Tom Lane [Fri, 8 Feb 2013 15:41:15 +0000 (10:41 -0500)]
doc: Fix mistakes in the most recent set of release notes.
Improve description of the vacuum_freeze_table_age bug (it's much more
serious than we realized at the time the fix was committed), and correct
attribution of pg_upgrade -O/-o fix (Marti Raudsepp contributed that,
but Bruce forgot to credit him in the commit log).
No need to back-patch right now, it'll happen when the next set of
release notes are prepared.
Instead of hardcoding a specific link, give a general link to the
download section of the web site. This gives the user more download
options and the sysadmins more flexibility. Also, the previously
presented link didn't work for devel versions.
Tom Lane [Fri, 8 Feb 2013 03:53:00 +0000 (22:53 -0500)]
Fix performance issue in EXPLAIN (ANALYZE, TIMING OFF).
Commit af7914c6627bcf0b0ca614e9ce95d3f8056602bf, which added the TIMING
option to EXPLAIN, had an oversight: if the TIMING option is disabled
then control in InstrStartNode() goes through an elog(DEBUG2) call, which
typically does nothing but takes a noticeable amount of time to do it.
Tweak the logic to avoid that.
In HEAD, also change the elog(DEBUG2)'s in instrument.c to elog(ERROR).
It's not very clear why they weren't like that to begin with, but this
episode shows that not complaining more vociferously about misuse is
likely to do little except allow bugs to remain hidden.
While at it, adjust some code that was making possibly-dangerous
assumptions about flag bits being in the rightmost byte of the
instrument_options word.
Problem reported by Pavel Stehule (via Tomas Vondra).
Tom Lane [Fri, 8 Feb 2013 00:13:59 +0000 (19:13 -0500)]
Make contrib/btree_gist's GiST penalty function a bit saner.
The previous coding supposed that the first differing bytes in two varlena
datums must have the same sign difference as their overall comparison
result. This is obviously bogus for text strings in non-C locales, and
probably wrong for numeric, and even for bytea I think it was wrong on
machines where char is signed. When the assumption failed, the function
could deliver a zero or negative penalty in situations where such a result
is quite ridiculous, leading the core GiST code to make very bad page-split
decisions.
To fix, take the absolute values of the byte-level differences. Also,
switch the code to using unsigned char not just char, so that the behavior
will be consistent whether char is signed or not.
Per investigation of a trouble report from Tomas Vondra. Back-patch to all
supported branches.
Tom Lane [Thu, 7 Feb 2013 23:22:19 +0000 (18:22 -0500)]
Fix erroneous range-union logic for varlena types in contrib/btree_gist.
gbt_var_bin_union() failed to do the right thing when the existing range
needed to be widened at both ends rather than just one end. This could
result in an invalid index in which keys that are present would not be
found by searches, because the searches would not think they need to
descend to the relevant leaf pages. This error affected all the varlena
datatypes supported by btree_gist (text, bytea, bit, numeric).
Per investigation of a trouble report from Tomas Vondra. (There is also
an issue in gbt_var_penalty(), but that should only result in inefficiency
not wrong answers. I'm committing this separately so that we have a git
state in which it can be tested that bad penalty results don't produce
invalid indexes.) Back-patch to all supported branches.
Tom Lane [Thu, 7 Feb 2013 22:44:02 +0000 (17:44 -0500)]
Repair bugs in GiST page splitting code for multi-column indexes.
When considering a non-last column in a multi-column GiST index,
gistsplit.c tries to improve on the split chosen by the opclass-specific
pickSplit function by considering penalties for the next column. However,
there were two bugs in this code: it failed to recompute the union keys for
the leftmost index columns, even though these might well change after
reassigning tuples; and it included the old union keys in the recomputation
for the columns it did recompute, so that those keys couldn't get smaller
even if they should. The first problem could result in an invalid index
in which searches wouldn't find index entries that are in fact present;
the second would make the index less efficient to search.
Both of these errors were caused by misuse of gistMakeUnionItVec, whose
API was designed in a way that just begged such errors to be made. There
is no situation in which it's safe or useful to compute the union keys for
a subset of the index columns, and there is no caller that wants any
previous union keys to be included in the computation; so the undocumented
choice to treat the union keys as in/out rather than pure output parameters
is a waste of code as well as being dangerous.
