]> granicus.if.org Git - postgresql/log
postgresql
18 years agoDelay write of pg_stats file to once every five minutes, during
Bruce Momjian [Thu, 27 Apr 2006 00:06:59 +0000 (00:06 +0000)]
Delay write of pg_stats file to once every five minutes, during
shutdown, or when requested by a backend:

It changes so the file is only written once every 5 minutes (changeable
of course, I just picked something) instead of once every half second.
It's still written when the stats collector shuts down, just as before.
And it is now also written on backend request. A backend requests a
rewrite by simply sending a special stats message. It operates on the
assumption that the backends aren't actually going to read the
statistics file very often, compared to how frequent it's written today.

Magnus Hagander

18 years agoAdd tablespace display to psql \l+.
Bruce Momjian [Wed, 26 Apr 2006 23:15:45 +0000 (23:15 +0000)]
Add tablespace display to psql \l+.

Philip Yarra

18 years agoImprove /contrib/pg_buffercache installation to use BEGIN/COMMIT,
Bruce Momjian [Wed, 26 Apr 2006 22:50:17 +0000 (22:50 +0000)]
Improve /contrib/pg_buffercache installation to use BEGIN/COMMIT,
improve examples.

Mark Kirkwood

18 years agoSplit out pg_freespace views to one for relations and another for pages,
Bruce Momjian [Wed, 26 Apr 2006 22:46:09 +0000 (22:46 +0000)]
Split out pg_freespace views to one for relations and another for pages,
pg_freespacemap_relations and pg_freespacemap_pages.

Mark Kirkwood

18 years agoAdjust /contrib/pg_freespace to show index free space as NULL (FSM only
Bruce Momjian [Wed, 26 Apr 2006 22:41:18 +0000 (22:41 +0000)]
Adjust /contrib/pg_freespace to show index free space as NULL (FSM only
tracks index pages, not free space on pages):

1/ Index free bytes set to NULL
2/ Comment added to the README briefly mentioning the index business
3/ Columns reordered more logically
4/ 'Blockid' column removed
5/ Free bytes column renamed to just 'bytes' instead of 'blockfreebytes'

Mark Kirkwood

18 years agoMove ltree parentsel() selectivity function into /contrib/ltree.
Bruce Momjian [Wed, 26 Apr 2006 22:33:36 +0000 (22:33 +0000)]
Move ltree parentsel() selectivity function into /contrib/ltree.

18 years agoIn pg_resetxlog.c, uint -> uint32, for Win32 port.
Bruce Momjian [Wed, 26 Apr 2006 21:52:31 +0000 (21:52 +0000)]
In pg_resetxlog.c, uint -> uint32, for Win32 port.

18 years agoAdd missing ControlFile.checkPointCopy.ThisTimeLineID line for 'guess'
Bruce Momjian [Wed, 26 Apr 2006 18:56:56 +0000 (18:56 +0000)]
Add missing ControlFile.checkPointCopy.ThisTimeLineID line for 'guess'
pg_resetxlog.

Simon

18 years agoUpdate catalog version for ltree changes.
Bruce Momjian [Wed, 26 Apr 2006 18:30:10 +0000 (18:30 +0000)]
Update catalog version for ltree changes.

18 years agoEnhanced containment selectivity function for /contrib/ltree
Bruce Momjian [Wed, 26 Apr 2006 18:28:34 +0000 (18:28 +0000)]
Enhanced containment selectivity function for /contrib/ltree

Matteo Beccati

18 years agoAllow pg_resetxlog -f to reset pg_control counters using xlog
Bruce Momjian [Wed, 26 Apr 2006 02:17:15 +0000 (02:17 +0000)]
Allow pg_resetxlog -f to reset pg_control counters using xlog
information, and add a -r option to reset pg_control without affecting
xlog.

yuanjia lee

18 years agoDone:
Bruce Momjian [Wed, 26 Apr 2006 02:15:20 +0000 (02:15 +0000)]
Done:

* -Allow WAL information to recover corrupted pg_controldata

18 years agoRevise large-object access routines to avoid running with CurrentMemoryContext
Tom Lane [Wed, 26 Apr 2006 00:34:57 +0000 (00:34 +0000)]
Revise large-object access routines to avoid running with CurrentMemoryContext
set to the large object context ("fscxt"), as this is inevitably a source of
transaction-duration memory leaks.  Not sure why we'd not noticed it before;
maybe people weren't touching a whole lot of LOs in the same transaction
before the 8.1 pg_dump changes.  Per report from Wayne Conrad.

