OSDN Git Service

pg-rex/syncrep.git
13 years agoBefore exiting walreceiver, fsync() all the WAL received.
Heikki Linnakangas [Mon, 17 Jan 2011 10:22:24 +0000 (12:22 +0200)]
Before exiting walreceiver, fsync() all the WAL received.

Otherwise WAL recovery will replay the un-flushed WAL after walreceiver has
exited, which can lead to a non-recoverable standby if the system crashes hard
at that point.

13 years agoIn test_fsync, use #define for printf format of ops/sec.
Bruce Momjian [Sun, 16 Jan 2011 13:36:43 +0000 (08:36 -0500)]
In test_fsync, use #define for printf format of ops/sec.

13 years agoUse O_DIRECT in O_SYNC test of different size. Restructure O_DIRECT
Bruce Momjian [Sun, 16 Jan 2011 00:40:49 +0000 (19:40 -0500)]
Use O_DIRECT in O_SYNC test of different size.  Restructure O_DIRECT
error reporting to be more consistent.

13 years agoReverse number of stars used for test_fsync details.
Bruce Momjian [Sat, 15 Jan 2011 23:40:10 +0000 (18:40 -0500)]
Reverse number of stars used for test_fsync details.

13 years agoIn test_fsync, warn about options without o_direct that are not used by
Bruce Momjian [Sat, 15 Jan 2011 23:27:10 +0000 (18:27 -0500)]
In test_fsync, warn about options without o_direct that are not used by
Postgres, and cases where o_direct does not work with certain file systems.

13 years agoMove a couple of declarations to reflect where the routines really are.
Tom Lane [Sat, 15 Jan 2011 21:09:05 +0000 (16:09 -0500)]
Move a couple of declarations to reflect where the routines really are.

13 years agoAdd .gitignore to silence git complaints about parser/scanner output files.
Tom Lane [Sat, 15 Jan 2011 21:05:28 +0000 (16:05 -0500)]
Add .gitignore to silence git complaints about parser/scanner output files.

13 years agoHave test_fsync output details that fdatasync is the default
Bruce Momjian [Sat, 15 Jan 2011 20:00:20 +0000 (15:00 -0500)]
Have test_fsync output details that fdatasync is the default
wal_sync_method on Linux.

13 years agoRestructure test_fync to use modular C so there is less duplicate code
Bruce Momjian [Sat, 15 Jan 2011 19:42:12 +0000 (14:42 -0500)]
Restructure test_fync to use modular C so there is less duplicate code
and it can be enhanced easier.

13 years agoEnumerate available tablespaces after starting the backup
Magnus Hagander [Sat, 15 Jan 2011 18:18:14 +0000 (19:18 +0100)]
Enumerate available tablespaces after starting the backup

This closes a race condition where if a tablespace was created
after the enumeration happened but before the do_pg_start_backup()
was called, the backup would be incomplete. Now that it's done
while we are in backup mode, WAL replay will recreate it during
restore.

Noted by Heikki.

13 years agoImprove output display of test_fsync.
Bruce Momjian [Sat, 15 Jan 2011 17:24:05 +0000 (12:24 -0500)]
Improve output display of test_fsync.

13 years agoApply patch for test_fsync to add tests for O_DIRECT.
Bruce Momjian [Sat, 15 Jan 2011 16:54:43 +0000 (11:54 -0500)]
Apply patch for test_fsync to add tests for O_DIRECT.

Adjusted patch by Josh Berkus

13 years agoTreat a WAL sender process that hasn't started streaming yet as a regular
Heikki Linnakangas [Sat, 15 Jan 2011 14:27:40 +0000 (16:27 +0200)]
Treat a WAL sender process that hasn't started streaming yet as a regular
backend, as far as the postmaster shutdown logic is concerned. That means,
fast shutdown will wait for WAL sender processes to exit before signaling
bgwriter to finish. This avoids race conditions between a base backup stopping
or starting, and bgwriter writing the shutdown checkpoint WAL record. We don't
want e.g the end-of-backup WAL record to be written after the shutdown
checkpoint.

13 years agoUse a lexer and grammar for parsing walsender commands
Magnus Hagander [Fri, 14 Jan 2011 15:30:33 +0000 (16:30 +0100)]
Use a lexer and grammar for parsing walsender commands

Makes it easier to parse mainly the BASE_BACKUP command
with it's options, and avoids having to manually deal
with quoted identifiers in the label (previously broken),
and makes it easier to add new commands and options in
the future.

