendriesch, reviewed by Julien Rouhaud and myself
The jsonb_path_ops code calculated hash values inconsistently in some cases
involving nested arrays and objects. This would result in queries possibly
not finding entries that they should find, when using a jsonb_path_ops GIN
index for the search. The problem cases involve JSONB values that contain
both scalars and sub-objects at the same nesting level, for example an
array containing both scalars and sub-arrays. To fix, reset the current
stack->hash after processing each value or sub-object, not before; and
don't try to be cute about the outermost level's initial hash.
Correcting this means that existing jsonb_path_ops indexes may now be
inconsistent with the new hash calculation code. The symptom is the same
--- searches not finding entries they should find --- but the specific
rows affected are likely to be different. Users will need to REINDEX
jsonb_path_ops indexes to make sure that all searches work as expected.
Per bug #13756 from Daniel Cheng. Back-patch to 9.4 where the faulty
logic was introduced.
Up until now, the total amount of data that could be passed to a
background worker at startup was one datum, which can be a small as
4 bytes on some systems. That's enough to pass a dsm_handle or an
array index, but not much else. Add a bgw_extra flag to the
BackgroundWorker struct, allowing up to 128 bytes to be passed to
a new worker on any platform.
Use this to fix a problem I recently discovered with the parallel
context machinery added in 9.5: the master assigns each worker an
array index, and each worker subsequently assigns itself an array
index, and there's nothing to guarantee that the two sets of indexes
match, leading to chaos.
Normally, I would not back-patch the change to add bgw_extra, since it
is basically a feature addition. However, since 9.5 is still in beta
and there seems to be no other sensible way to repair the broken
parallel context machinery, back-patch to 9.5. Existing background
worker code can ignore the bgw_extra field without a problem, but
might need to be recompiled since the structure size has changed.
Report and patch by me. Review by Amit Kapila.
Rather than filling a temporary array and then copying values to the
output array, we can generate the required random permutation in-place
using the Fisher-Yates shuffle algorithm. This is shorter as well as
more efficient than before. It's pretty unlikely that anyone would
notice a speed improvement, but shorter code is better.
Nathan Wagner, edited a bit by me
The user can whitelist specified extension(s) in the foreign server's
options, whereupon we will treat immutable functions and operators of those
extensions as candidates to be sent for remote execution.
Whitelisting an extension in this way basically promises that the extension
exists on the remote server and behaves compatibly with the local instance.
We have no way to prove that formally, so we have to rely on the user to
get it right. But this seems like something that people can usually get
right in practice.
We might in future allow functions and operators to be whitelisted
individually, but extension granularity is a very convenient special case,
so it got done first.
The patch as-committed lacks any regression tests, which is unfortunate,
but introducing dependencies on other extensions for testing purposes
would break "make installcheck" scenarios, which is worse. I have some
ideas about klugy ways around that, but it seems like material for a
separate patch. For the moment, leave the problem open.
Paul Ramsey, hacked up a bit more by me
Commit a1480ec1d3 purported to fix the
problems with commit b2ccb5f4e6, but it
didn't completely fix them. The problem is that the checks were
performed in the wrong order, leading to a race condition. If the
sender attached, sent a message, and detached after the receiver
called shm_mq_get_sender and before the receiver called
shm_mq_counterparty_gone, we'd incorrectly return SHM_MQ_DETACHED
before all messages were read. Repair by reversing the order of
operations, and add a long comment explaining why this new logic is
(hopefully) correct.
Commit d1b7c1ffe7 introduced a mechanism
for serializing a ParamListInfo structure to be passed to a parallel
worker. However, this mechanism failed to handle external expanded
values, as pointed out by Noah Misch. Repair.
Moreover, plpgsql_param_fetch requires adjustment because the
serialization mechanism needs it to skip evaluating unused parameters
just as we would do when it is called from copyParamList, but params
== estate->paramLI in that case. To fix, make the bms_is_member test
in that function unconditional.
Finally, have setup_param_list set a new ParamListInfo field,
paramMask, to the parameters actually used in the expression, so that
we don't try to fetch those that are not needed when serializing a
parameter list. This isn't necessary for correctness, but it makes
the performance of the parallel executor code comparable to what we
do for cases involving cursors.
Design suggestions and extensive review by Noah Misch. Patch by me.
On insert the CheckForSerializableConflictIn() test was performed
before the page(s) which were going to be modified had been locked
(with an exclusive buffer content lock). If another process
acquired a relation SIReadLock on the heap and scanned to a page on
which an insert was going to occur before the page was so locked,
a rw-conflict would be missed, which could allow a serialization
anomaly to be missed. The window between the check and the page
lock was small, so the bug was generally not noticed unless there
was high concurrency with multiple processes inserting into the
same table.
This was reported by Peter Bailis as bug #11732, by Sean Chittenden
as bug #13667, and by others.
The race condition was eliminated in heap_insert() by moving the
check down below the acquisition of the buffer lock, which had been
the very next statement. Because of the loop locking and unlocking
multiple buffers in heap_multi_insert() a check was added after all
inserts were completed. The check before the start of the inserts
was left because it might avoid a large amount of work to detect a
serialization anomaly before performing the all of the inserts and
the related WAL logging.
While investigating this bug, other SSI bugs which were even harder
to hit in practice were noticed and fixed, an unnecessary check
(covered by another check, so redundant) was removed from
heap_update(), and comments were improved.
Back-patch to all supported branches.
Kevin Grittner and Thomas Munro
A lookbehind constraint is like a lookahead constraint in that it consumes
no text; but it checks for existence (or nonexistence) of a match *ending*
at the current point in the string, rather than one *starting* at the
current point. This is a long-requested feature since it exists in many
other regex libraries, but Henry Spencer had never got around to
implementing it in the code we use.
Just making it work is actually pretty trivial; but naive copying of the
logic for lookahead constraints leads to code that often spends O(N^2) time
to scan an N-character string, because we have to run the match engine
from string start to the current probe point each time the constraint is
checked. In typical use-cases a lookbehind constraint will be written at
the start of the regex and hence will need to be checked at every character
--- so O(N^2) work overall. To fix that, I introduced a third copy of the
core DFA matching loop, paralleling the existing longest() and shortest()
loops. This version, matchuntil(), can suspend and resume matching given
a couple of pointers' worth of storage space. So we need only run it
across the string once, stopping at each interesting probe point and then
resuming to advance to the next one.
I also put in an optimization that simplifies one-character lookahead and
lookbehind constraints, such as "(?=x)" or "(?<!\w)", into AHEAD and BEHIND
constraints, which already existed in the engine. This avoids the overhead
of the LACON machinery entirely for these rather common cases.
The net result is that lookbehind constraints run a factor of three or so
slower than Perl's for multi-character constraints, but faster than Perl's
for one-character constraints ... and they work fine for variable-length
constraints, which Perl gives up on entirely. So that's not bad from a
competitive perspective, and there's room for further optimization if
anyone cares. (In reality, raw scan rate across a large input string is
probably not that big a deal for Postgres usage anyway; so I'm happy if
it's linear.)
Commit b0b0d84b3d purported to make it
possible to relaunch workers using the same parallel context, but it had
an unpleasant race condition: we might reinitialize after the workers
have sent their last control message but before they have dettached the
DSM, leaving to crashes. Repair by introducing a new ParallelContext
operation, ReinitializeParallelDSM.
