The postmaster was coded to treat any unexpected exit of the startup
process (i.e., the WAL replay process) as a catastrophic crash, and not try
to restart it. This was OK so long as the startup process could not have
any sibling postmaster children. However, if a hot-standby backend
crashes, we SIGQUIT the startup process along with everything else, and the
resulting exit is hardly "unexpected". Treating it as such meant we failed
to restart a standby server after any child crash at all, not only a crash
of the WAL replay process as intended. Adjust that. Back-patch to 9.0
where hot standby was introduced.
Although we will not even issue an XLOG_TBLSPC_DROP WAL record unless
removal of the tablespace's directories succeeds, that does not guarantee
that the same operation will succeed during WAL replay. Foreseeable
reasons for it to fail include temp files created in the tablespace by Hot
Standby backends, wrong directory permissions on a standby server, etc etc.
The original coding threw ERROR if replay failed to remove the directories,
but that is a serious overreaction. Throwing an error aborts recovery,
and worse means that manual intervention will be needed to get the database
to start again, since otherwise the same error will recur on subsequent
attempts to replay the same WAL record. And the consequence of failing to
remove the directories is only that some probably-small amount of disk
space is wasted, so it hardly seems justified to throw an error.
Accordingly, arrange to report such failures as LOG messages and keep going
when a failure occurs during replay.
Back-patch to 9.0 where Hot Standby was introduced. In principle such
problems can occur in earlier releases, but Hot Standby increases the odds
of trouble significantly. Given the lack of field reports of such issues,
I'm satisfied with patching back as far as the patch applies easily.
Instead, everything that needs the Archive object now gets it as a
parameter. This is necessary infrastructure for parallel pg_dump,
but is also amply justified by the ugliness of the current code
(though a lot more than this is needed to fix that problem).
Change various places in the code that are referencing the global
Archive object g_fout to instead reference the Archive object fout
which is already being passed as a parameter. For parallel pg_dump to
work, we're going to need multiple Archive(Handle) objects, so the
real solution here is to pass down the Archive object to everywhere
that it needs to go, but we might as well pick the low-hanging fruit
first.
Originally, most of this code assumed that no Postgres backends could be
running concurrently with it, and so no locking could be needed. That
assumption fails in Hot Standby. While it's still true that Hot Standby
backends should never change values like nextXid, they can examine them,
and consistency is important in some cases such as when computing a
snapshot. Therefore, prudence requires that WAL replay code obtain the
relevant locks when modifying such variables, even though it can examine
them without taking a lock. We were following that coding rule in some
places but not all. This commit applies the coding rule uniformly to all
updates of ShmemVariableCache and MultiXactState fields; a search of the
replay routines did not find any other cases that seemed to be at risk.
In addition, this commit fixes a longstanding thinko in replay of NEXTOID
and checkpoint records: we tried to advance nextOid only if it was behind
the value in the WAL record, but the comparison would draw the wrong
conclusion if OID wraparound had occurred since the previous value.
Better to just unconditionally assign the new value, since OID assignment
shouldn't be happening during replay anyway.
The additional locking seems to be more in the nature of future-proofing
than fixing any live bug, so I am not going to back-patch it. The NEXTOID
fix will be back-patched separately.
RestoreBkpBlocks was in the habit of zeroing and refilling the target
buffer; which was perfectly safe when the code was written, but is unsafe
during Hot Standby operation. The reason is that we have coding rules
that allow backends to continue accessing a tuple in a heap relation while
holding only a pin on its buffer. Such a backend could see transiently
zeroed data, if WAL replay had occasion to change other data on the page.
This has been shown to be the cause of bug #6425 from Duncan Rance (who
deserves kudos for developing a sufficiently-reproducible test case) as
well as Bridget Frey's re-report of bug #6200. It most likely explains the
original report as well, though we don't yet have confirmation of that.
To fix, change the code so that only bytes that are supposed to change will
change, even transiently. This actually saves cycles in RestoreBkpBlocks,
since it's not writing the same bytes twice.
Also fix seq_redo, which has the same disease, though it has to work a bit
harder to meet the requirement.
So far as I can tell, no other WAL replay routines have this type of bug.
In particular, the index-related replay routines, which would certainly be
broken if they had to meet the same standard, are not at risk because we
do not have coding rules that allow access to an index page when not
holding a buffer lock on it.