In passing, refactor the case statement in the walsender
to put each command in it's own function.

13 years agoExit from base backups when shutdown is requested
Magnus Hagander [Fri, 14 Jan 2011 11:36:45 +0000 (12:36 +0100)]
Exit from base backups when shutdown is requested

When the exit waits until the whole backup completes, it may take
a very long time.

In passing, add back an error check in the main loop so we detect
clients that disconnect much earlier if the backup is large.

13 years agoCode review for postmaster.pid contents changes.
Tom Lane [Fri, 14 Jan 2011 00:01:28 +0000 (19:01 -0500)]
Code review for postmaster.pid contents changes.

Fix broken test for pre-existing postmaster, caused by wrong code for
appending lines to the lockfile; don't write a failed listen_address
setting into the lockfile; don't arbitrarily change the location of the
data directory in the lockfile compared to previous releases; provide more
consistent and useful definitions of the socket path and listen_address
entries; avoid assuming that pg_ctl has the same DEFAULT_PGSOCKET_DIR as
the postmaster; assorted code style improvements.

13 years agoRevert incorrect memory-conservation hack in inheritance_planner().
Tom Lane [Thu, 13 Jan 2011 19:33:19 +0000 (14:33 -0500)]
Revert incorrect memory-conservation hack in inheritance_planner().

This reverts commit d1001a78ce612a16ea622b558f5fc2b68c45ab4c of 2010-12-05,
which was broken as reported by Jeff Davis.  The problem is that the
individual planning steps may have side-effects on substructures of
PlannerGlobal, not only the current PlannerInfo root.  Arranging to keep
all such side effects in the main planning context is probably possible,
but it would change this from a quick local hack into a wide-ranging and
rather fragile endeavor.  Which it's not worth.

13 years agoMake sure walsender state is only read while holding the spinlock
Magnus Hagander [Thu, 13 Jan 2011 17:51:13 +0000 (18:51 +0100)]
Make sure walsender state is only read while holding the spinlock

Noted by Robert Haas.

13 years agoMore libpq documentation adjustments from Leslie S Satenstein, reviewed
Bruce Momjian [Thu, 13 Jan 2011 17:07:51 +0000 (12:07 -0500)]
More libpq documentation adjustments from Leslie S Satenstein, reviewed
by Robert Haas.

13 years agoFix the logic in libpqrcv_receive() to determine if there's any incoming data
Heikki Linnakangas [Thu, 13 Jan 2011 15:51:28 +0000 (17:51 +0200)]
Fix the logic in libpqrcv_receive() to determine if there's any incoming data
that can be read without blocking. It used to conclude that there isn't, even
though there was data in the socket receive buffer. That lead walreceiver to
flush the WAL after every received chunk, potentially causing big performance
issues.

Backpatch to 9.0, because the performance impact can be very significant.

13 years agoWorkaround for recursive make breakage
Peter Eisentraut [Thu, 13 Jan 2011 07:32:06 +0000 (09:32 +0200)]
Workaround for recursive make breakage

Changing a file two directory levels deep under src/backend/ would not
cause the postgres binary to be rebuilt.  This change fixes it, but no
one knows why.

13 years agoDon't run regression tests in SQL_ASCII encoding by default
Peter Eisentraut [Thu, 13 Jan 2011 07:13:12 +0000 (09:13 +0200)]
Don't run regression tests in SQL_ASCII encoding by default

Instead, run them in the encoding that the locale selects, which is
more representative of real use.

Also document how locale and encoding for regression test runs can be
selected.

13 years agoApply libpq documentation patches submitted by Leslie S Satenstein and
Bruce Momjian [Thu, 13 Jan 2011 01:49:49 +0000 (20:49 -0500)]
Apply libpq documentation patches submitted by Leslie S Satenstein and
reviewed by Robert Haas.

13 years agoFix PlanRowMark/ExecRowMark structures to handle inheritance correctly.
Tom Lane [Thu, 13 Jan 2011 01:47:02 +0000 (20:47 -0500)]
Fix PlanRowMark/ExecRowMark structures to handle inheritance correctly.

In an inherited UPDATE/DELETE, each target table has its own subplan,
because it might have a column set different from other targets.  This
means that the resjunk columns we add to support EvalPlanQual might be
at different physical column numbers in each subplan.  The EvalPlanQual
rewrite I did for 9.0 failed to account for this, resulting in possible
misbehavior or even crashes during concurrent updates to the same row,
as seen in a recent report from Gordon Shannon.  Revise the data structure
so that we track resjunk column numbers separately for each subplan.