Adjust execParallel.c to use this new support, so that we can rescan a
Gather node by relaunching workers but without needing to recreate the
DSM.
Amit Kapila, with some adjustments by me. Extracted from latest parallel
sequential scan patch.
The original Gather code failed to mark a Gather node as not able to
do projection, but it couldn't, even though it did call initialize its
projection info via ExecAssignProjectionInfo. There doesn't seem to
be any good reason for this node not to have projection capability,
so clean things up so that it does. Without this, plans using Gather
nodes might need to carry extra Result nodes to do projection.
A bug in the original free space computation made it possible to
return a page which wasn't actually able to fit the item. Since the
insertion code isn't prepared to deal with PageAddItem failing, a PANIC
resulted ("failed to add BRIN tuple [to new page]"). Add a macro to
encapsulate the correct computation, and use it in
brin_getinsertbuffer's callers before calling that routine, to raise an
early error.
I became aware of the possiblity of a problem in this area while working
on ccc4c07499. There's no archived discussion about it, but it's
easy to reproduce a problem in the unpatched code with something like
CREATE TABLE t (a text);
CREATE INDEX ti ON t USING brin (a) WITH (pages_per_range=1);
for length in `seq 8000 8196`
do
psql -f - <<EOF
TRUNCATE TABLE t;
INSERT INTO t VALUES ('z'), (repeat('a', $length));
EOF
done
Backpatch to 9.5, where BRIN was introduced.
Further tweak commit_ts.c so that on a standby the state is completely
consistent with what that in the master, rather than behaving
differently in the cases that the settings differ. Now in standby and
master the module should always be active or inactive in lockstep.
Author: Petr Jelínek, with some further tweaks by Álvaro Herrera.
Backpatch to 9.5, where commit timestamps were introduced.
Discussion: http://www.postgresql.org/message-id/5622BF9D.2010409@2ndquadrant.com
Bernd Helmle complained that CreateReplicationSlot() was assigning the
same value to the same variable twice, so we could remove one of them.
Code inspection reveals that we can actually remove both assignments:
according to the author the assignment was there for beauty of the
strlen line only, and another possible fix to that is to put the strlen
in its own line, so do that.
To be consistent within the file, refactor all duplicated strlen()
calls, which is what we do elsewhere in the backend anyway. In
basebackup.c, snprintf already returns the right length; no need for
strlen afterwards.
Backpatch to 9.4, where replication slots were introduced, to keep code
identical. Some of this is older, but the patch doesn't apply cleanly
and it's only of cosmetic value anyway.
Discussion: http://www.postgresql.org/message-id/BE2FD71DEA35A2287EA5F018@eje.credativ.lan
If the counterparty writes some data into the queue and then detaches,
it's wrong to return SHM_MQ_DETACHED right away. If we do that, we
fail to read whatever was written.
The shm_mq mechanism was intended to optionally notice when the process
on the other end of the queue fails to attach to the queue. It does
this by allowing the user to pass a BackgroundWorkerHandle; if the
background worker in question is launched and dies without attaching
to the queue, then we know it never will. This logic works OK in
blocking mode, but when called with nowait = true we fail to notice
that this has happened due to an asymmetry in the logic. Repair.
Reported off-list by Rushabh Lathia. Patch by me.
Once upon a time we did not have a separate CREATEROLE privilege, and
CREATEUSER effectively meant SUPERUSER. When we invented CREATEROLE
(in 8.1) we also added SUPERUSER so as to have a less confusing keyword
for this role property. However, we left CREATEUSER in place as a
deprecated synonym for SUPERUSER, because of backwards-compatibility
concerns. It's still there and is still confusing people, as for example
in bug #13694 from Justin Catterson. 9.6 will be ten years or so later,
which surely ought to be long enough to end the deprecation and just
remove these old keywords. Hence, do so.
Commit 816e336f12 added the wrong error
check to async.c; sending restrictions is restricted to the leader,
not altogether unsafe.
Commit 3bd909b220 added ExecShutdownNode
to traverse the planstate tree and call shutdown functions, but made
a Gather node, the only node that actually has such a function, abort
the tree traversal, which is wrong.
Commit bda76c1c8c caused both plus and
minus infinity to be rendered as "infinity", which is not only wrong
but inconsistent with the pre-9.4 behavior of to_json(). Fix that by
duplicating the coding in date_out/timestamp_out/timestamptz_out more
closely. Per bug #13687 from Stepan Perlov. Back-patch to 9.4, like
the previous commit.
In passing, also re-pgindent json.c, since it had gotten a bit messed up by
recent patches (and I was already annoyed by indentation-related problems
in back-patching this fix ...)
Per a report from Shay Rojansky, Npgsql sends ssl_renegotiation_limit=0
in the startup packet because it does not support renegotiation; other
clients which have not attempted to support renegotiation might well
behave similarly. The recent removal of this parameter forces them to
break compatibility with either current PostgreSQL versions, or
previous ones. Per discussion, the best solution is to accept the
parameter but only allow a value of 0.
Shay Rojansky, edited a little by me.
Commit 0e57b4d8bd contained some clever
logic that attempted to make sure that we couldn't get confused about
whether the last thing we cached was a strcoll() result or a strxfrm()
result, but it wasn't quite clever enough, because we can perform
further abbreviations after having already performed some comparisons.
Introduce an explicit flag in the hopes of making this watertight.
Peter Geoghegan, reviewed by me.
pg_regprefix was doing nothing with lookahead constraints, which would
be fine if it were the right kind of nothing, but it isn't: we have to
terminate our search for a fixed prefix, not just pretend the LACON arc
isn't there. Otherwise, if the current state has both a LACON outarc and a
single plain-color outarc, we'd falsely conclude that the color represents
an addition to the fixed prefix, and generate an extracted index condition
that restricts the indexscan too much. (See added regression test case.)
Terminating the search is conservative: we could traverse the LACON arc
(thus assuming that the constraint can be satisfied at runtime) and then
examine the outarcs of the linked-to state. But that would be a lot more
work than it seems worth, because writing a LACON followed by a single
plain character is a pretty silly thing to do.
This makes a difference only in rather contrived cases, but it's a bug,
so back-patch to all supported branches.
Using this API, one backend can set up a ParallelHeapScanDesc to
which multiple backends can then attach. Each tuple in the relation
will be returned to exactly one of the scanning backends. Only
forward scans are supported, and rescans must be carefully
coordinated.
This is not exposed to the planner or executor yet.
The original version of this code was written by me. Amit Kapila
reviewed it, tested it, and improved it, including adding support for
synchronized scans, per review comments from Jeff Davis. Extensive
testing of this and related patches was performed by Haribabu Kommi.
Final cleanup of this patch by me.
This may allow some callers to avoid the overhead involved in tearing
down a parallel context and then setting up a new one, which means
releasing the DSM and then allocating and populating a new one. I
suspect we'll want to revise the Gather node to make use of this new
capability, but even if not it may be useful elsewhere and requires
very little additional code.
Revert our previous addition of "all" flags to copyins() and copyouts();
they're no longer needed, and were never anything but an unsightly hack.
Improve a couple of infelicities in the REG_DEBUG code for dumping
the NFA data structure, including adding code to count the total
number of states and arcs.
