The only interesting-for-performance case wherein we force heapscan here
is when we're rebuilding the relcache init file, and the only such case
that is likely to be examining a catalog big enough to be syncscanned is
RelationBuildTupleDesc. But the early-exit optimization in that code gets
broken if we start the scan at a random place within the catalog, so that
allowing syncscan is actually a big deoptimization if pg_attribute is large
(at least for the normal case where the rows for core system catalogs have
never been changed since initdb). Hence, prevent syncscan here. Per my
testing pursuant to complaints from Jeff Frost and Greg Sabino Mullane,
though neither of them seem to have actually hit this specific problem.
Back-patch to 8.3, where syncscan was introduced.
Previously, casts to name could generate invalidly-encoded results.
Also, make these functions match namein() more exactly, by consistently
using palloc0() instead of ad-hoc zeroing code.
Back-patch to all supported branches.
Karl Schnaitter and Tom Lane
The previous coding presented a significant bottleneck when dumping
databases containing many thousands of schemas, since the total time
spent searching would increase roughly as O(N^2) in the number of objects.
Noted by Jeff Janes, though I rewrote his proposed patch to use the
existing findObjectByOid infrastructure.
Since this is a longstanding performance bug, backpatch to all supported
versions.
If a seqscan encounters many consecutive pages containing only dead tuples,
it can remain in the loop in heapgettup for a long time, and there was no
CHECK_FOR_INTERRUPTS anywhere in that loop. This meant there were
real-world situations where a query would be effectively uncancelable for
long stretches. Add a check placed to occur once per page, which should be
enough to provide reasonable response time without adding any measurable
overhead.
Report and patch by Merlin Moncure (though I tweaked it a bit).
Back-patch to all supported branches.
When the column name is an unqualified name, rather than table.column,
the error message complains about too many dotted names, which is
wrong. Report by Peter Eisentraut based on examination of the
sepgsql regression test output, but the problem also affects COMMENT.
New wording as suggested by Tom Lane.
There is no reason to do this as early as possible in postmaster startup,
and good reason not to do it until we have completely created the
postmaster's lock file, namely that it might contribute to pg_ctl thinking
that postmaster startup has timed out. (This would require a rather
unusual amount of time to be spent scanning temp file directories, but we
have at least one field report of it happening reproducibly.)
Back-patch to 9.1. Before that, pg_ctl didn't wait for additional info to
be added to the lock file, so it wasn't a problem.
Note that this is not a complete fix to the slow-start issue in 9.1,
because we still had identify_system_timezone being run during postmaster
start in 9.1. But that's at least a reasonably well-defined delay, with
an easy workaround if needed, whereas the temp-files scan is not so
predictable and cannot be avoided.
If the tablespace directory is missing entirely, we allow DROP TABLESPACE
to go through, on the grounds that it should be possible to clean up the
catalog entry in such a situation. However, we forgot that the pg_tblspc
symlink might still be there. We should try to remove the symlink too
(but not fail if it's no longer there), since not doing so can lead to
weird behavior subsequently, as per report from Michael Nolan.
There was some discussion of adding dependency links to prevent DROP
TABLESPACE when the catalogs still contain references to the tablespace.
That might be worth doing too, but it's an orthogonal question, and in
any case wouldn't be back-patchable.
Back-patch to 9.0, which is as far back as the logic looks like this.
We could possibly do something similar in 8.x, but given the lack of
reports I'm not sure it's worth the trouble, and anyway the case could
not arise in the form the logic is meant to cover (namely, a post-DROP
transaction rollback having resurrected the pg_tablespace entry after
some or all of the filesystem infrastructure is gone).
We previously recognized that citext wouldn't get marked as collatable
during pg_upgrade from a pre-9.1 installation, and hacked its
create-from-unpackaged script to manually perform the necessary catalog
adjustments. However, we overlooked the fact that domains over citext,
as well as the citext[] array type, need the same adjustments. Extend
the script to handle those cases.
Also, the documentation suggested that this was only an issue in pg_upgrade
scenarios, which is quite wrong; loading any dump containing citext from a
pre-9.1 server will also result in the type being wrongly marked.
I approached the documentation problem by changing the 9.1.2 release note
paragraphs about this issue, which is historically inaccurate. But it
seems better than having the information scattered in multiple places, and
leaving incorrect info in the 9.1.2 notes would be bad anyway. We'll still
need to mention the issue again in the 9.1.4 notes, but perhaps they can
just reference 9.1.2 for fix instructions.
Per report from Evan Carroll. Back-patch into 9.1.
When inserting the downlinks for a split gist page, we used hold the locks
on the child pages until the insertion into the parent - and recursively its
parent if it had to be split too - were all completed. Change that so that
the locks on child pages are released after the insertion in the immediate
parent is done, before recursing further up the tree.