Hence, rather than just making a minimal patch, I've changed the API of
gistMakeUnionItVec to remove the "startkey" parameter (it now always
processes all index columns) and treat the attr/isnull arrays as purely
output parameters.
In passing, also get rid of a couple of unnecessary and dangerous uses
of static variables in gistutil.c. It's remarkable that the one in
gistMakeUnionKey hasn't given us portability troubles before now, because
in addition to posing a re-entrancy hazard, it was unsafely assuming that
a static char[] array would have at least Datum alignment.
Per investigation of a trouble report from Tomas Vondra. (There are also
some bugs in contrib/btree_gist to be fixed, but that seems like material
for a separate patch.) Back-patch to all supported branches.
Tom Lane [Thu, 7 Feb 2013 19:44:00 +0000 (14:44 -0500)]
Fix possible failure to send final transaction counts to stats collector.
Normally, we suppress sending a tabstats message to the collector unless
there were some actual table stats to send. However, during backend exit
we should force out the message if there are any transaction commit/abort
counts to send, else the session's last few commit/abort counts will never
get reported at all. We had logic for this, but the short-circuit test
at the top of pgstat_report_stat() ignored the "force" flag, with the
consequence that session-ending transactions that touched no database-local
tables would not get counted. Seems to be an oversight in my commit 641912b4d17fd214a5e5bae4e7bb9ddbc28b144b, which added the "force" flag.
That was back in 8.3, so back-patch to all supported versions.
Tom Lane [Mon, 4 Feb 2013 21:25:01 +0000 (16:25 -0500)]
Prevent execution of enum_recv() from SQL.
This function was misdeclared to take cstring when it should take internal.
This at least allows crashing the server, and in principle an attacker
might be able to use the function to examine the contents of server memory.
The correct fix is to adjust the system catalog contents (and fix the
regression tests that should have caught this but failed to). However,
asking users to correct the catalog contents in existing installations
is a pain, so as a band-aid fix for the back branches, install a check
in enum_recv() to make it throw error if called with a cstring argument.
We will later revert this in HEAD in favor of correcting the catalogs.
Our thanks to Sumit Soni (via Secunia SVCRP) for reporting this issue.
Simon Riggs [Mon, 4 Feb 2013 10:29:22 +0000 (10:29 +0000)]
Reset master xmin when hot_standby_feedback disabled.
If walsender has xmin of standby then ensure we
reset the value to 0 when we change from hot_standby_feedback=on
to hot_standby_feedback=off.
Tom Lane [Sun, 3 Feb 2013 20:56:45 +0000 (15:56 -0500)]
Perform line wrapping and indenting by default in ruleutils.c.
This patch changes pg_get_viewdef() and allied functions so that
PRETTY_INDENT processing is always enabled. Per discussion, only the
PRETTY_PAREN processing (that is, stripping of "unnecessary" parentheses)
poses any real forward-compatibility risk, so we may as well make dump
output look as nice as we safely can.
Also, set the default wrap length to zero (i.e, wrap after each SELECT
or FROM list item), since there's no very principled argument for the
former default of 80-column wrapping, and most people seem to agree this
way looks better.
Marko Tiikkaja, reviewed by Jeevan Chalke, further hacking by Tom Lane
Tom Lane [Sat, 2 Feb 2013 19:21:24 +0000 (14:21 -0500)]
Prevent "\g filename" from affecting subsequent commands after an error.
In the previous coding, psql's state variable saying that output should
go to a file was only reset after successful completion of a query
returning tuples. Thus for example,
regression=# select 1/0
regression-# \g somefile
ERROR: division by zero
regression=# select 1/2;
regression=#
... huh, I wonder where that output went. Even more oddly, the state
was not reset even if it's the file that's causing the failure:
This seems to me not to satisfy the principle of least surprise.
\g is certainly not documented in a way that suggests its effects are
at all persistent.
To fix, adjust the code so that the flag is reset at exit from SendQuery
no matter what happened.
Noted while reviewing the \gset patch, which had comparable issues.
Arguably this is a bug fix, but I'll refrain from back-patching for now.