Back-patch to 9.0 where Hot Standby was added.
All other WAL redo routines either call RestoreBkpBlocks() or Assert that
they haven't been passed any backup blocks. Make this one do likewise.
Also, fix incorrect routine name in its failure message.
This reverts commit 500cf66d55. As was
more or less expected, a small minority of platforms won't accept
denormalized input even with the recent changes. It doesn't seem
especially helpful to test this if we're going to have to provide an
alternate expected-file to allow failure.
Further improve on commit c75e143646.
Instead of building both .o files and binaries in the same make rule,
just rely on the normal .c -> .o rule. This will ensure that
dependency tracking is used when enabled. To do this, disable the
implicit direct .c -> binary rule globally, which will also prevent
the original problem (*.dSYM junk) from reappearing elsewhere.
When testing bits (but not when setting or clearing them), we now
won't check whether the map has been extended. This significantly
improves performance in the case where the visibility map doesn't
exist yet, by avoiding an extra system call per tuple. To make
sure backends notice eventually, send an smgr inval on VM extension.
Dean Rasheed, with minor modifications by me.
This was submitted with the previous patch, but I'm committing it
separately to ease backing it out if these results prove too unportable.
Marti Raudsepp, after a proposal by Jeroen Vermeulen
On some platforms, strtod() reports ERANGE for a denormalized value (ie,
one that can be represented as distinct from zero, but is too small to have
full precision). On others, it doesn't. It seems better to try to accept
these values consistently, so add a test to see if the result value
indicates a true out-of-range condition. This should be okay per Single
Unix Spec. On machines where the underlying math isn't IEEE standard, the
behavior for such small numbers may not be very consistent, but then it
wouldn't be anyway.
Marti Raudsepp, after a proposal by Jeroen Vermeulen
In dry-run mode, just the name of the file to be removed is printed to
stdout; this is so the user can easily plug it into another program
through a pipe. If debug mode is also specified, a more verbose message
is printed to stderr.
Author: Gabriele Bartolini
Reviewer: Josh Kupershmidt
Don't quote the output of format_procedure(); it's already quoted quite
enough. Remove the fn_name field, which was now just dead weight. Fix
remaining expected-output files.
Like the XML data type, we simply store JSON data as text, after checking
that it is valid. More complex operations such as canonicalization and
comparison may come later, but this is enough for not.
There are a few open issues here, such as whether we should attempt to
detect UTF-8 surrogate pairs represented as \uXXXX\uYYYY, but this gets
the basic framework in place.
If there was a wait-until-free process in the head of the wait queue,
followed by an exclusive locker, the exclusive locker was not be woken up
as it should.
The sequence USAGE privilege is sufficiently similar to the SQL
standard that it seems reasonable to show in the information schema.
Also add some compatibility notes about it on the GRANT reference
page.
Add result object functions .colnames, .coltypes, .coltypmods to
obtain information about the result column names and types, which was
previously not possible in the PL/Python SPI interface.
reviewed by Abhijit Menon-Sen
In some hopeless situations, certain library functions in libpq and
libpgport quit the program. Use abort() for that instead of exit(),
so we don't interfere with the normal exit codes the program might
use, we clearly signal the abnormal termination, and the caller has a
chance of catching the termination.
This was originally pointed out by Debian's Lintian program.
When a backend needs to flush the WAL, and someone else is already flushing
the WAL, wait until it releases the WALInsertLock and check if we still need
to do the flush or if the other backend already did the work for us, before
acquiring WALInsertLock. This helps group commit, because when the WAL flush
finishes, all the backends that were waiting for it can be woken up in one
go, and the can all concurrently observe that they're done, rather than
waking them up one by one in a cascading fashion.
This is based on a new LWLock function, LWLockWaitUntilFree(), which has
peculiar semantics. If the lock is immediately free, it grabs the lock and
returns true. If it's not free, it waits until it is released, but then
returns false without grabbing the lock. This is used in XLogFlush(), so
that when the lock is acquired, the backend flushes the WAL, but if it's
not, the backend first checks the current flush location before retrying.
Original patch and benchmarking by Peter Geoghegan and Simon Riggs, although
this patch as committed ended up being very different from that.