Add a couple of missed error checks.
Add some more documentation in the README file, and some regression tests
illustrating cases that exceeded the state-count limit and/or took
unreasonable amounts of time before this set of patches.
Back-patch to all supported branches.
This code previously counted the number of NFA states it created, and
complained if a limit was exceeded, so as to prevent bizarre regex patterns
from consuming unreasonable time or memory. That's fine as far as it went,
but the code paid no attention to how many arcs linked those states. Since
regexes can be contrived that have O(N) states but will need O(N^2) arcs
after fixempties() processing, it was still possible to blow out memory,
and take a long time doing it too. To fix, modify the bookkeeping to count
space used by both states and arcs.
I did not bother with including the "color map" in the accounting; it
can only grow to a few megabytes, which is not a lot in comparison to
what we're allowing for states+arcs (about 150MB on 64-bit machines
or half that on 32-bit machines).
Looking at some of the larger real-world regexes captured in the Tcl
regression test suite suggests that the most that is likely to be needed
for regexes found in the wild is under 10MB, so I believe that the current
limit has enough headroom to make it okay to keep it as a hard-wired limit.
In connection with this, redefine REG_ETOOBIG as meaning "regular
expression is too complex"; the previous wording of "nfa has too many
states" was already somewhat inapropos because of the error code's use
for stack depth overrun, and it was not very user-friendly either.
Back-patch to all supported branches.
The previous coding would create a new intermediate state every time it
wanted to interchange the ordering of two constraint arcs. Certain regex
features such as \Y can generate large numbers of parallel constraint arcs,
and if we needed to reorder the results of that, we created unreasonable
numbers of intermediate states. To improve matters, keep a list of
already-created intermediate states associated with the state currently
being considered by the outer loop; we can re-use such states to place all
the new arcs leading to the same destination or source.
I also took the trouble to redefine push() and pull() to have a less risky
API: they no longer delete any state or arc that the caller might possibly
have a pointer to, except for the specifically-passed constraint arc.
This reduces the risk of re-introducing the same type of error seen in
the failed patch for CVE-2007-4772.
Back-patch to all supported branches.
The previous coding took something like O(N^4) time to fully process a
chain of N EMPTY arcs. We can't really do much better than O(N^2) because
we have to insert about that many arcs, but we can do lots better than
what's there now. The win comes partly from using mergeins() to amortize
de-duplication of arcs across multiple source states, and partly from
exploiting knowledge of the ordering of arcs for each state to avoid
looking at arcs we don't need to consider during the scan. We do have
to be a bit careful of the possible reordering of arcs introduced by
the sort-merge coding of the previous commit, but that's not hard to
deal with.
Back-patch to all supported branches.
Change the singly-linked in-arc and out-arc lists to be doubly-linked,
so that arc deletion is constant time rather than having worst-case time
proportional to the number of other arcs on the connected states.
Modify the bulk arc transfer operations copyins(), copyouts(), moveins(),
moveouts() so that they use a sort-and-merge algorithm whenever there's
more than a small number of arcs to be copied or moved. The previous
method is O(N^2) in the number of arcs involved, because it performs
duplicate checking independently for each copied arc. The new method may
change the ordering of existing arcs for the destination state, but nothing
really cares about that.
Provide another bulk arc copying method mergeins(), which is unused as
of this commit but is needed for the next one. It basically is like
copyins(), but the source arcs might not all come from the same state.
Replace the O(N^2) bubble-sort algorithm used in carcsort() with a qsort()
call.
These changes greatly improve the performance of regex compilation for
large or complex regexes, at the cost of extra space for arc storage during
compilation. The original tradeoff was probably fine when it was made, but
now we care more about speed and less about memory consumption.
Back-patch to all supported branches.
It's possible to construct regular expressions that contain loops of
constraint arcs (that is, ^ $ AHEAD BEHIND or LACON arcs). There's no use
in fully traversing such a loop at execution, since you'd just end up in
the same NFA state without having consumed any input. Worse, such a loop
leads to infinite looping in the pullback/pushfwd stage of compilation,
because we keep pushing or pulling the same constraints around the loop
in a vain attempt to move them to the pre or post state. Such looping was
previously recognized in CVE-2007-4772; but the fix only handled the case
of trivial single-state loops (that is, a constraint arc leading back to
its source state) ... and not only that, it was incorrect even for that
case, because it broke the admittedly-not-very-clearly-stated API contract
of the pull() and push() subroutines. The first two regression test cases
added by this commit exhibit patterns that result in assertion failures
because of that (though there seem to be no ill effects in non-assert
builds). The other new test cases exhibit multi-state constraint loops;
in an unpatched build they will run until the NFA state-count limit is
exceeded.
To fix, remove the code added for CVE-2007-4772, and instead create a
general-purpose constraint-loop-breaking phase of regex compilation that
executes before we do pullback/pushfwd. Since we never need to traverse
a constraint loop fully, we can just break the loop at any chosen spot,
if we add clone states that can replicate any sequence of arc transitions
that would've traversed just part of the loop.
Also add some commentary clarifying why we have to have all these
machinations in the first place.
This class of problems has been known for some time --- we had a report
from Marc Mamin about two years ago, for example, and there are related
complaints in the Tcl bug tracker. I had discussed a fix of this kind
off-list with Henry Spencer, but didn't get around to doing something
about it until the issue was rediscovered by Greg Stark recently.
Back-patch to all supported branches.
Prior to commit 0709b7ee72, access to
variables within a spinlock-protected critical section had to be done
through a volatile pointer, but that should no longer be necessary.
Michael Paquier
Prior to commit 0709b7ee72, access to
variables within a spinlock-protected critical section had to be done
through a volatile pointer, but that should no longer be necessary.
Thomas Munro
In order for this to be safe, the code which hands true serializability
will need to taught that the SIRead locks taken by a parallel worker
pertain to the same transaction as those taken by the parallel leader.
Some further changes may be needed as well. Until the necessary
adaptations are made, don't generate parallel plans in serializable
mode, and if a previously-generated parallel plan is used after
serializable mode has been activated, run it serially.
This fixes a bug in commit 7aea8e4f2d.
In the previous coding, before returning from ExecutorRun, we'd shut
down all parallel workers. This was dead wrong if ExecutorRun was
called with a non-zero tuple count; it had the effect of truncating
the query output. To fix, give ExecutePlan control over whether to
enter parallel mode, and have it refuse to do so if the tuple count
is non-zero. Rewrite the Gather logic so that it can cope with being
called outside parallel mode.
Commit 7aea8e4f2d is largely to blame
for this problem, though this patch modifies some subsequently-committed
code which relied on the guarantees it purported to make.
Commit 7aea8e4f2d was overoptimistic
about the degree of safety associated with running various functions
in parallel mode. Functions that take a table name or OID as an
argument are at least parallel-restricted, because the table might be
temporary, and we currently don't allow parallel workers to touch
temporary tables. Functions that take a query as an argument are
outright unsafe, because the query could be anything, including a
parallel-unsafe query.
Also, the queue of pending notifications is backend-private, so adding
to it from a worker doesn't behave correctly. We could fix this by
transferring the worker's queue of pending notifications to the master
during worker cleanup, but that seems like more trouble than it's
worth for now. In addition to adjusting the pg_proc.h markings, also
add an explicit check for this in async.c.
check_role() tries to verify that the user has permission to become the
requested role, but this is inappropriate in a parallel worker, which
needs to exactly recreate the master's authorization settings. So skip
the check in that case.