Alvaro Herrera [Fri, 1 Feb 2013 20:50:04 +0000 (17:50 -0300)]
Move Assert() definitions to c.h
This way, they can be used by frontend and backend code. We already
supported that, but doing it this way allows us to mix true frontend
files with backend files compiled in frontend environment.
Alvaro Herrera [Fri, 1 Feb 2013 15:00:40 +0000 (12:00 -0300)]
Fix typo in freeze_table_age implementation
The original code used freeze_min_age instead of freeze_table_age. The
main consequence of this mistake is that lowering freeze_min_age would
cause full-table scans to occur much more frequently, which causes
serious issues because the number of writes required is much larger.
That feature (freeze_min_age) is supposed to affect only how soon tuples
are frozen; some pages should still be skipped due to the visibility
map.
Backpatch to 8.4, where the freeze_table_age feature was introduced.
Alvaro Herrera [Fri, 1 Feb 2013 13:43:09 +0000 (10:43 -0300)]
Fill tuple before HeapSatisfiesHOTAndKeyUpdate
Failing to do this results in almost all updates to system catalogs
being non-HOT updates, because the OID column would differ (not having
been set for the new tuple), which is an indexed column.
While at it, make sure to set the tableoid early in both old and new
tuples as well. This isn't of much consequence, since that column is
seldom (never?) indexed.
This is specified in the SQL standard. The CREATE RECURSIVE VIEW
specification is transformed into a normal CREATE VIEW statement with a
WITH RECURSIVE clause.
Alvaro Herrera [Thu, 31 Jan 2013 22:12:35 +0000 (19:12 -0300)]
Restrict infomask bits to set on multixacts
We must only set the bit(s) for the strongest lock held in the tuple;
otherwise, a multixact containing members with exclusive lock and
key-share lock will behave as though only a share lock is held.
This bug was introduced in commit 0ac5ad5134, somewhere along
development, when we allowed a singleton FOR SHARE lock to be
implemented without a MultiXact by using a multi-bit pattern.
I overlooked that GetMultiXactIdHintBits() needed to be tweaked as well.
Previously, we could have the bits for FOR KEY SHARE and FOR UPDATE
simultaneously set and it wouldn't cause a problem.
Simon Riggs [Thu, 31 Jan 2013 19:29:32 +0000 (19:29 +0000)]
Switch timelines if we crash soon after promotion.
Previous patch to skip checkpoints at end of recovery didn't
correctly perform crash recovery, fumbling the timeline switch.
Now we record the minRecoveryPointTLI of the newly selected
timeline, so that we crash recover to the correct timeline.
Tom Lane [Thu, 31 Jan 2013 17:12:23 +0000 (12:12 -0500)]
Reject nonzero day fields in AT TIME ZONE INTERVAL functions.
It's not sensible for an interval that's used as a time zone value to be
larger than a day. When we changed the interval type to contain a separate
day field, check_timezone() was adjusted to reject nonzero day values, but
timetz_izone(), timestamp_izone(), and timestamptz_izone() evidently were
overlooked.
While at it, make the error messages for these three cases consistent.
Magnus Hagander [Thu, 31 Jan 2013 14:03:24 +0000 (15:03 +0100)]
Properly zero-pad the day-of-year part of the win32 build number
This ensure the version number increases over time. The first three digits
in the version number is still set to the actual PostgreSQL version
number, but the last one is intended to be an ever increasing build number,
which previosly failed when it changed between 1, 2 and 3 digits long values.
Tatsuo Ishii [Thu, 31 Jan 2013 06:53:58 +0000 (15:53 +0900)]
Add --aggregate-interval option.
The new option specifies length of aggregation interval (in
seconds). May be used only together with -l. With this option, the log
contains per-interval summary (number of transactions, min/max latency
and two additional fields useful for variance estimation).
Patch contributed by Tomas Vondra, reviewed by Pavel Stehule. Slight
change by Tatsuo Ishii, suggested by Robert Hass to emit an error
message indicating that the option is not currently supported on
Windows.
Tom Lane [Thu, 31 Jan 2013 01:02:23 +0000 (20:02 -0500)]
Fix plpgsql's reporting of plan-time errors in possibly-simple expressions.
exec_simple_check_plan and exec_eval_simple_expr attempted to call
GetCachedPlan directly. This meant that if an error was thrown during
planning, the resulting context traceback would not include the line
normally contributed by _SPI_error_callback. This is already inconsistent,
but just to be really odd, a re-execution of the very same expression
*would* show the additional context line, because we'd already have cached
the plan and marked the expression as non-simple.
