Bruce Momjian [Fri, 1 Feb 2008 02:41:10 +0000 (02:41 +0000)]
Update item:
< * Improve deadlock detection when deleting items from shared buffers
> * Improve deadlock detection when a page cleaning lock conflicts
> with a shared buffer that is pinned
Tom Lane [Thu, 31 Jan 2008 20:29:30 +0000 (20:29 +0000)]
Remove the old table of "supported platforms" in favor of a link to the
buildfarm plus a narrative description of the CPU types and operating systems
on which Postgres is likely to work. Now that we've almost completely
decoupled CPU and OS considerations, the former tabular style isn't all that
enlightening anyway. Perhaps more importantly, no one seems particularly
interested in maintaining the table by hand when we have the buildfarm.
Tom Lane [Thu, 31 Jan 2008 18:58:30 +0000 (18:58 +0000)]
Fix pg_GSS_error to use conn->errorMessage more sanely, ie, actually
work with the PQExpBuffer code instead of fighting it. This avoids an
unnecessary limit on message length and fixes the latent bug that
errorMessage.len wasn't getting set.
Tom Lane [Thu, 31 Jan 2008 18:40:02 +0000 (18:40 +0000)]
Improve pg_autovacuum documentation to clarify that the enabled field cannot
prevent anti-wraparound vacuuming, and to caution against setting unreasonably
small values of freeze_max_age. Also put in a notice that this catalog is
likely to disappear entirely in some future release. Per discussion of
bug #3898 from Steven Flatt.
Magnus Hagander [Thu, 31 Jan 2008 09:21:17 +0000 (09:21 +0000)]
Add pid to the pgident event name on win32.
Should fix a problem where two clusters are running under
two different service accounts and get colliding names,
causing only the first cluster to contain the pgident
event description.
Tom Lane [Wed, 30 Jan 2008 19:46:48 +0000 (19:46 +0000)]
Add checks to TRUNCATE, CLUSTER, and REINDEX to prevent performing these
operations when the current transaction has any open references to the
target relation or index (implying it has an active query using the relation).
The need for this was previously recognized in connection with ALTER TABLE,
but anything that summarily eliminates tuples or moves them around would
confuse an active scan.
While this patch does not in itself fix bug #3883 (the deadlock would happen
before the new check fires), it will discourage people from attempting the
sequence of operations that creates a deadlock risk, so it's at least a
partial response to that problem.
In passing, add a previously-missing check to REINDEX to prevent trying to
reindex another backend's temp table. This isn't a security problem since
only a superuser would get past the schema permission checks, but if we are
testing for this in other utility commands then surely REINDEX should too.
Tom Lane [Wed, 30 Jan 2008 18:35:55 +0000 (18:35 +0000)]
Add a GUC variable "synchronize_seqscans" to allow clients to disable the new
synchronized-scanning behavior, and make pg_dump disable sync scans so that
it will reliably preserve row ordering. Per recent discussions.
Tom Lane [Wed, 30 Jan 2008 04:11:19 +0000 (04:11 +0000)]
Don't putenv() a string that is allocated in a context that will go away
soon. I suspect this explains bug #3902, though I'm still not able to
reproduce that.
Tom Lane [Tue, 29 Jan 2008 02:06:30 +0000 (02:06 +0000)]
Fix up closePGconn() so that PQreset() will work on GSSAPI/SSPI connections;
the patch for those features put its cleanup code into freePGconn() which is
really the wrong place. Remove redundant code from freePGconn() and add
comments in hopes of preventing similar mistakes in future.
Noticed while trying (futilely) to reproduce bug #3902.
Tom Lane [Tue, 29 Jan 2008 02:03:39 +0000 (02:03 +0000)]
Arrange to ignore SIGPIPE during SSL_read() and SSL_shutdown(), as these
are known to write on the socket sometimes and thus we are vulnerable to
being killed by the signal if the server happens to go away unexpectedly.
Noticed while trying (futilely) to reproduce bug #3902.
This bug has been there all along, but since the situation is usually
only of interest to developers, I chose not to back-patch the changes.