This fixes a bug in commit 924bcf4f16.
Starting a parallel worker transaction changes our notion of which XIDs
are in-progress or committed, and our notion of the current command
counter ID. Therefore, our view of these caches prior to starting
this transaction may no longer valid. Defend against that by clearing
them.
This fixes a bug in commit 924bcf4f16.
Commit 924bcf4f16 failed to enforce
parallel mode checks during the commit of a parallel worker, because
we exited parallel mode prior to ending the transaction so that we
could pop the active snapshot. Re-establish parallel mode during
parallel worker commit. Without this, it's far too easy for unsafe
actions during the pre-commit sequence to crash the server instead of
hitting the error checks as intended.
Just to be extra paranoid, adjust a couple of the sanity checks in
xact.c to check not only IsInParallelMode() but also
IsParallelWorker().
Commit 924bcf4f16 correctly forbade
parallel workers to modify the command counter while in parallel mode,
but it inexplicably neglected to actually transfer the current command
counter from leader to workers. This can result in the workers seeing
a different set of tuples from the leader, which is bad. Repair.
Commit 2bd9e412f9 introduced a mechanism
for relaying protocol messages from a background worker to another
backend via a shm_mq. However, there was no provision for shutting
down the communication channel. Therefore, a protocol message sent
late in the shutdown sequence, such as a DEBUG message resulting from
cranking up log_min_messages, could crash the server. To fix, install
an on_dsm_detach callback that disables sending messages to the shm_mq
when the associated DSM is detached.
The function failed to adhere to its specification that the "tcategory"
argument should not be examined when the input value is NULL. This
resulted in a crash in some cases. Per bug #13680 from Boyko Yordanov.
In passing, re-pgindent some recent changes in jsonb.c, and fix a rather
ungrammatical comment.
Diagnosis and patch by Michael Paquier, cosmetic changes by me
This fixes a long-standing bug which was discovered while investigating
the interaction between the new join pushdown code and the EvalPlanQual
machinery: if a ForeignScan appears on the inner side of a paramaterized
nestloop, an EPQ recheck would re-return the original tuple even if
it no longer satisfied the pushed-down quals due to changed parameter
values.
This fix adds a new member to ForeignScan and ForeignScanState and a
new argument to make_foreignscan, and requires changes to FDWs which
push down quals to populate that new argument with a list of quals they
have chosen to push down. Therefore, I'm only back-patching to 9.5,
even though the bug is not new in 9.5.
Etsuro Fujita, reviewed by me and by Kyotaro Horiguchi.
Per Mark Johnston, this resolves a build error on FreeBSD related
to the fact that dtrace is modifying the generated object files
under the hood. Consequently, without this, dtrace gets reinvoked
at install time because the object files have been updated. This
is a pretty hacky fix, but it shouldn't hurt anything, and it's
not clear that it's worth expending any more effort for a feature
that not too many people are using.
Patch by Mark Johnston. This is arguably back-patchable as a bug
fix to the build system, but I'm not certain enough of the
consequences to try that. Let's see what the buildfarm (and
our packagers) think of this change on master first.
Postmaster child processes that aren't supposed to be attached to shared
memory were not bothering to close the shared memory mapping handle they
inherit from the postmaster process. That's mostly harmless, since the
handle vanishes anyway when the child process exits -- but the syslogger
process, if used, doesn't get killed and restarted during recovery from a
backend crash. That meant that Windows doesn't see the shared memory
mapping as becoming free, so it doesn't delete it and the postmaster is
unable to create a new one, resulting in failure to recover from crashes
whenever logging_collector is turned on.
Per report from Dmitry Vasilyev. It's a bit astonishing that we'd not
figured this out long ago, since it's been broken from the very beginnings
of out native Windows support; probably some previously-unexplained trouble
reports trace to this.
A secondary problem is that on Cygwin (perhaps only in older versions?),
exec() may not detach from the shared memory segment after all, in which
case these child processes did remain attached to shared memory, posing
the risk of an unexpected shared memory clobber if they went off the rails
somehow. That may be a long-gone bug, but we can deal with it now if it's
still live, by detaching within the infrastructure introduced here to deal
with closing the handle.
Back-patch to all supported branches.
Tom Lane and Amit Kapila
Many functions stored JsonbIteratorToken values in variables of other
integer types. Also, standardize order relative to other declarations.
Expect compilers to generate the same code before and after this change.
Cache strxfrm() blobs across calls made to the text SortSupport
abbreviation routine. This can speed up sorting if the same string
needs to be abbreviated many times in a row.
Also, cache the result of the previous strcoll() comparison, so that
if we're asked to compare the same strings agin, we do need to call
strcoll() again.
Perhaps surprisingly, these optimizations don't seem to hurt even when
they don't help. memcmp() is really cheap compared to strcoll() or
strxfrm().
Peter Geoghegan, reviewed by me.
If we do some byte-swapping while abbreviating, we can do comparisons
using integer arithmetic rather than memcmp.
Peter Geoghegan, reviewed and slightly revised by me.
This flag has proven to be a recipe for bugs, and it doesn't seem like
it can really buy anything in terms of performance. So let's just
*always* set the process latch when we receive SIGUSR1 instead of
trying to do it only when needed.
Per my recent proposal on pgsql-hackers.
During expand_security_quals, we take the security barrier quals on an
RTE and create a subquery which evaluates the quals. During this, we
have to replace any variables in the outer query which refer to the
original RTE with references to the columns from the subquery.
We need to also perform that replacement for any Vars in the
append_rel_list.
Only backpatching to 9.5 as we only go through this process in 9.4 for
auto-updatable security barrier views, which UNION ALL queries aren't.
Discovered by Haribabu Kommi
Patch by Dean Rasheed
For some reason, neither of the compilers I usually use noticed the
uninitialized-variable problem I introduced in commit 7e2a18a916.
That's hardly a good enough excuse though. Committing with brown paper bag
on head.
In addition to putting the operations in the right order, move the
declaration of "now" inside the loop; there's no need for it to be
outside, and that does wake up older gcc enough to notice any similar
future problem.
Back-patch to 9.4; earlier versions lack the time-to-SIGKILL stanza
so there's no bug.
The postmaster now checks every minute or so (worst case, at most two
minutes) that postmaster.pid is still there and still contains its own PID.
If not, it performs an immediate shutdown, as though it had received
SIGQUIT.
The original goal behind this change was to ensure that failed buildfarm
runs would get fully cleaned up, even if the test scripts had left a
postmaster running, which is not an infrequent occurrence. When the
buildfarm script removes a test postmaster's $PGDATA directory, its next
check on postmaster.pid will fail and cause it to exit. Previously, manual
intervention was often needed to get rid of such orphaned postmasters,
since they'd block new test postmasters from obtaining the expected socket
address.
However, by checking postmaster.pid and not something else, we can provide
additional robustness: manual removal of postmaster.pid is a frequent DBA
mistake, and now we can at least limit the damage that will ensue if a new
postmaster is started while the old one is still alive.
Back-patch to all supported branches, since we won't get the desired
improvement in buildfarm reliability otherwise.