Backpatched as far as 8.1, but the problem doubtless goes all the way back.
I'm disinclined to spend the time to try to verify that the older branches
would still work if patched, seeing that this code was significantly modified
for 8.0 and again for 8.1, and that we don't have any trouble reports before
8.1.  (Maybe the leaks were smaller before?)

18 years agoArrange to cache btree metapage data in the relcache entry for the index,
Tom Lane [Tue, 25 Apr 2006 22:46:05 +0000 (22:46 +0000)]
Arrange to cache btree metapage data in the relcache entry for the index,
thereby saving a visit to the metapage in most index searches/updates.
This wouldn't actually save any I/O (since in the old regime the metapage
generally stayed in cache anyway), but it does provide a useful decrease
in bufmgr traffic in high-contention scenarios.  Per my recent proposal.

18 years agoAdjust SGML spacing.
Bruce Momjian [Tue, 25 Apr 2006 21:02:33 +0000 (21:02 +0000)]
Adjust SGML spacing.

18 years agoBack out the rest of the RESET CONNECTION patch.
Tom Lane [Tue, 25 Apr 2006 18:58:37 +0000 (18:58 +0000)]
Back out the rest of the RESET CONNECTION patch.

18 years agoThe 8.1 planner removes WHERE quals from the plan when the quals are
Tom Lane [Tue, 25 Apr 2006 16:54:09 +0000 (16:54 +0000)]
The 8.1 planner removes WHERE quals from the plan when the quals are
implied by the predicate of a partial index being used to scan a table.
However, this optimization is unsafe in an UPDATE, DELETE, or SELECT FOR
UPDATE query, because the quals need to be rechecked by EvalPlanQual if
there's an update conflict.  Per example from Jean-Samuel Reynaud.

18 years agoRemove equals sign from environment variable with not defaults for
Bruce Momjian [Tue, 25 Apr 2006 15:19:16 +0000 (15:19 +0000)]
Remove equals sign from environment variable with not defaults for
configure.

18 years agoMake configure environment variable options without defaults not have an
Bruce Momjian [Tue, 25 Apr 2006 15:17:04 +0000 (15:17 +0000)]
Make configure environment variable options without defaults not have an
equals sign.

18 years agoAdd documentation about configure environment variabbles used.
Bruce Momjian [Tue, 25 Apr 2006 15:14:16 +0000 (15:14 +0000)]
Add documentation about configure environment variabbles used.

 David Wheeler

18 years agoCall SET varaiables "configuration_parameter"s in the documentation.
Bruce Momjian [Tue, 25 Apr 2006 14:56:04 +0000 (14:56 +0000)]
Call SET varaiables "configuration_parameter"s in the documentation.

18 years agoChange "parameter" to "varname" in the alter role and alter user
Bruce Momjian [Tue, 25 Apr 2006 14:47:29 +0000 (14:47 +0000)]
Change "parameter" to "varname" in the alter role and alter user
documentation sections.  Also update SET/RESET to be consistent.

Nicolas Barbier

18 years agoI now see we support RESET SESSION AUTHORIZATION, so it seems we have to
Bruce Momjian [Tue, 25 Apr 2006 14:33:41 +0000 (14:33 +0000)]
I now see we support RESET SESSION AUTHORIZATION, so it seems we have to
use RESET CONNECTION:

< * Add RESET SESSION command to reset all session state
> * Add RESET CONNECTION command to reset all session state
447c447
<   notify the protocol when a RESET SESSION command is used.
>   notify the protocol when a RESET CONNECTION command is used.

18 years agoRESET SESSION is more precise:
Bruce Momjian [Tue, 25 Apr 2006 14:31:03 +0000 (14:31 +0000)]
RESET SESSION is more precise:

< * Add RESET CONNECTION command to reset all session state
> * Add RESET SESSION command to reset all session state
447c447
<   notify the protocol when a RESET CONNECTION command is used.
>   notify the protocol when a RESET SESSION command is used.

18 years agoBack out RESET CONNECTION until there is more discussion.
Bruce Momjian [Tue, 25 Apr 2006 14:11:59 +0000 (14:11 +0000)]
Back out RESET CONNECTION until there is more discussion.

18 years agoAdd RESET CONNECTION, to reset all aspects of a session.
Bruce Momjian [Tue, 25 Apr 2006 14:09:21 +0000 (14:09 +0000)]
Add RESET CONNECTION, to reset all aspects of a session.