Tom Lane [Sun, 27 Jan 2008 19:12:28 +0000 (19:12 +0000)]
Minor editorial improvements in documentation of session_replication_role;
in particular correct the obsolete claim that it can't be changed once
any plans have been cached.
Tom Lane [Sat, 26 Jan 2008 19:55:08 +0000 (19:55 +0000)]
Change StatementCancelHandler() to check the DoingCommandRead flag to decide
whether to execute an immediate interrupt, rather than testing whether
LockWaitCancel() cancelled a lock wait. The old way misclassified the case
where we were blocked in ProcWaitForSignal(), and arguably would misclassify
any other future additions of new ImmediateInterruptOK states too. This
allows reverting the old kluge that gave LockWaitCancel() a return value,
since no callers care anymore. Improve comments in the various
implementations of PGSemaphoreLock() to explain that on some platforms, the
assumption that semop() exits after a signal is wrong, and so we must ensure
that the signal handler itself throws elog if we want cancel or die interrupts
to be effective. Per testing related to bug #3883, though this patch doesn't
solve those problems fully.
Perhaps this change should be back-patched, but since pre-8.3 branches aren't
really relying on autovacuum to respond to SIGINT, it doesn't seem critical
for them.
Tom Lane [Fri, 25 Jan 2008 20:42:10 +0000 (20:42 +0000)]
Fix two different copy-and-paste-os in CSV log rotation logic; one that led to
a double-pfree crash and another that effectively disabled size-based rotation
for CSV logs. Also suppress a memory leak and make some trivial cosmetic
improvements. Per bug #3901 from Chris Hoover and additional code-reading.
Andrew Dunstan [Fri, 25 Jan 2008 15:28:35 +0000 (15:28 +0000)]
Fix example of de-escaping bytea argument, per Florian Weimer. Also fix example
of escaping bytea return value. Both cases did not handle backslash values properly.
Tom Lane [Fri, 25 Jan 2008 04:46:07 +0000 (04:46 +0000)]
Release any detoasted copies of arrays that are made temporarily in
ri_FetchConstraintInfo, to avoid a query-duration memory leak when that
routine is called by RI_FKey_keyequal_upd_fk (which isn't executed in a
short-lived context). This problem was latent when the routine was added
in February, but it didn't become serious until the varvarlena patch made
it quite likely that the fields being examined would be "toasted" (ie, have
short headers). Per report from Stephen Denne.
Tom Lane [Wed, 23 Jan 2008 21:26:13 +0000 (21:26 +0000)]
Prevent integer overflow within the integer-datetimes version of
TimestampTzPlusMilliseconds. An integer argument of more than INT_MAX/1000
milliseconds (ie, about 35 minutes) would provoke a wrong result, resulting
in incorrect enforcement of statement_timestamp values larger than that.
Bug was introduced in my rewrite of 2006-06-20, which fixed some other
overflow risks, but missed this one :-( Per report from Elein.
Andrew Dunstan [Wed, 23 Jan 2008 00:55:47 +0000 (00:55 +0000)]
Document the fix for perl 5.10 with this comment:
* The temporary enabling of the caller opcode here is to work around a
* bug in perl 5.10, which unkindly changed the way its Safe.pm works, without
* notice. It is quite safe, as caller is informational only, and in any case
* we only enable it while we load the 'strict' module.
Tom Lane [Mon, 21 Jan 2008 02:46:11 +0000 (02:46 +0000)]
Fix RS_isRegis() to agree exactly with RS_compile()'s idea of what's a valid
regis. Correct the latter's oversight that a bracket-expression needs to be
terminated. Reduce the ereports to elogs, since they are now not expected to
ever be hit (thus addressing Alvaro's original complaint).
In passing, const-ify the string argument to RS_compile.
Tom Lane [Sun, 20 Jan 2008 21:13:55 +0000 (21:13 +0000)]
Fix psql \h output for case of no parameters (ie, list all the known commands)
to format properly for the actually needed column width, instead of having
a hard-wired assumption about the longest command name length. Also make it
respond to the current screen width. In passing, const-ify the constant
table.