I also chose to move responsibility for identifying the physical column
numbers back to executor startup, instead of assuming that numbers derived
during preprocess_targetlist would stay valid throughout subsequent
massaging of the plan.  That's a bit slower, so we might want to consider
undoing it someday; but it would complicate the patch considerably and
didn't seem justifiable in a bug fix that has to be back-patched to 9.0.

13 years agoRevert patch adding support for logging the current role.
Robert Haas [Wed, 12 Jan 2011 16:59:21 +0000 (11:59 -0500)]
Revert patch adding support for logging the current role.

This reverts commit a8a8867912c46a68c9ac14903b3dba2fab8f7097, committed
by me earlier today (2011-01-12).  This isn't safe inside an aborted
transaction.

Noted by Tom Lane.

13 years agoAdd support for logging the current role.
Robert Haas [Wed, 12 Jan 2011 16:34:53 +0000 (11:34 -0500)]
Add support for logging the current role.

Stephen Frost, with some editorialization by me.

13 years agoUnbreak regression tests, apparently broken by commit 4c8e20f
Andrew Dunstan [Wed, 12 Jan 2011 03:27:20 +0000 (22:27 -0500)]
Unbreak regression tests, apparently broken by commit 4c8e20f

13 years agoRe-add recursive coverage target in src/backend/
Peter Eisentraut [Tue, 11 Jan 2011 22:26:20 +0000 (00:26 +0200)]
Re-add recursive coverage target in src/backend/

This was lost during the recent recursive make change.

13 years agoAdd some subsection headings
Peter Eisentraut [Tue, 11 Jan 2011 20:47:58 +0000 (22:47 +0200)]
Add some subsection headings

13 years agoTrack walsender state in shared memory and expose in pg_stat_replication
Magnus Hagander [Tue, 11 Jan 2011 20:25:28 +0000 (21:25 +0100)]
Track walsender state in shared memory and expose in pg_stat_replication

13 years agoAdd missing function prototype, for consistency
Magnus Hagander [Tue, 11 Jan 2011 20:12:12 +0000 (21:12 +0100)]
Add missing function prototype, for consistency

13 years agoAdjust basebackup.c to suppress compiler warnings.
Tom Lane [Tue, 11 Jan 2011 18:41:13 +0000 (13:41 -0500)]
Adjust basebackup.c to suppress compiler warnings.

Some versions of gcc complain about "variable `tablespaces' might be
clobbered by `longjmp' or `vfork'" with the original coding.  Fix by
moving the PG_TRY block into a separate subroutine.

13 years agoTweak create_index_paths()'s test for whether to consider a bitmap scan.
Tom Lane [Tue, 11 Jan 2011 17:12:04 +0000 (12:12 -0500)]
Tweak create_index_paths()'s test for whether to consider a bitmap scan.

Per my note of a couple days ago, create_index_paths would refuse to
consider any path at all for GIN indexes if the selectivity estimate came
out as 1.0; not even if you tried to force it with enable_seqscan.  While
this isn't really a bad outcome in practice, it could be annoying for
testing purposes.  Adjust the test for "is this path only useful for
sorting" so that it doesn't fire on paths with nil pathkeys, which will
include all GIN paths.

13 years agoTypo fix
Magnus Hagander [Tue, 11 Jan 2011 09:45:16 +0000 (10:45 +0100)]
Typo fix

Josh Kupershmidt

13 years agoReset walsender ps title in the main loop
Magnus Hagander [Tue, 11 Jan 2011 09:04:54 +0000 (10:04 +0100)]
Reset walsender ps title in the main loop

When in streaming mode we can never get out, so it will never
be required, but after a base backup (or other operations)
we can get back to the loop, so the title needs to be cleared.

13 years agoSet process title to indicate base backup is running
Magnus Hagander [Mon, 10 Jan 2011 20:50:10 +0000 (21:50 +0100)]
Set process title to indicate base backup is running

13 years agoLeave temporary files out of streaming base backups.
Heikki Linnakangas [Mon, 10 Jan 2011 17:42:05 +0000 (19:42 +0200)]
Leave temporary files out of streaming base backups.

13 years agoA toast relid field are no longer needed in pg_upgrade's rel arrays, so
Bruce Momjian [Mon, 10 Jan 2011 16:45:22 +0000 (11:45 -0500)]
A toast relid field are no longer needed in pg_upgrade's rel arrays, so
remove them.  Also other renaming.