Hans-J?rgen Sch?nig

18 years agoUpdate inheritance constraint items:
Bruce Momjian [Tue, 25 Apr 2006 02:58:34 +0000 (02:58 +0000)]
Update inheritance constraint items:

<  o %Prevent child tables from altering or dropping constraints
<           like CHECK that were inherited from the parent table
<    like CHECK that are inherited by child tables
<
<    Dropping constraints should only be possible with CASCADE.
<
>    like CHECK that are inherited by child tables unless CASCADE
>    is used
>  o %Prevent child tables from altering or dropping constraints
>           like CHECK that were inherited from the parent table

18 years agoFix bug reported by Michael Fuhr on Sun, 23 Apr 2006 09:52:31 -0600
Tatsuo Ishii [Tue, 25 Apr 2006 00:40:54 +0000 (00:40 +0000)]
Fix bug reported by Michael Fuhr on Sun, 23 Apr 2006 09:52:31 -0600
(pgsql-committers@postgresql.org).

18 years agoDone:
Bruce Momjian [Tue, 25 Apr 2006 00:26:53 +0000 (00:26 +0000)]
Done:

> * -Add transaction_timestamp(), statement_timestamp(), clock_timestamp()

18 years agoAdd statement_timestamp(), clock_timestamp(), and
Bruce Momjian [Tue, 25 Apr 2006 00:25:22 +0000 (00:25 +0000)]
Add statement_timestamp(), clock_timestamp(), and
transaction_timestamp() (just like now()).

Also update statement_timeout() to mention it is statement arrival time
that is measured.

Catalog version updated.

18 years agoWord wrap entry
Bruce Momjian [Tue, 25 Apr 2006 00:06:23 +0000 (00:06 +0000)]
Word wrap entry

18 years agoUpdate SQL-standard INTERVAL item:
Bruce Momjian [Tue, 25 Apr 2006 00:01:44 +0000 (00:01 +0000)]
Update SQL-standard INTERVAL item:

        o Support ISO INTERVAL syntax if units cannot be determined from
          the string, and are supplied after the string

          The SQL standard states that the units after the string specify
          the units of the string, e.g. INTERVAL '2' MINUTE should
          return '00:02:00'.  The current behavior has the units
          restrict the interval value to the specified unit or unit range,
          INTERVAL '70' SECOND returns '00:00:10'.

          For syntax that isn't uniquely ISO or PG syntax, like '1' or
          '1:30', treat as ISO if there is a range specification clause,
          and as PG if there no clause is present, e.g. interpret
          '1:30' MINUTE TO SECOND as '1 minute 30 seconds', and
          interpret '1:30' as '1 hour, 30 minutes'.

          This makes common cases like SELECT INTERVAL '1' MONTH
          SQL-standard results.  The SQL standard supports a limited
          number of unit combinations and doesn't support unit names
          in the string.  The PostgreSQL syntax is more flexible in
          the range of units supported, e.g. PostgreSQL supports
          '1 year 1 hour', while the SQL standard does not.

18 years agoDone;
Bruce Momjian [Mon, 24 Apr 2006 22:59:46 +0000 (22:59 +0000)]
Done;

   o -Add support for day-time syntax, INTERVAL '1 2:03:04'DAY TO
      SECOND

18 years agoBack out patch, unintended.
Bruce Momjian [Mon, 24 Apr 2006 22:59:19 +0000 (22:59 +0000)]
Back out patch, unintended.

18 years agoDone:
Bruce Momjian [Mon, 24 Apr 2006 22:56:11 +0000 (22:56 +0000)]
Done:

     o -Add support for day-time syntax, INTERVAL '1 2:03:04' DAY TO
        SECOND

18 years agoSuppress more compiler warnings caused by macro tests.
Bruce Momjian [Mon, 24 Apr 2006 22:24:58 +0000 (22:24 +0000)]
Suppress more compiler warnings caused by macro tests.

18 years agoAdd one more paren to macro.
Bruce Momjian [Mon, 24 Apr 2006 22:17:04 +0000 (22:17 +0000)]
Add one more paren to macro.

18 years agoSuprress compiler warning in gcc 4.2.
Bruce Momjian [Mon, 24 Apr 2006 22:06:32 +0000 (22:06 +0000)]
Suprress compiler warning in gcc 4.2.