Prior to commit 0709b7ee72, access to
variables within a spinlock-protected critical section had to be done
through a volatile pointer, but that should no longer be necessary.
This continues work begun in df4077cda2
and 6ba4ecbf47.
Thomas Munro and Michael Paquier
For time masks, like HH24, MI, SS, CC, MM, do not count the negative
sign as part of the zero-padding length specified by the mask, e.g. have
to_char('-4 years'::interval, 'YY') return '-04', not '-4'.
Report by Craig Ringer
The tsquery, ltxtquery and query_int data types have a common ancestor.
Having acquired check_stack_depth() calls independently, each was
missing at least one call. Back-patch to 9.0 (all supported versions).
A range type can name another range type as its subtype, and a record
type can bear a column of another record type. Consequently, functions
like range_cmp() and record_recv() are recursive. Functions at risk
include operator family members and referents of pg_type regproc
columns. Treat as recursive any such function that looks up and calls
the same-purpose function for a record column type or the range subtype.
Back-patch to 9.0 (all supported versions).
An array type's element type is never itself an array type, so array
functions are unaffected. Recursion depth proportional to array
dimensionality, found in array_dim_to_jsonb(), is fine thanks to MAXDIM.
Sufficiently-deep recursion heretofore elicited a SIGSEGV. If an
application constructs PostgreSQL json or jsonb values from arbitrary
user input, application users could have exploited this to terminate all
active database connections. That applies to 9.3, where the json parser
adopted recursive descent, and later versions. Only row_to_json() and
array_to_json() were at risk in 9.2, both in a non-security capacity.
Back-patch to 9.2, where the json type was introduced.
Oskari Saarenmaa, reviewed by Michael Paquier.
Security: CVE-2015-5289
Similar to 7d8db3e, given that INSERT+RETURNING requires SELECT rights
on the table, apply the SELECT policies as WCOs to the tuples being
inserted. Apply the same logic to UPDATE+RETURNING.
Back-patch to 9.5 where RLS was added.
The WithCheckOptions list in Query are only populated during rewrite and
do not need to be written out or read in as part of a Query structure.
Further, move WithCheckOptions to the bottom and add comments to clarify
that it is only populated during rewrite.
Back-patch to 9.5 with a catversion bump, as we are still in alpha.
In 020235a575 I lowered the autovacuum_*freeze_max_age minimums to
allow for easier testing of wraparounds. I did not touch the
corresponding per-table limits. While those don't matter for the purpose
of wraparound, it seems more consistent to lower them as well.
It's noteworthy that the previous reloption lower limit for
autovacuum_multixact_freeze_max_age was too high by one magnitude, even
before 020235a575.
Discussion: 26377.1443105453@sss.pgh.pa.us
Backpatch: back to 9.0 (in parts), like the prior patch
To allow users to force RLS to always be applied, even for table owners,
add ALTER TABLE .. FORCE ROW LEVEL SECURITY.
row_security=off overrides FORCE ROW LEVEL SECURITY, to ensure pg_dump
output is complete (by default).
Also add SECURITY_NOFORCE_RLS context to avoid data corruption when
ALTER TABLE .. FORCE ROW SECURITY is being used. The
SECURITY_NOFORCE_RLS security context is used only during referential
integrity checks and is only considered in check_enable_rls() after we
have already checked that the current user is the owner of the relation
(which should always be the case during referential integrity checks).
Back-patch to 9.5 where RLS was added.
On reflection, the submitted patch didn't really work to prevent the
request size from exceeding MaxAllocSize, because of the fact that we'd
happily round nbuckets up to the next power of 2 after we'd limited it to
max_pointers. The simplest way to enforce the limit correctly is to
round max_pointers down to a power of 2 when it isn't one already.
(Note that the constraint to INT_MAX / 2, if it were doing anything useful
at all, is properly applied after that.)
Limit the size of the hashtable pointer array to not more than
MaxAllocSize, per reports from Kouhei Kaigai and others of "invalid memory
alloc request size" failures. There was discussion of allowing the array
to get larger than that by using the "huge" palloc API, but so far no proof
that that is actually a good idea, and at this point in the 9.5 cycle major
changes from old behavior don't seem like the way to go.
Fix a rather serious secondary bug in the new code, which was that it
didn't ensure nbuckets remained a power of 2 when recomputing it for the
multiple-batch case.
Clean up sloppy division of labor between ExecHashIncreaseNumBuckets and
its sole call site.
Null path elements and, where the object is an array, invalid integer
elements now cause an error.
Incorrect behaviour noted by Thom Brown, patch from Dmitry Dolgov.
Backpatch to 9.5 where jsonb_set was introduced
Specifically, make its effect independent from the row_security GUC, and
make it affect permission checks pertinent to views the BYPASSRLS role
owns. The row_security GUC thereby ceases to change successful-query
behavior; it can only make a query fail with an error. Back-patch to
9.5, where BYPASSRLS was introduced.
Without CASCADE, if an extension has an unfullfilled dependency on
another extension, CREATE EXTENSION ERRORs out with "required extension
... is not installed". That is annoying, especially when that dependency
is an implementation detail of the extension, rather than something the
extension's user can make sense of.
In addition to CASCADE this also includes a small set of regression
tests around CREATE EXTENSION.
Author: Petr Jelinek, editorialized by Michael Paquier, Andres Freund
Reviewed-By: Michael Paquier, Andres Freund, Jeff Janes
Discussion: 557E0520.3040800@2ndquadrant.com
The existing hint talked about "may only contain letters", but the
actual requirement is more strict: only lower case letters are allowed.
Reported-By: Rushabh Lathia
Author: Rushabh Lathia
Discussion: AGPqQf2x50qcwbYOBKzb4x75sO_V3g81ZsA8+Ji9iN5t_khFhQ@mail.gmail.com
Backpatch: 9.4-, where replication slots were added
Four related issues:
1) attnos/varnos/resnos for EXCLUDED were out of sync when a column
after one dropped in the underlying relation was referenced.
2) References to whole-row variables (i.e. EXCLUDED.*) lead to errors.
3) It was possible to reference system columns in the EXCLUDED pseudo
relations, even though they would not have valid contents.
4) References to EXCLUDED were rewritten by the RLS machinery, as
EXCLUDED was treated as if it were the underlying relation.
To fix the first two issues, generate the excluded targetlist with
dropped columns in mind and add an entry for whole row
variables. Instead of unconditionally adding a wholerow entry we could
pull up the expression if needed, but doing it unconditionally seems
simpler. The wholerow entry is only really needed for ruleutils/EXPLAIN
support anyway.
The remaining two issues are addressed by changing the EXCLUDED RTE to
have relkind = composite. That fits with EXCLUDED not actually being a
real relation, and allows to treat it differently in the relevant
places. scanRTEForColumn now skips looking up system columns when the
RTE has a composite relkind; fireRIRrules() already had a corresponding
check, thereby preventing RLS expansion on EXCLUDED.
Also add tests for these issues, and improve a few comments around
excluded handling in setrefs.c.