13 years agoBackend support for streaming base backups
Magnus Hagander [Mon, 10 Jan 2011 13:03:55 +0000 (14:03 +0100)]
Backend support for streaming base backups

Add BASE_BACKUP command to walsender, allowing it to stream a
base backup to the client (in tar format). The syntax is still
far from ideal, that will be fixed in the switch to use a proper
grammar for walsender.

No client included yet, will come as a separate commit.

Magnus Hagander and Heikki Linnakangas

13 years agoUpdate contrib/pg_trgm for new GIN extractQuery API.
Tom Lane [Sun, 9 Jan 2011 23:04:20 +0000 (18:04 -0500)]
Update contrib/pg_trgm for new GIN extractQuery API.

No actual change in functionality ... just get rid of uselessly complex
code to pass the number of keys via extra_data.

13 years agoUpdate contrib/hstore for new GIN extractQuery API.
Tom Lane [Sun, 9 Jan 2011 21:43:56 +0000 (16:43 -0500)]
Update contrib/hstore for new GIN extractQuery API.

In particular, make hstore @> '' succeed for all hstores, likewise
hstore ?& '{}'.  Previously the results were inconsistent and could
depend on whether you were using a GiST index, GIN index, or seqscan.

13 years agoImprove comment.
Tom Lane [Sun, 9 Jan 2011 20:24:22 +0000 (15:24 -0500)]
Improve comment.

13 years agoSplit pg_start_backup() and pg_stop_backup() into two pieces
Magnus Hagander [Sun, 9 Jan 2011 20:00:28 +0000 (21:00 +0100)]
Split pg_start_backup() and pg_stop_backup() into two pieces

Move the actual functionality into a separate function that's
easier to call internally, and change the SQL-callable function
to be a wrapper calling this.

Also create a pg_abort_backup() function, only callable internally,
that does only the most vital parts of pg_stop_backup(), making it
safe(r) to call from error handlers.

13 years agoFix crash in the new GiST insertion code, when an update splits the root page.
Heikki Linnakangas [Sun, 9 Jan 2011 19:09:58 +0000 (21:09 +0200)]
Fix crash in the new GiST insertion code, when an update splits the root page.

This bug was exercised by contrib/intarray/bench, as noted by Tom Lane.

13 years agoFix up core tsquery GIN support for new extractQuery API.
Tom Lane [Sun, 9 Jan 2011 19:34:50 +0000 (14:34 -0500)]
Fix up core tsquery GIN support for new extractQuery API.

No need for the empty-prefix-match kluge to force a full scan anymore.

13 years agoUse array_contains_nulls instead of ARR_HASNULL on user-supplied arrays.
Tom Lane [Sun, 9 Jan 2011 18:09:07 +0000 (13:09 -0500)]
Use array_contains_nulls instead of ARR_HASNULL on user-supplied arrays.

This applies the fix for bug #5784 to remaining places where we wish
to reject nulls in user-supplied arrays.  In all these places, there's
no reason not to allow a null bitmap to be present, so long as none of
the current elements are actually null.

I did not change some other places where we are looking at system catalog
entries or aggregate transition values, as the presence of a null bitmap
in such an array would be suspicious.

13 years agoEnsure the directory for gram.h is created on win32
Magnus Hagander [Sun, 9 Jan 2011 16:01:15 +0000 (17:01 +0100)]
Ensure the directory for gram.h is created on win32

Result of bad testing of my last commit.

13 years agoProperly install gram.h on MSVC builds
Magnus Hagander [Sun, 9 Jan 2011 14:31:48 +0000 (15:31 +0100)]
Properly install gram.h on MSVC builds

This file is now needed by pgAdmin builds, which started
failing since it was missing in the installer builds.

13 years agoAdd pgreadlink() on Windows to read junction points
Magnus Hagander [Sun, 9 Jan 2011 14:06:55 +0000 (15:06 +0100)]
Add pgreadlink() on Windows to read junction points

Add support for reading back information about the symbolic
links we've created with pgsymlink(), which are actually
Junction Points. Just like pgsymlink() can only create directory
symlinks, pgreadlink() can only read directory symlinks.

13 years agoThere is no need to have to identical functions in ecpg thus removing one of them.
Michael Meskes [Sun, 9 Jan 2011 01:45:56 +0000 (02:45 +0100)]
There is no need to have to identical functions in ecpg thus removing one of them.