Report by Kris Jurka

18 years agoImprove our private implementation of cbrt() to give results of the
Tom Lane [Mon, 24 Apr 2006 20:36:32 +0000 (20:36 +0000)]
Improve our private implementation of cbrt() to give results of the
accuracy expected by the regression tests.  Per suggestion from
Martijn van Oosterhout.

18 years agoRemove compiler warning by casting SNPRINTF() call to void.
Bruce Momjian [Mon, 24 Apr 2006 19:51:13 +0000 (19:51 +0000)]
Remove compiler warning by casting SNPRINTF() call to void.

Report from Gevik Babakhani.

18 years agoFixed memory leak bugs found by Martijn Oosterhout.
Michael Meskes [Mon, 24 Apr 2006 09:45:22 +0000 (09:45 +0000)]
Fixed memory leak bugs found by Martijn Oosterhout.

18 years agoFixes for BCC 5.5 compile of libpq. Backpatch to 8.1.X.
Bruce Momjian [Mon, 24 Apr 2006 04:03:25 +0000 (04:03 +0000)]
Fixes for BCC 5.5 compile of libpq.  Backpatch to 8.1.X.

Mark Morgan Lloyd

18 years agoDon't add a shared dependency on the owner of a composite type in pg_class.
Alvaro Herrera [Mon, 24 Apr 2006 01:40:48 +0000 (01:40 +0000)]
Don't add a shared dependency on the owner of a composite type in pg_class.
We track the owner in pg_type instead, as that is the place where the owner is
changed on ALTER TYPE ... OWNER TO.

18 years agoAdd:
Bruce Momjian [Sun, 23 Apr 2006 04:00:06 +0000 (04:00 +0000)]
Add:

< * -Eventually enable escape_string_warning and standard_conforming_strings
> * -Enable escape_string_warning and standard_conforming_strings
> * Make standard_conforming_strings the default in 8.3?
>
>   When this is done, backslash-quote should be prohibited in non-E''
>   strings because of possible confusion over how such strings treat
>   backslashes.  Basically, '' is always safe for a literal single
>   quote, while \' might or might not be based on the backslash
>   handling rules.
>

18 years agoRemoves or minimizes some documentation mentions of backward
Bruce Momjian [Sun, 23 Apr 2006 03:39:52 +0000 (03:39 +0000)]
Removes or minimizes some documentation mentions of backward
compatibility for release 7.2 and earlier.  I have not altered any
mentions of release 7.3 or later.  The release notes were not modified,
so the changes are still documented, just not in the main docs.

18 years agoAdd pgrowlocks
Tatsuo Ishii [Sun, 23 Apr 2006 02:17:59 +0000 (02:17 +0000)]
Add pgrowlocks

18 years agoAdd new contrib function pgrowlocks. See README.pgrowlocks for more details.
Tatsuo Ishii [Sun, 23 Apr 2006 01:12:58 +0000 (01:12 +0000)]
Add new contrib function pgrowlocks. See README.pgrowlocks for more details.

18 years agoUpdate:
Bruce Momjian [Sat, 22 Apr 2006 18:18:24 +0000 (18:18 +0000)]
Update:

<  o Fix psql's backslash commands more consistent
>  o Fix psql's \d commands more consistent

18 years agoUpdate:
Bruce Momjian [Sat, 22 Apr 2006 18:17:57 +0000 (18:17 +0000)]
Update:

<  o Fix psql's \dn for various schema combinations (Neil)
>  o Fix psql's backslash commands more consistent
625a626
>    http://archives.postgresql.org/pgsql-hackers/2004-11/msg00014.php

18 years agoUpdate:
Bruce Momjian [Sat, 22 Apr 2006 16:06:44 +0000 (16:06 +0000)]
Update:

        o Fix psql's \dn for various schema combinations (Neil)

          http://archives.postgresql.org/pgsql-hackers/2004-11/msg00014.php

18 years agoSuggest Win32 users user E'' strings and double backslashes used for
Bruce Momjian [Sat, 22 Apr 2006 03:03:11 +0000 (03:03 +0000)]
Suggest Win32 users user E'' strings and double backslashes used for
patch separators in COPY.

Backpatch doubleing backslashes suggestion to 8.1.

18 years agoSimplify ParamListInfo data structure to support only numbered parameters,
Tom Lane [Sat, 22 Apr 2006 01:26:01 +0000 (01:26 +0000)]
Simplify ParamListInfo data structure to support only numbered parameters,
not named ones, and replace linear searches of the list with array indexing.
The named-parameter support has been dead code for many years anyway,
and recent profiling suggests that the searching was costing a noticeable
amount of performance for complex queries.