When default_text_search_config, default_tablespace, or temp_tablespaces
setting is set per-user or per-database, with an "ALTER USER/DATABASE SET
..." statement, don't throw an error if the text search configuration or
tablespace does not exist. In case of text search configuration, even if
it doesn't exist in the current database, it might exist in another
database, where the setting is intended to have its effect. This behavior
is now the same as search_path's.
Tablespaces are cluster-wide, so the same argument doesn't hold for
tablespaces, but there's a problem with pg_dumpall: it dumps "ALTER USER
SET ..." statements before the "CREATE TABLESPACE" statements. Arguably
that's pg_dumpall's fault - it should dump the statements in such an order
that the tablespace is created first and then the "ALTER USER SET
default_tablespace ..." statements after that - but it seems better to be
consistent with search_path and default_text_search_config anyway. Besides,
you could still create a dump that throws an error, by creating the
tablespace, running "ALTER USER SET default_tablespace", then dropping the
tablespace and running pg_dumpall on that.
Backpatch to all supported versions.
btcostestimate() makes an estimate of the number of index tuples that will
be visited based on knowledge of which index clauses can actually bound the
scan within nbtree. However, it forgot to account for partial indexes in
this calculation, with the result that the cost of the index scan could be
significantly overestimated for a partial index. Fix that by merging the
predicate with the abbreviated indexclause list, in the same way as we do
with the full list to estimate how many heap tuples will be visited.
Also, slightly increase the "fudge factor" that's meant to give preference
to smaller indexes over larger ones. While this is applied to all indexes,
it's most important for partial indexes since it can be the only factor
that makes a partial index look cheaper than a similar full index.
Experimentation shows that the existing value is so small as to easily get
swamped by noise such as page-boundary-roundoff behavior. I'm tempted to
kick it up more than this, but will refrain for now.
Per report from Ruben Blanco. These are long-standing issues, but given
the lack of prior complaints I'm not going to risk changing planner
behavior in back branches by back-patching.
In commit 57664ed25e, I made the planner
wrap non-simple-variable outputs of appendrel children (IOW, child SELECTs
of UNION ALL subqueries) inside PlaceHolderVars, in order to solve some
issues with EquivalenceClass processing. However, this means that any
upper-level WHERE clauses mentioning such outputs will now contain
PlaceHolderVars after they're pushed down into the appendrel child,
and that prevents indxpath.c from recognizing that they could be matched
to index expressions. To fix, add explicit stripping of PlaceHolderVars
from index operands, same as we have long done for RelabelType nodes.
Add a regression test covering both this and the plain-UNION case (which
is a totally different code path, but should also be able to do it).
Per bug #6416 from Matteo Beccati. Back-patch to 9.1, same as the
previous change.
Formerly we passed an empty list to each per-child-table invocation of
grouping_planner, and then merged the results into the global list.
However, that fails if there's a CTE attached to the statement, because
create_ctescan_plan uses the list to find the plan referenced by a CTE
reference; so it was unable to find any CTEs attached to the outer UPDATE
or DELETE. But there's no real reason not to use the same list throughout
the process, and doing so is simpler and faster anyway.
Per report from Josh Berkus of "could not find plan for CTE" failures.
Back-patch to 9.1 where we added support for WITH attached to UPDATE or
DELETE. Add some regression test cases, too.
Much more could be done here, but at least now we have *some* automated
test coverage of that mechanism. In particular this tests the writable-CTE
case reported by Phil Sorber.
In passing, remove isolationtester's arbitrary restriction on the number of
steps in a permutation list. I used this so that a single spec file could
be used to run several related test scenarios, but there are other possible
reasons to want a step series that's not exactly a permutation. Improve
documentation and fix a couple other nits as well.
We can't just skip initializing such subplans, because the referencing CTE
node will expect to find the subplan available when it initializes. That
in turn means that ExecInitModifyTable must allow the case (which actually
it needed to do anyway, since there's no guarantee that ModifyTable is
exactly at the top of the CTE plan tree). So move the complaint about not
being allowed in EvalPlanQual mode to execution instead of initialization.
Testing turned up yet another problem, which is that we'd try to
re-initialize the result relation's index list, leading to leaks and
dangling pointers.
Per report from Phil Sorber. Back-patch to 9.1 where data-modifying CTEs
were introduced.