This reduces the number of lwlocks that are held simultaneously. Holding
many locks is bad for concurrency, and in extreme cases you can even hit
the limit of 100 simultaneously held lwlocks in a backend. If you're really
unlucky, you can hit the limit while in a critical section, which brings
down the whole system.
This fixes bug #6629 reported by Tom Forbes. Backpatch to 9.1. The page
splitting code was rewritten in 9.1, and the old code did not have this
problem.
The original coding failed to reset ImmediateInterruptOK before returning,
which would potentially allow a subsequent query-cancel interrupt to be
accepted at an unsafe point. This is a really nasty bug since it's so hard
to predict the consequences, but they could be unpleasant.
Also, ensure that signal handlers are serviced before this function
returns, even if the semaphore is already set. This should make the
behavior more like Unix.
Back-patch to all supported versions.
Commit 62c7bd31c8 had assorted problems, most
visibly that it broke PREPARE TRANSACTION in the presence of session-level
advisory locks (which should be ignored by PREPARE), as per a recent
complaint from Stephen Rees. More abstractly, the patch made the
LockMethodData.transactional flag not merely useless but outright
dangerous, because in point of fact that flag no longer tells you anything
at all about whether a lock is held transactionally. This fix therefore
removes that flag altogether. We now rely entirely on the convention
already in use in lock.c that transactional lock holds must be owned by
some ResourceOwner, while session holds are never so owned. Setting the
locallock struct's owner link to NULL thus denotes a session hold, and
there is no redundant marker for that.
PREPARE TRANSACTION now works again when there are session-level advisory
locks, and it is also able to transfer transactional advisory locks to the
prepared transaction, but for implementation reasons it throws an error if
we hold both types of lock on a single lockable object. Perhaps it will be
worth improving that someday.
Assorted other minor cleanup and documentation editing, as well.
Back-patch to 9.1, except that in the 9.1 branch I did not remove the
LockMethodData.transactional flag for fear of causing an ABI break for
any external code that might be examining those structs.
This backatches Heikki's patch in 140a4fbf1a
to make sure the documentation on the website gets updated, since
we're regularly receiving complains about this link.
Normally whole-row Vars are printed as "tabname.*". However, that does not
work at top level of a targetlist, because per SQL standard the parser will
think that the "*" should result in column-by-column expansion; which is
not at all what a whole-row Var implies. We used to just print the table
name in such cases, which works most of the time; but it fails if the table
name matches a column name available anywhere in the FROM clause. This
could lead for instance to a view being interpreted differently after dump
and reload. Adding parentheses doesn't fix it, but there is a reasonably
simple kluge we can use instead: attach a no-op cast, so that the "*" isn't
syntactically at top level anymore. This makes the printing of such
whole-row Vars a lot more consistent with other Vars, and may indeed fix
more cases than just the reported one; I'm suspicious that cases involving
schema qualification probably didn't work properly before, either.
Per bug report and fix proposal from Abbas Butt, though this patch is quite
different in detail from his.
Back-patch to all supported versions.
If it fails to open a new log file, the syslogger assumes there's something
wrong with its parameters (such as log_directory), and stops attempting
automatic time-based or size-based log file rotations. Sending it SIGHUP
is supposed to start that up again. However, the original coding for that
was really bogus, involving clobbering a couple of GUC variables and hoping
that SIGHUP processing would restore them. Get rid of that technique in
favor of maintaining a separate flag showing we've turned rotation off.
Per report from Mark Kirkwood.
Also, the syslogger will automatically attempt to create the log_directory
directory if it doesn't exist, but that was only happening at startup.
For consistency and ease of use, it should do the same whenever the value
of log_directory is changed by SIGHUP.
Back-patch to all supported branches.
Before 9.1, PL/Python functions returning composite types could return
a string and it would be parsed using record_in. The 9.1 changes made
PL/Python only expect dictionaries, tuples, or objects supporting
getattr as output of composite functions, resulting in a regression
and a confusing error message, as the strings were interpreted as
sequences and the code for transforming lists to database tuples was
used. Fix this by treating strings separately as before, before
checking for the other types.
The reason why it's important to support string to database tuple
conversion is that trigger functions on tables with composite columns
get the composite row passed in as a string (from record_out).
Without supporting converting this back using record_in, this makes it
impossible to implement pass-through behavior for these columns, as
PL/Python no longer accepts strings for composite values.
A better solution would be to fix the code that transforms composite
inputs into Python objects to produce dictionaries that would then be
correctly interpreted by the Python->PostgreSQL counterpart code. But
that would be too invasive to backpatch to 9.1, and it is too late in
the 9.2 cycle to attempt it. It should be revisited in the future,
though.
Reported as bug #6559 by Kirill Simonov.