18 years agoRevert Intel compiler bug comment.
Bruce Momjian [Sat, 22 Apr 2006 00:29:41 +0000 (00:29 +0000)]
Revert Intel compiler bug comment.

18 years agoAdd mentioun of Intel compiler bug:
Bruce Momjian [Sat, 22 Apr 2006 00:27:30 +0000 (00:27 +0000)]
Add mentioun of Intel compiler bug:

Intel compiler has a bug in checking for division by NAN
# (NaN == 0), -mp1 fixes it, so someday we might way to add it.

18 years agoUse special gcc -W flags only if we are using gcc, not if we are using
Bruce Momjian [Fri, 21 Apr 2006 23:11:58 +0000 (23:11 +0000)]
Use special gcc -W flags only if we are using gcc, not if we are using
the Intel compiler.

Jeremy Drake

18 years agoRemove from TODO ability to edit pg_hba.conf, but add GRANT connection
Bruce Momjian [Fri, 21 Apr 2006 20:46:22 +0000 (20:46 +0000)]
Remove from TODO ability to edit pg_hba.conf, but add GRANT connection
permission item:

<  o %Allow pg_hba.conf settings to be controlled via SQL
>  o %Allow per-database permissions to be set via GRANT
<    This would add a function to load the SQL table from
<           pg_hba.conf, and one to writes its contents to the flat file.
<    The table should have a line number that is a float so rows
<    can be inserted between existing rows, e.g. row 2.5 goes
<    between row 2 and row 3.
>    Allow database connection checks based on GRANT rules in
>    addition to the existing access checks in pg_hba.conf.

18 years agoAdd some optional code (conditionally compiled under #ifdef LWLOCK_STATS)
Tom Lane [Fri, 21 Apr 2006 16:45:12 +0000 (16:45 +0000)]
Add some optional code (conditionally compiled under #ifdef LWLOCK_STATS)
to track the number of LWLock acquisitions and the number of times we
block waiting for an LWLock, on a per-process basis.  After having needed
this twice in the past few months, seems like it should go into CVS.

18 years agoEliminate some no-longer-needed workarounds for palloc's old behavior
Tom Lane [Thu, 20 Apr 2006 17:50:18 +0000 (17:50 +0000)]
Eliminate some no-longer-needed workarounds for palloc's old behavior
of rejecting palloc(0).  Also, tweak like_selectivity() to avoid assuming
the presented pattern is nonempty; although that assumption is valid,
it doesn't really help much, and the new coding is more correct anyway
since it properly handles redundant wildcards.  In combination these
changes should eliminate a Coverity warning noted by Martijn.

18 years agoAdd "retry another address" log message on statistics collector socket
Bruce Momjian [Thu, 20 Apr 2006 10:51:32 +0000 (10:51 +0000)]
Add "retry another address" log message on statistics collector socket
failure, to reduce confusion in the log file.

18 years agoEnsure that we validate the page header of the first page of a WAL file
Tom Lane [Thu, 20 Apr 2006 04:07:38 +0000 (04:07 +0000)]
Ensure that we validate the page header of the first page of a WAL file
whenever we start to read within that file.  The first page carries
extra identification information that really ought to be checked, but
as the code stood, this was only checked when we switched sequentially
into a new WAL file, or if by chance the starting checkpoint record was
within the first page.  This patch ensures that we will detect bogus
'long header' information before we start replaying the WAL sequence.

18 years agoRecognize __ppc64__, which seems to be Apple's spelling of the predefined
Tom Lane [Wed, 19 Apr 2006 23:11:15 +0000 (23:11 +0000)]
Recognize __ppc64__, which seems to be Apple's spelling of the predefined
symbol for PPC64 hardware.  I hadn't known that Apple supported PPC64 at
all, but darn if there aren't 64-bit variant libraries in OS X as well
as support in their gcc.

18 years agoAdd detail:
Bruce Momjian [Wed, 19 Apr 2006 22:35:42 +0000 (22:35 +0000)]
Add detail:

<    encoding.
>    encoding.  This requires using mblen() to determine if the
>    backslash is inside or outside a multi-byte sequence.