Reported-By: Peter Geoghegan, Geoff Winkless
Author: Andres Freund, Amit Langote, Peter Geoghegan
Discussion: CAEzk6fdzJ3xYQZGbcuYM2rBd2BuDkUksmK=mY9UYYDugg_GgZg@mail.gmail.com,
CAM3SWZS+CauzbiCEcg-GdE6K6ycHE_Bz6Ksszy8AoixcMHOmsA@mail.gmail.com
Backpatch: 9.5, where ON CONFLICT was introduced
Some of the functions in regex compilation and execution recurse, and
therefore could in principle be driven to stack overflow. The Tcl crew
has seen this happen in practice in duptraverse(), though their fix was
to put in a hard-wired limit on the number of recursive levels, which is
not too appetizing --- fortunately, we have enough infrastructure to check
the actually available stack. Greg Stark has also seen it in other places
while fuzz testing on a machine with limited stack space. Let's put guards
in to prevent crashes in all these places.
Since the regex code would leak memory if we simply threw elog(ERROR),
we have to introduce an API that checks for stack depth without throwing
such an error. Fortunately that's not difficult.
In cfindloop(), if the initial call to shortest() reports that a
zero-length match is possible at the current search start point, but then
it is unable to construct any actual match to that, it'll just loop around
with the same start point, and thus make no progress. We need to force the
start point to be advanced. This is safe because the loop over "begin"
points has already tried and failed to match starting at "close", so there
is surely no need to try that again.
This bug was introduced in commit e2bd904955,
wherein we allowed continued searching after we'd run out of match
possibilities, but evidently failed to think hard enough about exactly
where we needed to search next.
Because of the way this code works, such a match failure is only possible
in the presence of backrefs --- otherwise, shortest()'s judgment that a
match is possible should always be correct. That probably explains how
come the bug has escaped detection for several years.
The actual fix is a one-liner, but I took the trouble to add/improve some
comments related to the loop logic.
After fixing that, the submitted test case "()*\1" didn't loop anymore.
But it reported failure, though it seems like it ought to match a
zero-length string; both Tcl and Perl think it does. That seems to be from
overenthusiastic optimization on my part when I rewrote the iteration match
logic in commit 173e29aa5deefd9e71c183583ba37805c8102a72: we can't just
"declare victory" for a zero-length match without bothering to set match
data for capturing parens inside the iterator node.
Per fuzz testing by Greg Stark. The first part of this is a bug in all
supported branches, and the second part is a bug since 9.2 where the
iteration rewrite happened.
Commit 9662143f0c added infrastructure to
allow regular-expression operations to be terminated early in the event
of SIGINT etc. However, fuzz testing by Greg Stark disclosed that there
are still cases where regex compilation could run for a long time without
noticing a cancel request. Specifically, the fixempties() phase never
adds new states, only new arcs, so it doesn't hit the cancel check I'd put
in newstate(). Add one to newarc() as well to cover that.
Some experimentation of my own found that regex execution could also run
for a long time despite a pending cancel. We'd put a high-level cancel
check into cdissect(), but there was none inside the core text-matching
routines longest() and shortest(). Ordinarily those inner loops are very
very fast ... but in the presence of lookahead constraints, not so much.
As a compromise, stick a cancel check into the stateset cache-miss
function, which is enough to guarantee a cancel check at least once per
lookahead constraint test.
Making this work required more attention to error handling throughout the
regex executor. Henry Spencer had apparently originally intended longest()
and shortest() to be incapable of incurring errors while running, so
neither they nor their subroutines had well-defined error reporting
behaviors. However, that was already broken by the lookahead constraint
feature, since lacon() can surely suffer an out-of-memory failure ---
which, in the code as it stood, might never be reported to the user at all,
but just silently be treated as a non-match of the lookahead constraint.
Normalize all that by inserting explicit error tests as needed. I took the
opportunity to add some more comments to the code, too.
Back-patch to all supported branches, like the previous patch.
Module initialization was still not completely correct after commit
6b61955135, per crash report from Takashi Ohnishi. To fix, instead of
trying to monkey around with the value of the GUC setting directly, add
a separate boolean flag that enables the feature on a standby, but only
for the startup (recovery) process, when it sees that its master server
has the feature enabled.
Discussion: http://www.postgresql.org/message-id/ca44c6c7f9314868bdc521aea4f77cbf@MP-MSGSS-MBX004.msg.nttdata.co.jp
Also change the deactivation routine to delete all segment files rather
than leaving the last one around. (This doesn't need separate
WAL-logging, because on recovery we execute the same deactivation
routine anyway.)
In passing, clean up the code structure somewhat, particularly so that
xlog.c doesn't know so much about when to activate/deactivate the
feature.
Thanks to Fujii Masao for testing and Petr Jelínek for off-list discussion.
Back-patch to 9.5, where commit_ts was introduced.
If some existing listener is far behind, incoming new listener sessions
would start from that session's read pointer and then need to advance over
many already-committed notification messages, which they have no interest
in. This was expensive in itself and also thrashed the pg_notify SLRU
buffers a lot more than necessary. We can improve matters considerably
in typical scenarios, without much added cost, by starting from the
furthest-ahead read pointer, not the furthest-behind one. We do have to
consider only sessions in our own database when doing this, which requires
an extra field in the data structure, but that's a pretty small cost.
Back-patch to 9.0 where the current LISTEN/NOTIFY logic was introduced.
Matt Newell, slightly adjusted by me
A Gather executor node runs any number of copies of a plan in an equal
number of workers and merges all of the results into a single tuple
stream. It can also run the plan itself, if the workers are
unavailable or haven't started up yet. It is intended to work with
the Partial Seq Scan node which will be added in future commits.
It could also be used to implement parallel query of a different sort
by itself, without help from Partial Seq Scan, if the single_copy mode
is used. In that mode, a worker executes the plan, and the parallel
leader does not, merely collecting the worker's results. So, a Gather
node could be inserted into a plan to split the execution of that plan
across two processes. Nested Gather nodes aren't currently supported,
but we might want to add support for that in the future.
There's nothing in the planner to actually generate Gather nodes yet,
so it's not quite time to break out the champagne. But we're getting
close.
Amit Kapila. Some designs suggestions were provided by me, and I also
reviewed the patch. Single-copy mode, documentation, and other minor
changes also by me.
If a transaction or subtransaction creates a ParallelContext but ends
without calling InitializeParallelDSM, the previous code would
seg fault. Fix that.
When considering which policies should be included, rather than look at
individual bits of the query (eg: if a RETURNING clause exists, or if a
WHERE clause exists which is referencing the table, or if it's a
FOR SHARE/UPDATE query), consider any case where we've determined
the user needs SELECT rights on the relation while doing an UPDATE or
DELETE to be a case where we apply SELECT policies, and any case where
we've deteremind that the user needs UPDATE rights on the relation while
doing a SELECT to be a case where we apply UPDATE policies.
This simplifies the logic and addresses concerns that a user could use
UPDATE or DELETE with a WHERE clauses to determine if rows exist, or
they could use SELECT .. FOR UPDATE to lock rows which they are not
actually allowed to modify through UPDATE policies.
Use list_append_unique() to avoid adding the same quals multiple times,
as, on balance, the cost of checking when adding the quals will almost
always be cheaper than keeping them and doing busywork for each tuple
during execution.
Back-patch to 9.5 where RLS was added.
We seem to have lost a line somewhere along the way in the comment block
that discusses async.c's locks, because it suddenly refers to "both locks"
without previously having mentioned more than one. Add a sentence to make
that read more sanely. Also, refer to the "pos of the slowest backend"
not the "tail of the slowest backend", since we have no per-backend value
called "tail".