13 years agoFix assorted corner-case bugs in contrib/intarray.
Tom Lane [Sun, 9 Jan 2011 05:39:21 +0000 (00:39 -0500)]
Fix assorted corner-case bugs in contrib/intarray.

The array containment operators now behave per mathematical expectation
for empty arrays (ie, an empty array is contained in anything).
Both these operators and the query_int operators now work as expected in
GiST and GIN index searches, rather than having corner cases where the
index searches gave different answers.

Also, fix unexpected failures where the operators would claim that an array
contained nulls, when in fact there was no longer any null present (similar
to bug #5784).  The restriction to not have nulls is still there, as
removing it would take a lot of added code complexity and probably slow
things down significantly.

Also, remove the arbitrary restriction to 1-D arrays; unlike the other
restriction, this was buying us nothing performance-wise.

Assorted cosmetic improvements and marginal performance improvements, too.

13 years agoAdd array_contains_nulls() function in arrayfuncs.c.
Tom Lane [Sun, 9 Jan 2011 01:24:08 +0000 (20:24 -0500)]
Add array_contains_nulls() function in arrayfuncs.c.

This will support fixing contrib/intarray (and probably other places)
so that they don't have to fail on arrays that contain a null bitmap
but no live null entries.

13 years agoFix up gincostestimate for new extractQuery API.
Tom Lane [Sun, 9 Jan 2011 01:13:33 +0000 (20:13 -0500)]
Fix up gincostestimate for new extractQuery API.

The only reason this wasn't crashing while testing the core anyarray
operators was that it was disabled for those cases because of passing the
wrong type information to get_opfamily_proc :-(.  So fix that too, and make
it insist on finding the support proc --- in hindsight, silently doing
nothing is not as sane a coping mechanism as all that.

13 years agoIn ecpg's parser removed a fixed length limit for constants defining an array dimension.
Michael Meskes [Sat, 8 Jan 2011 17:35:37 +0000 (18:35 +0100)]
In ecpg's parser removed a fixed length limit for constants defining an array dimension.

13 years agoRemove pg_am.amindexnulls.
Tom Lane [Sat, 8 Jan 2011 21:08:05 +0000 (16:08 -0500)]
Remove pg_am.amindexnulls.

The only use we have had for amindexnulls is in determining whether an
index is safe to cluster on; but since the addition of the amclusterable
flag, that usage is pretty redundant.

In passing, clean up assorted sloppiness from the last patch that touched
pg_am.h: Natts_pg_am was wrong, and ambuildempty was not documented.

13 years agoRefactor GIN's handling of duplicate search entries.
Tom Lane [Sat, 8 Jan 2011 19:47:13 +0000 (14:47 -0500)]
Refactor GIN's handling of duplicate search entries.

The original coding could combine duplicate entries only when they
originated from the same qual condition.  In particular it could not
combine cases where multiple qual conditions all give rise to full-index
scan requests, which is an expensive case well worth optimizing.  Refactor
so that duplicates are recognized across all the quals.

13 years agoIn pg_upgrade, remove functions that did sequential array scans looking
Bruce Momjian [Sat, 8 Jan 2011 18:44:44 +0000 (13:44 -0500)]
In pg_upgrade, remove functions that did sequential array scans looking
up relations, but rather order old/new relations and use the same array
index value for both.  This should speed up pg_upgrade for databases
with many relations.

13 years agoIn pg_upgrade, clarify use of install_db_support_functions().
Bruce Momjian [Sat, 8 Jan 2011 14:11:48 +0000 (09:11 -0500)]
In pg_upgrade, clarify use of install_db_support_functions().

13 years agoIn pg_upgrade, remove unnecessary separate handling of toast tables now
Bruce Momjian [Sat, 8 Jan 2011 13:01:12 +0000 (08:01 -0500)]
In pg_upgrade, remove unnecessary separate handling of toast tables now
that we restore by oid;  they can be handled like regular tables when
creating the file mapping structure.

13 years agoRemove bogus claims regarding createuser defaults.
Robert Haas [Sat, 8 Jan 2011 11:12:05 +0000 (06:12 -0500)]
Remove bogus claims regarding createuser defaults.

Josh Kupershmidt

13 years agoUpdate GIN support function definitions for contrib/intarray.
Tom Lane [Sat, 8 Jan 2011 04:34:06 +0000 (23:34 -0500)]
Update GIN support function definitions for contrib/intarray.

The underlying C code still needs work, but this at least gets its
current regression test passing again.