18 years agoAdd:
Bruce Momjian [Wed, 19 Apr 2006 21:50:19 +0000 (21:50 +0000)]
Add:

>
>  o Add new version of PQescapeString() that doesn't double backslashes
>    that are part of a client-only multibyte sequence
>
>    Single-quote is not a valid byte in any supported client-only
>    encoding.
>
>  o Add new version of PQescapeString() that doesn't double
>    backslashes when standard_conforming_strings is true and
>    non-E strings are used

18 years agoFix problem that sscanf(buf, "%d", &val) eats leading white space, but
Bruce Momjian [Wed, 19 Apr 2006 18:49:09 +0000 (18:49 +0000)]
Fix problem that sscanf(buf, "%d", &val) eats leading white space, but
our to_* functions were not handling that.

18 years agoRemove use of lorder and tsort while building static libraries. There's
Tom Lane [Wed, 19 Apr 2006 16:32:08 +0000 (16:32 +0000)]
Remove use of lorder and tsort while building static libraries.  There's
no evidence that any currently-supported platform needs this, and good
reason to think that any platform that did need it couldn't use the static
libraries anyway --- libpq, at least, has circular references.  Removing
the code shuts up tsort warnings about the circular references on some
platforms.

18 years agoFix ancient memory leak in PQprintTuples(); our code no longer uses this
Tom Lane [Wed, 19 Apr 2006 16:15:29 +0000 (16:15 +0000)]
Fix ancient memory leak in PQprintTuples(); our code no longer uses this
routine, but perhaps some applications do.  Found by Martijn van Oosterhout
using Coverity.

18 years agoFix a couple of rather-pointless-but-easily-fixed Coverity warnings.
Tom Lane [Wed, 19 Apr 2006 16:02:17 +0000 (16:02 +0000)]
Fix a couple of rather-pointless-but-easily-fixed Coverity warnings.
Per Martijn van Oosterhout.

18 years agoC code whitespace inprovement for formatting.c.
Bruce Momjian [Wed, 19 Apr 2006 14:48:06 +0000 (14:48 +0000)]
C code whitespace inprovement for formatting.c.

18 years agoAdd:
Bruce Momjian [Wed, 19 Apr 2006 02:46:32 +0000 (02:46 +0000)]
Add:

>    o Add reporting of the current WAL file, perhaps as part of
>      partial log file archiving

18 years agoRemove kill mention in install docs, was for pre-7.0 releases.
Bruce Momjian [Tue, 18 Apr 2006 22:52:07 +0000 (22:52 +0000)]
Remove kill mention in install docs, was for pre-7.0 releases.

18 years agoMention "syntax" error as not logged by log_statement.
Bruce Momjian [Tue, 18 Apr 2006 12:41:15 +0000 (12:41 +0000)]
Mention "syntax" error as not logged by log_statement.
Backpatch.

18 years agoDocument that errors are not output by log_statement (was they were in
Bruce Momjian [Tue, 18 Apr 2006 00:52:23 +0000 (00:52 +0000)]
Document that errors are not output by log_statement (was they were in
8.0), and add as suggestion to use log_min_error_statement for this
purpose.  I also fixed the code so the first EXECUTE has it's prepare,
rather than the last which is what was in the current code.  Also remove
"protocol" prefix for SQL EXECUTE output because it is not accurate.

Backpatch to 8.1.X.

18 years agoFix the torn-page hazard for PITR base backups by forcing full page writes
Tom Lane [Mon, 17 Apr 2006 18:55:05 +0000 (18:55 +0000)]
Fix the torn-page hazard for PITR base backups by forcing full page writes
to occur between pg_start_backup() and pg_stop_backup(), even if the GUC
setting full_page_writes is OFF.  Per discussion, doing this in combination
with the already-existing checkpoint during pg_start_backup() should ensure
safety against partial page updates being included in the backup.  We do
not have to force full page writes to occur during normal PITR operation,
as I had first feared.

18 years agoDone:
Bruce Momjian [Mon, 17 Apr 2006 03:18:07 +0000 (03:18 +0000)]
Done:

<  o Update pg_dump and psql to use the new COPY libpq API (Christopher)
>  o -Update pg_dump and psql to use the new COPY libpq API (Christopher)

18 years agoRemove:
Bruce Momjian [Sat, 15 Apr 2006 18:13:11 +0000 (18:13 +0000)]
Remove:

<  o %Have pg_dump use multi-statement transactions for INSERT dumps

18 years agoDocument that pg_dump -d/-D prevents invalid data from canceling the
Bruce Momjian [Sat, 15 Apr 2006 18:11:16 +0000 (18:11 +0000)]
Document that pg_dump -d/-D prevents invalid data from canceling the
entire table load.