Jan Urbański
setrefs.c failed to do "rtoffset" adjustment of Vars in RETURNING lists,
which meant they were left with the wrong varnos when the RETURNING list
was in a subquery. That was never possible before writable CTEs, of
course, but now it's broken. The executor fails to notice any problem
because ExecEvalVar just references the ecxt_scantuple for any normal
varno; but EXPLAIN breaks when the varno is wrong, as illustrated in a
recent complaint from Bartosz Dmytrak.
Since the eventual rtoffset of the subquery is not known at the time
we are preparing its plan node, the previous scheme of executing
set_returning_clause_references() at that time cannot handle this
adjustment. Fortunately, it turns out that we don't really need to do it
that way, because all the needed information is available during normal
setrefs.c execution; we just have to dig it out of the ModifyTable node.
So, do that, and get rid of the kluge of early setrefs processing of
RETURNING lists. (This is a little bit of a cheat in the case of inherited
UPDATE/DELETE, because we are not passing a "root" struct that corresponds
exactly to what the subplan was built with. But that doesn't matter, and
anyway this is less ugly than early setrefs processing was.)
Back-patch to 9.1, where the problem became possible to hit.
Due to rather sloppy thinking (on my part, I'm afraid) about the
appropriate behavior for boundary conditions, pg_next_dst_boundary() gave
undefined, platform-dependent results when the input time is exactly the
last recorded DST transition time for the specified time zone, as a result
of fetching values one past the end of its data arrays.
Change its specification to be that it always finds the next DST boundary
*after* the input time, and adjust code to match that. The sole existing
caller, DetermineTimeZoneOffset, doesn't actually care about this
distinction, since it always uses a probe time earlier than the instant
that it does care about. So it seemed best to me to change the API to make
the result=1 and result=0 cases more consistent, specifically to ensure
that the "before" outputs always describe the state at the given time,
rather than hacking the code to obey the previous API comment exactly.
Per bug #6605 from Sergey Burladyan. Back-patch to all supported versions.
A number of utility programs were rather careless about paremeters
that can be set via both an option argument and a positional
argument. This leads to results which can violate the Principal
Of Least Astonishment. These changes refuse to use positional
arguments to override settings that have been made via positional
arguments. The changes are backpatched to all live branches.
When using synchronous replication, we waited for the commit record to be
replicated, but if we our transaction didn't write any other WAL records,
that's not required because we don't even flush the WAL locally to disk in
that case. This lead to long waits when committing a transaction that only
modified a temporary table. Bug spotted by Thom Brown.
cost_index tries to estimate the per-tuple costs of evaluating filter
conditions (a/k/a qpquals) by subtracting the estimated cost of the
indexqual conditions from that of the baserestrictinfo conditions. This is
correct so long as the indexquals list is a subset of the baserestrictinfo
list. However, in the presence of derived indexable conditions it's
completely wrong, leading to bogus or even negative scan cost estimates,
as seen for example in bug #6579 from Istvan Endredy. In practice the
problem isn't severe except in the specific case of a LIKE optimization on
a functional index containing a very expensive function.
A proper fix for this might change cost estimates by more than people would
like for stable branches, so in the back branches let's just clamp the cost
difference to be not less than zero. That will at least prevent completely
insane behavior, while not changing the results normally.
This aligns 9.1's behavior with that of older branches. HEAD is now even
laxer, ignoring missing schemas all the time, but that seems like too big
a change for a released branch. Per complaint from Robert Haas.
default tablespace, but part of a database that is in a user-defined
tablespace. Caused "file not found" error during upgrade.
Per bug report from Ants Aasma.
Backpatch to 9.1 and 9.0.
It's still non-deterministic in some sense ... but given fixed settings
and identical planning problems, it will now always choose the same plan,
so we probably shouldn't tar it with that brush. Per bug #6565 from
Guillaume Cottenceau. Back-patch to 9.0 where the behavior was fixed.
estimate_num_groups() gets unhappy with
create table empty();
select * from empty except select * from empty e2;
I can't see any actual use-case for such a query (and the table is illegal
per SQL spec), but it seems like a good idea that it not cause an assert
failure.
This was a thinko in previous commit. Now that stack base pointer is now set
in PostmasterMain and SubPostmasterMain, it doesn't need to be set in
PostgresMain anymore.
We used to only initialize the stack base pointer when starting up a regular
backend, not in other processes. In particular, autovacuum workers can run
arbitrary user code, and without stack-depth checking, infinite recursion
in e.g an index expression will bring down the whole cluster.
The comment about PL/Java using set_stack_base() is not yet true. As the
code stands, PL/java still modifies the stack_base_ptr variable directly.