13 years agoDocument why pg_upgrade preserves pg_auth.oid (because they are stored
Bruce Momjian [Sat, 8 Jan 2011 04:02:03 +0000 (23:02 -0500)]
Document why pg_upgrade preserves pg_auth.oid (because they are stored
in pg_largeobject_metadata).

13 years agoMore pg_upgrade relfilenode C comments added.
Bruce Momjian [Sat, 8 Jan 2011 03:57:30 +0000 (22:57 -0500)]
More pg_upgrade relfilenode C comments added.

13 years agoSimplify pg_upgrade C comment about what is preserved.
Bruce Momjian [Sat, 8 Jan 2011 03:46:41 +0000 (22:46 -0500)]
Simplify pg_upgrade C comment about what is preserved.

13 years agoIn pg_upgrade, track only one copy of namespace/relname in FileNameMap
Bruce Momjian [Sat, 8 Jan 2011 03:36:51 +0000 (22:36 -0500)]
In pg_upgrade, track only one copy of namespace/relname in FileNameMap
because the old and new values are identical.

13 years agoFix C indenting in pg_upgrade function.
Bruce Momjian [Sat, 8 Jan 2011 03:01:44 +0000 (22:01 -0500)]
Fix C indenting in pg_upgrade function.

13 years agoFix pg_upgrade of large object permissions by preserving pg_auth.oid,
Bruce Momjian [Sat, 8 Jan 2011 02:59:29 +0000 (21:59 -0500)]
Fix pg_upgrade of large object permissions by preserving pg_auth.oid,
which is stored in pg_largeobject_metadata.

No backpatch to 9.0 because you can't migrate from 9.0 to 9.0 with the
same catversion (because of tablespace conflict), and a pre-9.0
migration to 9.0 has not large object permissions to migrate.

13 years agoForce pg_upgrade's to preserve pg_class.oid, not pg_class.relfilenode.
Bruce Momjian [Sat, 8 Jan 2011 02:25:34 +0000 (21:25 -0500)]
Force pg_upgrade's to preserve pg_class.oid, not pg_class.relfilenode.
Toast tables have identical pg_class.oid and pg_class.relfilenode, but
for clarity it is good to preserve the pg_class.oid.

Update comments regarding what is preserved, and do some
variable/function renaming for clarity.

13 years agoOoops, forgot to fix contrib/tsearch2's references to GIN support funcs.
Tom Lane [Sat, 8 Jan 2011 02:10:48 +0000 (21:10 -0500)]
Ooops, forgot to fix contrib/tsearch2's references to GIN support funcs.

13 years agoFix the built-in GIN support procedure declarations in pg_proc.h.
Tom Lane [Sat, 8 Jan 2011 01:40:48 +0000 (20:40 -0500)]
Fix the built-in GIN support procedure declarations in pg_proc.h.

Add more "internal" arguments so that these pg_proc entries reflect the
current preferred API.  This is purely a cosmetic change, since GIN doesn't
actually consult the pg_proc entry when calling a support function.
Accordingly, no catversion bump.

13 years agoFix GIN to support null keys, empty and null items, and full index scans.
Tom Lane [Sat, 8 Jan 2011 00:16:24 +0000 (19:16 -0500)]
Fix GIN to support null keys, empty and null items, and full index scans.

Per my recent proposal(s).  Null key datums can now be returned by
extractValue and extractQuery functions, and will be stored in the index.
Also, placeholder entries are made for indexable items that are NULL or
contain no keys according to extractValue.  This means that the index is
now always complete, having at least one entry for every indexed heap TID,
and so we can get rid of the prohibition on full-index scans.  A full-index
scan is implemented much the same way as partial-match scans were already:
we build a bitmap representing all the TIDs found in the index, and then
drive the results off that.

Also, introduce a concept of a "search mode" that can be requested by
extractQuery when the operator requires matching to empty items (this is
just as cheap as matching to a single key) or requires a full index scan
(which is not so cheap, but it sure beats failing or giving wrong answers).
The behavior remains backward compatible for opclasses that don't return
any null keys or request a non-default search mode.

Using these features, we can now make the GIN index opclass for anyarray
behave in a way that matches the actual anyarray operators for &&, <@, @>,
and = ... which it failed to do before in assorted corner cases.

This commit fixes the core GIN code and ginarrayprocs.c, updates the
documentation, and adds some simple regression test cases for the new
behaviors using the array operators.  The tsearch and contrib GIN opclass
support functions still need to be looked over and probably fixed.