Tom Lane [Sat, 19 Jan 2008 17:43:42 +0000 (17:43 +0000)]
Make pg_regress clean out the testtablespace directory only on Windows.
On other platforms it's better to let the Makefile handle it, but we want
the regression tests to be invokable without make on Windows. A batch
file would be a better solution, but no time for that before 8.3.
Per my discovery that this breaks testing under SELinux, and subsequent
discussion.
Tom Lane [Thu, 17 Jan 2008 21:21:50 +0000 (21:21 +0000)]
Insert into getCopyDataMessage() the same logic that already existed in the
main code path for enlarging libpq's input buffer in one swoop when needing to
read a long data message. Without this, the code will double the buffer size,
read more data, notice it still hasn't got the whole message, and repeat till
it finally has a large enough buffer. Which wastes a lot of data-moving
effort and also memory (since malloc probably can't do anything very useful
with the freed-up smaller buffers). Not sure why this wasn't there already;
certainly the COPY data path is a place where we're quite likely to see long
data messages. I'm not backpatching though, since this is just a marginal
performance issue rather than a real bug.
Tom Lane [Thu, 17 Jan 2008 20:35:27 +0000 (20:35 +0000)]
Fix subselect.c to avoid assuming that a SubLink's testexpr references each
subquery output column exactly once left-to-right. Although this is the case
in the original parser output, it might not be so after rewriting and
constant-folding, as illustrated by bug #3882 from Jan Mate. Instead
scan the subquery's target list to obtain needed per-column information;
this is duplicative of what the parser did, but only a couple dozen lines
need be copied, and we can clean up a couple of notational uglinesses.
Bug was introduced in 8.2 as part of revision of SubLink representation.
Tom Lane [Thu, 17 Jan 2008 18:56:54 +0000 (18:56 +0000)]
Fix ALTER INDEX RENAME so that if the index belongs to a unique or primary key
constraint, the constraint is renamed as well. This avoids inconsistent
situations that could confuse pg_dump (not to mention humans). We might at
some point provide ALTER TABLE RENAME CONSTRAINT as a more general solution,
but there seems no reason not to allow doing it this way too. Per bug #3854
and related discussions.
Tom Lane [Wed, 16 Jan 2008 21:00:25 +0000 (21:00 +0000)]
Remove inappropriate cd commands, per David Wheeler. Also make
the PATH responsive to the installation prefix, which was the apparent
intent of the previous edit, but not well executed.
Teodor Sigaev [Wed, 16 Jan 2008 13:01:03 +0000 (13:01 +0000)]
Fix core dump with buffer-overrun by too long infinitive. Add checking of using
fixed length arrays to prevent array's overrun. Per report by
Hannes Dorbath <light@theendofthetunnel.de> and comments by Tom.
Tom Lane [Tue, 15 Jan 2008 22:18:20 +0000 (22:18 +0000)]
Be less wishy-washy in the documentation and comments about whether a
ParameterStatus message can be sent during COPY OUT: it's definitely
possible, since COPY from a SELECT subquery can trigger any user-defined
function.
Tom Lane [Tue, 15 Jan 2008 18:57:00 +0000 (18:57 +0000)]
Revise memory management for libxml calls. Instead of keeping libxml's data
in whichever context happens to be current during a call of an xml.c function,
use a dedicated context that will not go away until we explicitly delete it
(which we do at transaction end or subtransaction abort). This makes recovery
after an error much simpler --- we don't have to individually delete the data
structures created by libxml. Also, we need to initialize and cleanup libxml
only once per transaction (if there's no error) instead of once per function
call, so it should be a bit faster. We'll need to keep an eye out for
intra-transaction memory leaks, though. Alvaro and Tom.
Bruce Momjian [Mon, 14 Jan 2008 22:14:30 +0000 (22:14 +0000)]
Add:
>
> * Add the ability to automatically create materialized views
>
> Right now materialized views require the user to create triggers on the
> main table to keep the summary table current. SQL syntax should be able
> to manager the triggers and summary table automatically. A more
> sophisticated implementation would automatically retrieve from the
> summary table when the main table is referenced, if possible.