However, it's been discussed in the PL/Java mailing list that it should be
changed to use the function, because PL/Java is currently oblivious to the
register stack used on Itanium. There's another issues with PL/Java, namely
that the stack base pointer it sets is not really the base of the stack, it
could be something close to the bottom of the stack. That's a separate issue
that might need some further changes to this code, but that's a different
story.
Backpatch to all supported releases.
XLOG_GIN_UPDATE_META_PAGE and XLOG_GIN_DELETE_LISTPAGE records were printed
with a list link field labeled as "blkno", which was confusing, especially
when the link was empty (InvalidBlockNumber). Print the metapage block
number instead, since that's what's actually being updated. We could
include the link values too as a separate field, but not clear it's worth
the trouble.
Back-patch to 8.4 where the dubious code was added.
The original coding of the syslogger had an arbitrary limit of 20 large
messages concurrently in progress, after which it would just punt and dump
message fragments to the output file separately. Our ambitions are a bit
higher than that now, so allow the data structure to expand as necessary.
Reported and patched by Andrew Dunstan; some editing by Tom
dblink_exec leaked temporary database connections if any error occurred
after connection setup, for example
SELECT dblink_exec('...connect string...', 'select 1/0');
Add a PG_TRY block to ensure PQfinish gets done when it is needed.
(dblink_record_internal is on the hairy edge of needing similar treatment,
but seems not to be actively broken at the moment.)
Also, in 9.0 and up, only one of the three functions using tuplestore
return mode was properly checking that the query context would allow
a tuplestore result.
Noted while reviewing dblink patch. Back-patch to all supported branches.
Combining the loop workspace with the record of already-processed objects
might have been a cute trick, but it behaves horridly if there are many
dependency loops to repair: the time spent in the first step of findLoop()
grows as O(N^2). Instead use a separate flag array indexed by dump ID,
which we can check in constant time. The length of the workspace array
is now never more than the actual length of a dependency chain, which
should be reasonably short in all cases of practical interest. The code
is noticeably easier to understand this way, too.
Per gripe from Mike Roest. Since this is a longstanding performance bug,
backpatch to all supported versions.
The loop that matched owned sequences to their owning tables required time
proportional to number of owned sequences times number of tables; although
this work was only expended in selective-dump situations, which is probably
why the issue wasn't recognized long since. Refactor slightly so that we
can perform this work after the index array for findTableByOid has been
set up, reducing the time to O(M log N).
Per gripe from Mike Roest. Since this is a longstanding performance bug,
backpatch to all supported versions.
The DBLINK_GET_CONN and DBLINK_GET_NAMED_CONN macros did not set the
surrounding function's conname variable, causing errors to be incorrectly
reported as having occurred on the "unnamed" connection in some cases.
This bug was actually visible in two cases in the regression tests,
but apparently whoever added those cases wasn't paying attention.
Noted by Kyotaro Horiguchi, though this is different from his proposed
patch.
Back-patch to 8.4; 8.3 does not have the same type of error reporting
so the patch is not relevant.
Initialise ckptXidEpoch from starting checkpoint and maintain the correct
value as we roll forwards. This allows GetNextXidAndEpoch() to return the
correct epoch when executed during recovery. Backpatch to 9.0 when the
problem is first observable by a user.
Bug report from Daniel Farina
The COPY documentation says "COPY FROM matches the input against the null
string before removing backslashes". It is therefore reasonable to presume
that null markers like E'\\0' will work ... and they did, until someone put
the tests in the wrong order during microoptimization-driven rewrites.
Since then, we've been failing if the null marker is something that would
de-escape to an invalidly-encoded string. Since null markers generally
need to be something that can't appear in the data, this represents a
nontrivial loss of functionality; surprising nobody noticed it earlier.
Per report from Jeff Davis. Backpatch to 8.4 where this got broken.
For some reason, in the original coding of the PlaceHolderVar mechanism
I had supposed that PlaceHolderVars couldn't propagate into subqueries.
That is of course entirely possible. When it happens, we need to treat
an outer-level PlaceHolderVar much like an outer Var or Aggref, that is
SS_replace_correlation_vars() needs to replace the PlaceHolderVar with
a Param, and then when building the finished SubPlan we have to provide
the PlaceHolderVar expression as an actual parameter for the SubPlan.
The handling of the contained expression is a bit delicate but it can be
treated exactly like an Aggref's expression.
In addition to the missing logic in subselect.c, prepjointree.c was failing
to search subqueries for PlaceHolderVars that need their relids adjusted
during subquery pullup. It looks like everyplace else that touches
PlaceHolderVars got it right, though.
Per report from Mark Murawski. In 9.1 and HEAD, queries affected by this
oversight would fail with "ERROR: Upper-level PlaceHolderVar found where
not expected". But in 9.0 and 8.4, you'd silently get possibly-wrong
answers, since the value transmitted into the subquery wouldn't go to null
when it should.
Tom Lane