Another thing I intend to fix separately is that this is pretty inefficient
for cases where more than one scan condition needs a full-index search:
we'll run duplicate GinScanEntrys, each one of which builds a large bitmap.
There is some existing logic to merge duplicate GinScanEntrys but it needs
refactoring to make it work for entries belonging to different scan keys.

Note that most of gin.h has been split out into a new file gin_private.h,
so that gin.h doesn't export anything that's not supposed to be used by GIN
opclasses or the rest of the backend.  I did quite a bit of other code
beautification work as well, mostly fixing comments and choosing more
appropriate names for things.

13 years agoDocument pg_stat_replication, bump catversion since that was overlooked.
Robert Haas [Fri, 7 Jan 2011 16:06:55 +0000 (11:06 -0500)]
Document pg_stat_replication, bump catversion since that was overlooked.

Itagaki Takahiro, edited by me.

13 years agoImprove recovery.conf.sample comments.
Robert Haas [Fri, 7 Jan 2011 15:49:29 +0000 (10:49 -0500)]
Improve recovery.conf.sample comments.

Jehan-Guillaume de Rorthais, with some additional wordsmithing by me.

13 years agoNew system view pg_stat_replication displays activity of wal sender processes.
Itagaki Takahiro [Fri, 7 Jan 2011 11:35:38 +0000 (20:35 +0900)]
New system view pg_stat_replication displays activity of wal sender processes.

Itagaki Takahiro and Simon Riggs.

13 years agoImprove C comments about backend variables set by pg_upgrade_support
Bruce Momjian [Fri, 7 Jan 2011 03:44:57 +0000 (22:44 -0500)]
Improve C comments about backend variables set by pg_upgrade_support
functions.

13 years agoImprove markup of unnest example
Peter Eisentraut [Thu, 6 Jan 2011 21:54:05 +0000 (23:54 +0200)]
Improve markup of unnest example

13 years agoImprove array_upper example
Peter Eisentraut [Thu, 6 Jan 2011 21:31:20 +0000 (23:31 +0200)]
Improve array_upper example

The previous example didn't make it clear whether array_upper returned
the last element or the index of the last element.

13 years agoUpdate sequence_1.out for recent changes in sequence regression test.
Tom Lane [Thu, 6 Jan 2011 15:58:32 +0000 (10:58 -0500)]
Update sequence_1.out for recent changes in sequence regression test.

13 years agoUpdate documentation to say that \lo_import sets :LASTOID, not
Bruce Momjian [Thu, 6 Jan 2011 02:32:10 +0000 (21:32 -0500)]
Update documentation to say that \lo_import sets :LASTOID, not
lo_insert.

13 years agoRename pg_upgrade variables, for clarity.
Bruce Momjian [Thu, 6 Jan 2011 01:14:59 +0000 (20:14 -0500)]
Rename pg_upgrade variables, for clarity.

13 years agoUpdate pg_upgrade C comments.
Bruce Momjian [Wed, 5 Jan 2011 19:09:01 +0000 (14:09 -0500)]
Update pg_upgrade C comments.

13 years agoFix for recent pg_upgrade commit to fix large objects relfilenode
Bruce Momjian [Wed, 5 Jan 2011 18:52:44 +0000 (13:52 -0500)]
Fix for recent pg_upgrade commit to fix large objects relfilenode
handling.  (metadata user ids still an open issue).

13 years agoIn pg_upgrade, adjust transfer_relfile()'s parameters to be more logical.
Bruce Momjian [Wed, 5 Jan 2011 17:07:45 +0000 (12:07 -0500)]
In pg_upgrade, adjust transfer_relfile()'s parameters to be more logical.

13 years agoClarify pg_upgrade's creation of the map file structure. Also clean
Bruce Momjian [Wed, 5 Jan 2011 16:37:08 +0000 (11:37 -0500)]
Clarify pg_upgrade's creation of the map file structure.  Also clean
up pg_dump's calling of pg_upgrade_support functions.

13 years agoGive superusers REPLIACTION permission by default
Magnus Hagander [Wed, 5 Jan 2011 13:24:17 +0000 (14:24 +0100)]
Give superusers REPLIACTION permission by default

This can be overriden by using NOREPLICATION on the CREATE ROLE
statement, but by default they will have it, making it backwards
compatible and "less surprising" (given that superusers normally
override all checks).

13 years agoIn pg_upgrade, copy pg_largeobject_metadata and its index for 9.0+
Bruce Momjian [Wed, 5 Jan 2011 04:35:49 +0000 (23:35 -0500)]
In pg_upgrade, copy pg_largeobject_metadata and its index for 9.0+
servers because, like pg_largeobject, it is a system table whose
contents are not dumped by pg_dump --schema-only.