There are three main changes here:
1. No longer cause a start failure in a standby if the feature is
disabled in postgresql.conf but enabled in the master. This reverts one
part of commit 4f3924d9cd43; what we keep is the ability of the standby
to activate/deactivate the module (which includes creating and removing
segments as appropriate) during replay of such actions in the master.
2. Replay WAL records affecting commitTS even if the feature is
disabled. This means the standby will always have the same state as the
master after replay.
3. Have COMMIT PREPARE record the transaction commit time as well. We
were previously only applying it in the normal transaction commit path.
Author: Petr Jelínek
Discussion: http://www.postgresql.org/message-id/CAHGQGwHereDzzzmfxEBYcVQu3oZv6vZcgu1TPeERWbDc+gQ06g@mail.gmail.com
Discussion: http://www.postgresql.org/message-id/CAHGQGwFuzfO4JscM9LCAmCDCxp_MfLvN4QdB+xWsS-FijbjTYQ@mail.gmail.com
Additionally, I cleaned up nearby code related to replication origins,
which I found a bit hard to follow, and fixed a couple of typos.
Backpatch to 9.5, where this code was introduced.
Per bug reports from Fujii Masao and subsequent discussion.
This code provides infrastructure for a parallel leader to start up
parallel workers to execute subtrees of the plan tree being executed
in the master. User-supplied parameters from ParamListInfo are passed
down, but PARAM_EXEC parameters are not. Various other constructs,
such as initplans, subplans, and CTEs, are also not currently shared.
Nevertheless, there's enough here to support a basic implementation of
parallel query, and we can lift some of the current restrictions as
needed.
Amit Kapila and Robert Haas
When taking the UPDATE path in an INSERT .. ON CONFLICT .. UPDATE tables
with oids were not supported. The tuple generated by the update target
list was projected without space for an oid - a simple oversight.
Reported-By: Peter Geoghegan
Author: Andres Freund
Backpatch: 9.5, where ON CONFLICT was introduced
In 9.5 and master there is no need to support legacy truncation. This is
just committed separately to make it easier to backpatch the WAL logged
multixact truncation to 9.3 and 9.4 if we later decide to do so.
I bumped master's magic from 0xD086 to 0xD088 and 9.5's from 0xD085 to
0xD087 to avoid 9.5 reusing a value that has been in use on master while
keeping the numbers increasing between major versions.
Discussion: 20150621192409.GA4797@alap3.anarazel.de
Backpatch: 9.5
The fact that multixact truncations are not WAL logged has caused a fair
share of problems. Amongst others it requires to do computations during
recovery while the database is not in a consistent state, delaying
truncations till checkpoints, and handling members being truncated, but
offset not.
We tried to put bandaids on lots of these issues over the last years,
but it seems time to change course. Thus this patch introduces WAL
logging for multixact truncations.
This allows:
1) to perform the truncation directly during VACUUM, instead of delaying it
to the checkpoint.
2) to avoid looking at the offsets SLRU for truncation during recovery,
we can just use the master's values.
3) simplify a fair amount of logic to keep in memory limits straight,
this has gotten much easier
During the course of fixing this a bunch of additional bugs had to be
fixed:
1) Data was not purged from memory the member's SLRU before deleting
segments. This happened to be hard or impossible to hit due to the
interlock between checkpoints and truncation.
2) find_multixact_start() relied on SimpleLruDoesPhysicalPageExist - but
that doesn't work for offsets that haven't yet been flushed to
disk. Add code to flush the SLRUs to fix. Not pretty, but it feels
slightly safer to only make decisions based on actual on-disk state.
3) find_multixact_start() could be called concurrently with a truncation
and thus fail. Via SetOffsetVacuumLimit() that could lead to a round
of emergency vacuuming. The problem remains in
pg_get_multixact_members(), but that's quite harmless.
For now this is going to only get applied to 9.5+, leaving the issues in
the older branches in place. It is quite possible that we need to
backpatch at a later point though.
For the case this gets backpatched we need to handle that an updated
standby may be replaying WAL from a not-yet upgraded primary. We have to
recognize that situation and use "old style" truncation (i.e. looking at
the SLRUs) during WAL replay. In contrast to before, this now happens in
the startup process, when replaying a checkpoint record, instead of the
checkpointer. Doing truncation in the restartpoint is incorrect, they
can happen much later than the original checkpoint, thereby leading to
wraparound. To avoid "multixact_redo: unknown op code 48" errors
standbys would have to be upgraded before primaries.
A later patch will bump the WAL page magic, and remove the legacy
truncation codepaths. Legacy truncation support is just included to make
a possible future backpatch easier.
Discussion: 20150621192409.GA4797@alap3.anarazel.de
Reviewed-By: Robert Haas, Alvaro Herrera, Thomas Munro
Backpatch: 9.5 for now
This replaces ill-fated commit 5ddc72887a,
which was reverted because it broke active uses of FK cache entries. In
this patch, we still do nothing more to invalidatable cache entries than
mark them as needing revalidation, so we won't break active uses. To keep
down the overhead of InvalidateConstraintCacheCallBack(), keep a list of
just the currently-valid cache entries. (The entries are large enough that
some added space for list links doesn't seem like a big problem.) This
would still be O(N^2) when there are many valid entries, though, so when
the list gets too long, just force the "sinval reset" behavior to remove
everything from the list. I set the threshold at 1000 entries, somewhat
arbitrarily. Possibly that could be fine-tuned later. Another item for
future study is whether it's worth adding reference counting so that we
could safely remove invalidated entries. As-is, problem cases are likely
to end up with large and mostly invalid FK caches.
Like the previous attempt, backpatch to 9.3.
Jan Wieck and Tom Lane
Previously, a function call appearing at the top level of WHERE had a
hard-wired selectivity estimate of 0.3333333, a kludge conveniently dated
in the source code itself to July 1992. The expectation at the time was
that somebody would soon implement estimator support functions analogous
to those for operators; but no such code has appeared, nor does it seem
likely to in the near future. We do have an alternative solution though,
at least for immutable functions on single relations: creating an
expression index on the function call will allow ANALYZE to gather stats
about the function's selectivity. But the code in clause_selectivity()
failed to make use of such data even if it exists.
Refactor so that that will happen. I chose to make it try this technique
for any clause type for which clause_selectivity() doesn't have a special
case, not just functions. To avoid adding unnecessary overhead in the
common case where we don't learn anything new, make selfuncs.c provide an
API that hooks directly to examine_variable() and then var_eq_const(),
rather than the previous coding which laboriously constructed an OpExpr
only so that it could be expensively deconstructed again.
I preserved the behavior that the default estimate for a function call
is 0.3333333. (For any other expression node type, it's 0.5, as before.)
I had originally thought to make the default be 0.5 across the board, but
changing a default estimate that's survived for twenty-three years seems
like something not to do without a lot more testing than I care to put
into it right now.
Per a complaint from Jehan-Guillaume de Rorthais. Back-patch into 9.5,
but not further, at least for the moment.