The problem is easy to demonstrate in 9.2 and HEAD because planning of a
cached plan doesn't occur at all until GetCachedPlan is done. In earlier
versions, it could only be an issue if initial planning had succeeded, then
a replan was forced (already somewhat improbable for a simple expression),
and the replan attempt failed. Since the issue is mainly cosmetic in older
branches anyway, it doesn't seem worth the risk of trying to fix it there.
It is worth fixing in 9.2 since the instability of the context printout can
affect the results of GET STACKED DIAGNOSTICS, as per a recent discussion
on pgsql-novice.
To fix, introduce a SPI function that wraps GetCachedPlan while installing
the correct callback function. Use this instead of calling GetCachedPlan
directly from plpgsql.
Also introduce a wrapper function for extracting a SPI plan's
CachedPlanSource list. This lets us stop including spi_priv.h in
pl_exec.c, which was never a very good idea from a modularity standpoint.
In passing, fix a similar inconsistency that could occur in SPI_cursor_open,
which was also calling GetCachedPlan without setting up a context callback.
Tom Lane [Wed, 30 Jan 2013 19:16:16 +0000 (14:16 -0500)]
Fix grammar for subscripting or field selection from a sub-SELECT result.
Such cases should work, but the grammar failed to accept them because of
our ancient precedence hacks to convince bison that extra parentheses
around a sub-SELECT in an expression are unambiguous. (Formally, they
*are* ambiguous, but we don't especially care whether they're treated as
part of the sub-SELECT or part of the expression. Bison cares, though.)
Fix by adding a redundant-looking production for this case.
This is a fine example of why fixing shift/reduce conflicts via
precedence declarations is more dangerous than it looks: you can easily
cause the parser to reject cases that should work.
This has been wrong since commit 3db4056e22b0c6b2adc92543baf8408d2894fe91
or maybe before, and apparently some people have been working around it
by inserting no-op casts. That method introduces a dump/reload hazard,
as illustrated in bug #7838 from Jan Mate. Hence, back-patch to all
active branches.
Tom Lane [Tue, 29 Jan 2013 22:06:26 +0000 (17:06 -0500)]
Provide database object names as separate fields in error messages.
This patch addresses the problem that applications currently have to
extract object names from possibly-localized textual error messages,
if they want to know for example which index caused a UNIQUE_VIOLATION
failure. It adds new error message fields to the wire protocol, which
can carry the name of a table, table column, data type, or constraint
associated with the error. (Since the protocol spec has always instructed
clients to ignore unrecognized field types, this should not create any
compatibility problem.)
Support for providing these new fields has been added to just a limited set
of error reports (mainly, those in the "integrity constraint violation"
SQLSTATE class), but we will doubtless add them to more calls in future.
Pavel Stehule, reviewed and extensively revised by Peter Geoghegan, with
additional hacking by Tom Lane.
Beyond 21474, the number of accounts exceed the range for int4. Change the
initialization code to use bigint for account id columns when scale is large
enough, and switch to using int64s for the variables in pgbench code. The
threshold where we switch to bigints is set at 20000, because that's easier
to remember and document than 21474, and ensures that there is some headroom
when int4s are used.
Greg Smith, with various changes by Euler Taveira de Oliveira, Gurjeet
Singh and Satoshi Nagayasu.
Skip truncating ON COMMIT DELETE ROWS temp tables, if the transaction hasn't
touched any temporary tables.
We could try harder, and keep track of whether we've inserted to any temp
tables, rather than accessed them, and which temp tables have been inserted
to. But this is dead simple, and already covers many interesting scenarios.
Simon Riggs [Tue, 29 Jan 2013 00:06:15 +0000 (00:06 +0000)]
Fast promote mode skips checkpoint at end of recovery.
pg_ctl promote -m fast will skip the checkpoint at end of recovery so that we
can achieve very fast failover when the apply delay is low. Write new WAL record
XLOG_END_OF_RECOVERY to allow us to switch timeline correctly for downstream log
readers. If we skip synchronous end of recovery checkpoint we request a normal
spread checkpoint so that the window of re-recovery is low.