13 years agoIn pg_upgrade, fix backward logging display of link operations.
Bruce Momjian [Wed, 5 Jan 2011 02:33:36 +0000 (21:33 -0500)]
In pg_upgrade, fix backward logging display of link operations.

13 years agoSimplify functions and parameters used by pg_upgrade.
Bruce Momjian [Wed, 5 Jan 2011 00:11:00 +0000 (19:11 -0500)]
Simplify functions and parameters used by pg_upgrade.

13 years agoImprove psql tab completion for CREATE/ALTER ROLE [NO]REPLICATION.
Itagaki Takahiro [Tue, 4 Jan 2011 08:56:01 +0000 (17:56 +0900)]
Improve psql tab completion for CREATE/ALTER ROLE [NO]REPLICATION.

Missing support for VALID UNTIL in CREATE ROLE is also added.

13 years agoFix crash in ALTER OPERATOR CLASS/FAMILY .. SET SCHEMA.
Robert Haas [Tue, 4 Jan 2011 03:08:55 +0000 (22:08 -0500)]
Fix crash in ALTER OPERATOR CLASS/FAMILY .. SET SCHEMA.

In the previous coding, the parser emitted a List containing a C string,
which is no good, because copyObject() can't handle it.

Dimitri Fontaine

13 years agoUpdate comments in RecordTransactionCommit() to mention unlogged tables.
Robert Haas [Mon, 3 Jan 2011 15:29:22 +0000 (10:29 -0500)]
Update comments in RecordTransactionCommit() to mention unlogged tables.

13 years agoBump catversion, forgot in previous commit.
Magnus Hagander [Mon, 3 Jan 2011 11:50:30 +0000 (12:50 +0100)]
Bump catversion, forgot in previous commit.

13 years agoAdd views and functions to monitor hot standby query conflicts
Magnus Hagander [Mon, 3 Jan 2011 11:46:03 +0000 (12:46 +0100)]
Add views and functions to monitor hot standby query conflicts

Add the view pg_stat_database_conflicts and a column to pg_stat_database,
and the underlying functions to provide the information.

13 years agoAdd missing part of replication role docs
Magnus Hagander [Mon, 3 Jan 2011 11:00:09 +0000 (12:00 +0100)]
Add missing part of replication role docs

Noted by Peter E.

13 years agoperltidy run on the MSVC build system
Magnus Hagander [Mon, 3 Jan 2011 09:44:56 +0000 (10:44 +0100)]
perltidy run on the MSVC build system

Forgot this with previuos commit, line it up so it's easier to
submit (readable) patches against the MSVC build system.

13 years agoImplement remaining fields of information_schema.sequences view
Peter Eisentraut [Sun, 2 Jan 2011 13:08:08 +0000 (15:08 +0200)]
Implement remaining fields of information_schema.sequences view

Add new function pg_sequence_parameters that returns a sequence's start,
minimum, maximum, increment, and cycle values, and use that in the view.
(bug #5662; design suggestion by Tom Lane)

Also slightly adjust the view's column order and permissions after review of
SQL standard.

13 years agoFix typo.
Robert Haas [Sun, 2 Jan 2011 12:26:10 +0000 (07:26 -0500)]
Fix typo.

Noted by Magnus Hagander.

13 years agoBasic foreign table support.
Robert Haas [Sun, 2 Jan 2011 04:48:11 +0000 (23:48 -0500)]
Basic foreign table support.

Foreign tables are a core component of SQL/MED.  This commit does
not provide a working SQL/MED infrastructure, because foreign tables
cannot yet be queried.  Support for foreign table scans will need to
be added in a future patch.  However, this patch creates the necessary
system catalog structure, syntax support, and support for ancillary
operations such as COMMENT and SECURITY LABEL.

Shigeru Hanada, heavily revised by Robert Haas

13 years agoMinor wordsmithing.
Robert Haas [Sat, 1 Jan 2011 22:50:31 +0000 (17:50 -0500)]
Minor wordsmithing.

As suggested by Tom Lane, in response to a gripe from Leslie S Satenstein.

13 years agoFix pg_dump support for security labels on columns.
Robert Haas [Sat, 1 Jan 2011 22:43:37 +0000 (17:43 -0500)]
Fix pg_dump support for security labels on columns.

Along the way, correct an erroneous comment.