The comments here stated that this was just in case we ever had an
ALTER OPERATOR command that could remap an operator to a different
function. But those comments have been here for a long time, and no
such command has come about. In the absence of such a feature,
forcing the pg_proc OID to be looked up again each time we reread a
stored rule or similar is just a waste of cycles. Moreover, parallel
query needs a way to reread the exact same node tree that was written
out, not one that has been slightly stomped on. So just get rid of
this for now.
Per discussion with Tom Lane.
The old minimum values are rather large, making it time consuming to
test related behaviour. Additionally the current limits, especially for
multixacts, can be problematic in space-constrained systems. 10000000
multixacts can contain a lot of members.
Since there's no good reason for the current limits, lower them a good
bit. Setting them to 0 would be a bad idea, triggering endless vacuums,
so still retain a limit.
While at it fix autovacuum_multixact_freeze_max_age to refer to
multixact.c instead of varsup.c.
Reviewed-By: Robert Haas
Discussion: CA+TgmoYmQPHcrc3GSs7vwvrbTkbcGD9Gik=OztbDGGrovkkEzQ@mail.gmail.com
Backpatch: back to 9.0 (in parts)
Previously, ANALYZE simply ignored columns of datatypes that have neither
a btree nor hash opclass (which means they have no recognized equality
operator). Without a notion of equality, we can't identify most-common
values nor estimate the number of distinct values. But we can still
count nulls and compute the average physical column width, and those
stats might be of value. Moreover there are some tools out there that
don't work so well if rows are missing from pg_statistic. So let's
add suitable logic for this case.
While this is arguably a bug fix, it also has the potential to change
query plans, and the gain seems not worth taking a risk of that in
stable branches. So back-patch into 9.5 but not further.
Oleksandr Shulgin, rewritten a bit by me.
For parallel query, we need to be able to pass a Plan to a worker, so
that it knows what it's supposed to do. We could invent our own way
of serializing plans for that purpose, but piggybacking on the
existing node infrastructure seems like a much better idea.
Initially, we'll probably only support a limited number of nodes
within parallel workers, but this commit adds support for everything
in plannodes.h except CustomScan, because doing it all at once seems
easier than doing it piecemeal, and it makes testing this code easier,
too. CustomScan is excluded because making that work requires a
larger rework of that facility.
Amit Kapila, reviewed and slightly revised by me.
It's declared as being an array of bool, but it's printed
differently from the way bool and arrays of bool are handled
elsewhere.
Patch by Amit Kapila. Anomaly noted independently by Amit Kapila
and KaiGai Kohei.
Commit e956808328 introduces adding pages
to FSM for ordinary insert, but autoanalyze was able just cleanup
pending list without adding to FSM.
Also fix double call of IndexFreeSpaceMapVacuum() during ginvacuumcleanup()
Report from Fujii Masao
Patch by me
Review by Jeff Janes
This logic was missing from ExplainPreScanNode, from which I derived
planstate_tree_walker. But it shouldn't be missing, especially not
from a generic walker function, so add it.
KaiGai Kohei
mul_var() postpones propagating carries until it risks overflow in its
internal digit array. However, the logic failed to account for the
possibility of overflow in the carry propagation step, allowing wrong
results to be generated in corner cases. We must slightly reduce the
when-to-propagate-carries threshold to avoid that.
Discovered and fixed by Dean Rasheed, with small adjustments by me.
This has been wrong since commit d72f6c7503,
so back-patch to all supported branches.
This commit's parent made superfluous the bit's sole usage. Referential
integrity checks have long run as the subject table's owner, and that
now implies RLS bypass. Safe use of the bit was tricky, requiring
strict control over the SQL expressions evaluating therein. Back-patch
to 9.5, where the bit was introduced.
Based on a patch by Stephen Frost.
Every query of a single ENABLE ROW SECURITY table has two meanings, with
the row_security GUC selecting between them. With row_security=force
available, every function author would have been advised to either set
the GUC locally or test both meanings. Non-compliance would have
threatened reliability and, for SECURITY DEFINER functions, security.
Authors already face an obligation to account for search_path, and we
should not mimic that example. With this change, only BYPASSRLS roles
need exercise the aforementioned care. Back-patch to 9.5, where the
row_security GUC was introduced.
Since this narrows the domain of pg_db_role_setting.setconfig and
pg_proc.proconfig, one might bump catversion. A row_security=force
setting in one of those columns will elicit a clear message, so don't.
RemoveLocalLock() must consider the possibility that LockAcquireExtended()
failed to palloc the initial space for a locallock's lockOwners array.
I had evidently meant to cope with this hazard when the code was originally
written (commit 1785acebf2), but missed that
the pfree needed to be protected with an if-test. Just to make sure things
are left in a clean state, reset numLockOwners as well.
Per low-memory testing by Andreas Seltenreich. Back-patch to all supported
branches.
BSD find is not very smart and ends up writing double slashes into the
output in those cases. Also, xgettext is not very smart and splits the
file names incorrectly in those cases, resulting in slightly incorrect
file names being written into the POT file.
The shm_mq mechanism was built to send error (and notice) messages and
tuples between backends. However, shm_mq itself only deals in raw
bytes. Since commit 2bd9e412f9, we have
had infrastructure for one message to redirect protocol messages to a
queue and for another backend to parse them and do useful things with
them. This commit introduces a somewhat analogous facility for tuples
by adding a new type of DestReceiver, DestTupleQueue, which writes
each tuple generated by a query into a shm_mq, and a new
TupleQueueFunnel facility which reads raw tuples out of the queue and
reconstructs the HeapTuple format expected by the executor.
The TupleQueueFunnel abstraction supports reading from multiple tuple
streams at the same time, but only in round-robin fashion. Someone
could imaginably want other policies, but this should be good enough
to meet our short-term needs related to parallel query, and we can
always extend it later.
This also makes one minor addition to the shm_mq API that didn'
seem worth breaking out as a separate patch.
Extracted from Amit Kapila's parallel sequential scan patch. This
code was originally written by me, and then it was revised by Amit,
and then it was revised some more by me.
These functions have been looking up type info for every row they
process. Instead of doing that we only look them up the first time
through and stash the information in the aggregate state object.
Affects json_agg, json_object_agg, jsonb_agg and jsonb_object_agg.
There is plenty more work to do in making these more efficient,
especially the jsonb functions, but this is a virtually cost free
improvement that can be done right away.
Backpatch to 9.5 where the jsonb variants were introduced.
After an internal failure in shortest() or longest() while pinning down the
exact location of a match, find() forgot to free the DFA structure before
returning. This is pretty unlikely to occur, since we just successfully
ran the "search" variant of the DFA; but it could happen, and it would
result in a session-lifespan memory leak since this code uses malloc()
directly. Problem seems to have been aboriginal in Spencer's library,
so back-patch all the way.
In passing, correct a thinko in a comment I added awhile back about the
meaning of the "ntree" field.
I happened across these issues while comparing our code to Tcl's version
of the library.
Use IsBinaryCoercible() method instead of custom
is_expected_type/is_text_type functions which was introduced when tsearch2
was moved into core.
Per report by David E. Wheeler
Analysis by Tom Lane
Patch by me
Alvaro Herrera
Tom Lane
Bruce Momjian
Noah Misch
Joe Conway
Robert Haas
Peter Eisentraut
Stephen Frost
Teodor Sigaev
Andres Freund
Magnus Hagander
Kevin Grittner
Andrew Dunstan