18 years agoSupport the syntax
Tom Lane [Sat, 15 Apr 2006 17:45:46 +0000 (17:45 +0000)]
Support the syntax
CREATE AGGREGATE aggname (input_type) (parameter_list)
along with the old syntax where the input type was named in the parameter
list.  This fits more naturally with the way that the aggregate is identified
in DROP AGGREGATE and other utility commands; furthermore it has a natural
extension to handle multiple-input aggregates, where the basetype-parameter
method would get ugly.  In fact, this commit fixes the grammar and all the
utility commands to support multiple-input aggregates; but DefineAggregate
rejects it because the executor isn't fixed yet.
I didn't do anything about treating agg(*) as a zero-input aggregate instead
of artificially making it a one-input aggregate, but that should be considered
in combination with supporting multi-input aggregates.

18 years agoDone:
Bruce Momjian [Fri, 14 Apr 2006 22:26:22 +0000 (22:26 +0000)]
Done:

> * -Re-enable the GUC full_page_writes in 8.2 when reliability issues have

18 years agoMake the world safe for full_page_writes. Allow XLOG records that try to
Tom Lane [Fri, 14 Apr 2006 20:27:24 +0000 (20:27 +0000)]
Make the world safe for full_page_writes.  Allow XLOG records that try to
update no-longer-existing pages to fall through as no-ops, but make a note
of each page number referenced by such records.  If we don't see a later
XLOG entry dropping the table or truncating away the page, complain at
the end of XLOG replay.  Since this fixes the known failure mode for
full_page_writes = off, revert my previous band-aid patch that disabled
that GUC variable.

18 years agoRepair a low-probability race condition identified by Qingqing Zhou.
Tom Lane [Fri, 14 Apr 2006 03:38:56 +0000 (03:38 +0000)]
Repair a low-probability race condition identified by Qingqing Zhou.
If a process abandons a wait in LockBufferForCleanup (in practice,
only happens if someone cancels a VACUUM) just before someone else
sends it a signal indicating the buffer is available, it was possible
for the wakeup to remain in the process' semaphore, causing misbehavior
next time the process waited for an lmgr lock.  Rather than try to
prevent the race condition directly, it seems best to make the lock
manager robust against leftover wakeups, by having it repeat waiting
on the semaphore if the lock has not actually been granted or denied
yet.

18 years agoFix similar_escape() so that SIMILAR TO works properly for patterns involving
Tom Lane [Thu, 13 Apr 2006 18:01:31 +0000 (18:01 +0000)]
Fix similar_escape() so that SIMILAR TO works properly for patterns involving
alternatives ("|" symbol).  The original coding allowed the added ^ and $
constraints to be absorbed into the first and last alternatives, producing
a pattern that would match more than it should.  Per report from Eric Noriega.

I also changed the pattern to add an ARE director ("***:"), ensuring that
SIMILAR TO patterns do not change behavior if regex_flavor is changed.  This
is necessary to make the non-capturing parentheses work, and seems like a
good idea on general principles.

Back-patched as far as 7.4.  7.3 also has the bug, but a fix seems impractical
because that version's regex engine doesn't have non-capturing parens.

18 years agoFix HTML markup.
Bruce Momjian [Thu, 13 Apr 2006 12:20:36 +0000 (12:20 +0000)]
Fix HTML markup.

18 years agoUpdate tutorial URL's.
Bruce Momjian [Thu, 13 Apr 2006 12:14:41 +0000 (12:14 +0000)]
Update tutorial URL's.

18 years agoUpdate URL for SQL tutorial.
Bruce Momjian [Thu, 13 Apr 2006 12:09:08 +0000 (12:09 +0000)]
Update URL for SQL tutorial.

18 years agoUpdate AIX FAQ:
Bruce Momjian [Thu, 13 Apr 2006 11:41:02 +0000 (11:41 +0000)]
Update AIX FAQ:

At any rate, here's a revision to CVS HEAD to reflect some changes by
myself and by Seneca Cunningham for the AIX FAQ.  It touches on the
following issues:

1.  memcpy pointer patch for dynahash.c

2.  AIX memory management, which can, for 32 bit cases, bite people
    quite unexpectedly...

Chris Browne

18 years agoUpdate:
Bruce Momjian [Thu, 13 Apr 2006 10:50:13 +0000 (10:50 +0000)]
Update:

<   multiple I/O channels simultaneously.
>   multiple I/O channels simultaneously.  One idea is to create a
>   background reader that can pre-fetch sequential and index scan
>   pages needed by other backends.  This could be expanded to allow
>   concurrent reads from multiple devices in a partitioned table.