>
Tom Lane [Mon, 14 Jan 2008 19:27:41 +0000 (19:27 +0000)]
Prevent pg_dump from dumping the comment (if any) on the 'public' schema.
This is to avoid uselessly requiring superuser permissions to restore
the dump without errors. Pretty grotty, but no better alternative seems
available, at least not in the near term.
Tom Lane [Mon, 14 Jan 2008 18:46:17 +0000 (18:46 +0000)]
Fix an ancient oversight in libpq's handling of V3-protocol COPY OUT mode:
we need to be able to swallow NOTICE messages, and potentially also
ParameterStatus messages (although the latter would be a bit weird),
without exiting COPY OUT state. Fix it, and adjust the protocol documentation
to emphasize the need for this. Per off-list report from Alexander Galler.
Tom Lane [Mon, 14 Jan 2008 01:39:09 +0000 (01:39 +0000)]
Fix CREATE INDEX CONCURRENTLY so that it won't use synchronized scan for
its second pass over the table. It has to start at block zero, else the
"merge join" logic for detecting which TIDs are already in the index
doesn't work. Hence, extend heapam.c's API so that callers can enable or
disable syncscan. (I put in an option to disable buffer access strategy,
too, just in case somebody needs it.) Per report from Hannes Dorbath.
Tom Lane [Sat, 12 Jan 2008 21:14:08 +0000 (21:14 +0000)]
It turns out the LIBXML_TEST_VERSION macro calls xmlInitParser().
Therefore we must xmlCleanupParser(), or we risk leaving behind
dangling pointers to whatever memory context is current when xml_init()
is called. This seems to fix bug #3860, though we might still want
the more invasive solution being worked on by Alvaro.
Tom Lane [Sat, 12 Jan 2008 00:11:39 +0000 (00:11 +0000)]
Fix logical errors in constraint exclusion: we cannot assume that a CHECK
constraint yields TRUE for every row of its table, only that it does not
yield FALSE (a NULL result isn't disallowed). This breaks a couple of
implications that would be true in two-valued logic. I had put in one such
mistake in an 8.2.5 patch: foo IS NULL doesn't refute a strict operator
on foo. But there was another in the original 8.2 release: NOT foo doesn't
refute an expression whose truth would imply the truth of foo.
Per report from Rajesh Kumar Mallah.
To preserve the ability to do constraint exclusion with one partition
holding NULL values, extend relation_excluded_by_constraints() to check
for attnotnull flags, and add col IS NOT NULL expressions to the set of
constraints we hope to refute.
Tom Lane [Fri, 11 Jan 2008 18:39:41 +0000 (18:39 +0000)]
The original implementation of polymorphic aggregates didn't really get the
checking of argument compatibility right; although the problem is only exposed
with multiple-input aggregates in which some arguments are polymorphic and
some are not. Per bug #3852 from Sokolov Yura.
Tom Lane [Fri, 11 Jan 2008 17:00:45 +0000 (17:00 +0000)]
Fix an old error in clause_selectivity: the default selectivity estimate
for unhandled clause types ought to be 0.5, not 1.0. I fear I introduced
this silliness due to misreading the intent of the very-poorly-structured
code that was there when we inherited the file from Berkeley. The lack
of sanity in this behavior was exposed by an example from Sim Zacks.
(Arguably this is a bug fix and should be back-patched, but I'm a bit
hesitant to introduce a possible planner behavior change in the back
branches; it might detune queries that worked acceptably in the past.)
While at it, make estimation for DistinctExpr do something marginally
realistic, rather than just defaulting.
Tom Lane [Fri, 11 Jan 2008 04:02:18 +0000 (04:02 +0000)]
Fix a conceptual error in my patch of 2007-10-26 that avoided considering
clauseless joins of relations that have unexploited join clauses. Rather
than looking at every other base relation in the query, the correct thing is
to examine the other relations in the "initial_rels" list of the current
make_rel_from_joinlist() invocation, because those are what we actually have
the ability to join against. This might be a subset of the whole query in
cases where join_collapse_limit or from_collapse_limit or full joins have
prevented merging the whole query into a single join problem. This is a bit
untidy because we have to pass those rels down through a new PlannerInfo
field, but it's necessary. Per bug #3865 from Oleg Kharin.
Tom Lane [Fri, 11 Jan 2008 00:54:09 +0000 (00:54 +0000)]
Restructure the shutdown procedure for the archiver process to allow it to
finish archiving everything (when there's no error), and to eliminate various
hazards as best we can. This fixes a previous 8.3 patch that caused the
postmaster to kill and then restart the archiver during shutdown (!?).
The new behavior is that the archiver is allowed to run unmolested until
the bgwriter has exited; then it is sent SIGUSR2 to tell it to do a final
archiving cycle and quit. We only SIGQUIT the archiver if we want a panic
stop; this is important since SIGQUIT will also be sent to any active
archive_command. The postmaster also now doesn't SIGQUIT the stats collector
until the bgwriter is done, since the bgwriter can send stats messages in 8.3.
The postmaster will not exit until both the archiver and stats collector are
gone; this provides some defense (not too bulletproof) against conflicting
archiver or stats collector processes being started by a new postmaster
instance. We continue the prior practice that the archiver will check
for postmaster death immediately before issuing any archive_command; that
gives some additional protection against conflicting archivers.
Also, modify the archiver process to notice SIGTERM and refuse to issue any
more archive commands if it gets it. The postmaster doesn't ever send it
SIGTERM; we assume that any such signal came from init and is a notice of
impending whole-system shutdown. In this situation it seems imprudent to try
to start new archive commands --- if they aren't extremely quick they're
likely to get SIGKILL'd by init.
Tom Lane [Wed, 9 Jan 2008 21:52:36 +0000 (21:52 +0000)]
Fix CREATE INDEX CONCURRENTLY to not deadlock against an automatic or manual
VACUUM that is blocked waiting to get lock on the table being indexed.
Per report and fix suggestion from Greg Stark.
Tom Lane [Wed, 9 Jan 2008 20:42:29 +0000 (20:42 +0000)]
Fix some planner issues found while investigating Kevin Grittner's report
of poorer planning in 8.3 than 8.2:
1. After pushing a constant across an outer join --- ie, given
"a LEFT JOIN b ON (a.x = b.y) WHERE a.x = 42", we can deduce that b.y is
sort of equal to 42, in the sense that we needn't fetch any b rows where
it isn't 42 --- loop to see if any additional deductions can be made.
Previous releases did that by recursing, but I had mistakenly thought that
this was no longer necessary given the EquivalenceClass machinery.
2. Allow pushing constants across outer join conditions even if the
condition is outerjoin_delayed due to a lower outer join. This is safe
as long as the condition is strict and we re-test it at the upper join.
3. Keep the outer-join clause even if we successfully push a constant
across it. This is *necessary* in the outerjoin_delayed case, but
even in the simple case, it seems better to do this to ensure that the
join search order heuristics will consider the join as reasonable to
make. Mark such a clause as having selectivity 1.0, though, since it's
not going to eliminate very many rows after application of the constant
condition.
4. Tweak have_relevant_eclass_joinclause to report that two relations
are joinable when they have vars that are equated to the same constant.
We won't actually generate any joinclause from such an EquivalenceClass,
but again it seems that in such a case it's a good idea to consider
the join as worth costing out.
5. Fix a bug in select_mergejoin_clauses that was exposed by these
changes: we have to reject candidate mergejoin clauses if either side was
equated to a constant, because we can't construct a canonical pathkey list
for such a clause. This is an implementation restriction that might be
worth fixing someday, but it doesn't seem critical to get it done for 8.3.
Tom Lane [Tue, 8 Jan 2008 23:18:51 +0000 (23:18 +0000)]
lmgr.c:DescribeLockTag was never taught about virtual xids, per Greg Stark.
Also a couple of minor tweaks to try to future-proof the code a bit better
against future locktag additions.