18 years agoFix an ancient oversight in btree xlog replay. When trying to determine if an
Tom Lane [Thu, 13 Apr 2006 03:53:05 +0000 (03:53 +0000)]
Fix an ancient oversight in btree xlog replay.  When trying to determine if an
upper-level insertion completes a previously-seen split, we cannot simply grab
the downlink block number out of the buffer, because the buffer could contain
a later state of the page --- or perhaps the page doesn't even exist at all
any more, due to relation truncation.  These possibilities have been masked up
to now because the use of full_page_writes effectively ensured that no xlog
replay routine ever actually saw a page state newer than its own change.
Since we're deprecating full_page_writes in 8.1.*, there's no need to fix this
in existing release branches, but we need a fix in HEAD if we want to have any
hope of re-allowing full_page_writes.  Accordingly, adjust the contents of
btree WAL records so that we can always get the downlink block number from the
WAL record rather than having to depend on buffer contents.  Per report from
Kevin Grittner and Peter Brant.

Improve a few comments in related code while at it.

18 years agoFix pg_restore -n option to do what the man page says it does. The
Tom Lane [Wed, 12 Apr 2006 22:18:48 +0000 (22:18 +0000)]
Fix pg_restore -n option to do what the man page says it does.  The
original coding only worked if one of the selTypes restriction options
was also given.  Per report from Nick Johnson.

18 years agoChange int->int32, for clarity.
Bruce Momjian [Wed, 12 Apr 2006 21:46:31 +0000 (21:46 +0000)]
Change int->int32, for clarity.

jw.pgsql@sduept.com

18 years agoAdd, "How do I submit a patch" FAQ item.
Bruce Momjian [Wed, 12 Apr 2006 20:03:54 +0000 (20:03 +0000)]
Add, "How do I submit a patch" FAQ item.

18 years agoAdd second sentence:
Bruce Momjian [Wed, 12 Apr 2006 18:56:16 +0000 (18:56 +0000)]
Add second sentence:
    <P>The maximum table size, row size, and maximum number of columns
    can be quadrupled by increasing the default block size to 32k.  The
    maximum table size can also be increased using table partitioning.</P>

18 years agoUpdated Japanese FAQ, version 2.
Bruce Momjian [Tue, 11 Apr 2006 21:21:52 +0000 (21:21 +0000)]
Updated Japanese FAQ, version 2.

J.Kuwamura

18 years agoUpdate Japanese FAQ, backpatch.
Bruce Momjian [Tue, 11 Apr 2006 21:18:03 +0000 (21:18 +0000)]
Update Japanese FAQ, backpatch.

J.Kuwamura

18 years agoOops, sysctl, not sysconf, for FreeBSD.
Bruce Momjian [Tue, 11 Apr 2006 21:04:52 +0000 (21:04 +0000)]
Oops, sysctl, not sysconf, for FreeBSD.

18 years agoFix typo.
Neil Conway [Tue, 11 Apr 2006 20:26:40 +0000 (20:26 +0000)]
Fix typo.

18 years agoMore FreeBSD jail wording improvements.
Bruce Momjian [Tue, 11 Apr 2006 19:45:33 +0000 (19:45 +0000)]
More FreeBSD jail wording improvements.

18 years agoUpdate jail wording to mention non-root users.
Bruce Momjian [Tue, 11 Apr 2006 19:42:27 +0000 (19:42 +0000)]
Update jail wording to mention non-root users.

18 years agoAdd documentation about running postmasters in FreeBSD jails (use
Bruce Momjian [Tue, 11 Apr 2006 19:26:42 +0000 (19:26 +0000)]
Add documentation about running postmasters in FreeBSD jails (use
separate users).

18 years agoThere is updated Russian translation of the FAQ in the attachment.
Bruce Momjian [Tue, 11 Apr 2006 15:48:11 +0000 (15:48 +0000)]
There is updated Russian translation of the FAQ in the attachment.

Victor Vislobokov

18 years agoDone, per Tom:
Bruce Momjian [Tue, 11 Apr 2006 02:08:19 +0000 (02:08 +0000)]
Done, per Tom:

< * Allow user-defined functions retuning a domain value to enforce domain
> * -Allow user-defined functions retuning a domain value to enforce domain