Simon Riggs and Kyotaro Horiguchi, with input from Fujii Masao.
Review by Heikki Linnakangas
Alvaro Herrera [Mon, 28 Jan 2013 21:03:23 +0000 (18:03 -0300)]
REASSIGN OWNED: handle shared objects, too
Give away ownership of shared objects (databases, tablespaces) along
with local objects, per original code intention. Try to make the
documentation clearer, too.
Per discussion about DROP OWNED's brokenness, in bug #7748.
This is not backpatched because it'd require some refactoring of the
ALTER/SET OWNER code for databases and tablespaces.
Alvaro Herrera [Mon, 28 Jan 2013 20:46:47 +0000 (17:46 -0300)]
DROP OWNED: don't try to drop tablespaces/databases
My "fix" for bugs #7578 and #6116 on DROP OWNED at fe3b5eb08a1 not only
misstated that it applied to REASSIGN OWNED (which it did not affect),
but it also failed to fix the problems fully, because I didn't test the
case of owned shared objects. Thus I created a new bug, reported by
Thomas Kellerer as #7748, which would cause DROP OWNED to fail with a
not-for-user-consumption error message. The code would attempt to drop
the database, which not only fails to work because the underlying code
does not support that, but is a pretty dangerous and undesirable thing
to be doing as well.
This patch fixes that bug by having DROP OWNED only attempt to process
shared objects when grants on them are found, ignoring ownership.
Backpatch to 8.3, which is as far as the previous bug was backpatched.
Handle SPIErrors raised directly in PL/Python code.
If a PL/Python function raises an SPIError (or one if its subclasses)
directly with python's raise statement, treat it the same as an SPIError
generated internally. In particular, if the user sets the sqlstate
attribute, preserve that.
Oskari Saarenmaa and Jan Urbański, reviewed by Karl O. Pinc.
Tom Lane [Sat, 26 Jan 2013 21:18:42 +0000 (16:18 -0500)]
Make LATERAL implicit for functions in FROM.
The SQL standard does not have general functions-in-FROM, but it does
allow UNNEST() there (see the <collection derived table> production),
and the semantics of that are defined to include lateral references.
So spec compliance requires allowing lateral references within UNNEST()
even without an explicit LATERAL keyword. Rather than making UNNEST()
a special case, it seems best to extend this flexibility to any
function-in-FROM. We'll still allow LATERAL to be written explicitly
for clarity's sake, but it's now a noise word in this context.
In theory this change could result in a change in behavior of existing
queries, by allowing what had been an outer reference in a function-in-FROM
to be captured by an earlier FROM-item at the same level. However, all
pre-9.3 PG releases have a bug that causes them to match variable
references to earlier FROM-items in preference to outer references (and
then throw an error). So no previously-working query could contain the
type of ambiguity that would risk a change of behavior.
Per a suggestion from Andrew Gierth, though I didn't use his patch.
Bruce Momjian [Sat, 26 Jan 2013 19:51:59 +0000 (14:51 -0500)]
Update comments in new DROP IF EXISTS code; commit message update
DROP IF EXISTS with a missing schema in commit 7e2322dff30c04d90c0602d2b5ae24b4881db88b applies not only to tables, but
to DROP IF EXISTS with missing schemas for indexes, views, sequences,
and foreign tables. Yeah!
Bruce Momjian [Sat, 26 Jan 2013 18:33:24 +0000 (13:33 -0500)]
Issue ERROR if FREEZE mode can't be honored by COPY
Previously non-honored FREEZE mode was ignored. This also issues an
appropriate error message based on the cause of the failure, per
suggestion from Tom. Additional regression test case added.
Bruce Momjian [Sat, 26 Jan 2013 18:24:50 +0000 (13:24 -0500)]
Allow CREATE TABLE IF EXIST so succeed if the schema is nonexistent
Previously, CREATE TABLE IF EXIST threw an error if the schema was
nonexistent. This was done by passing 'missing_ok' to the function that
looks up the schema oid.
Tom Lane [Fri, 25 Jan 2013 21:58:55 +0000 (16:58 -0500)]
Fix plpython's handling of functions used as triggers on multiple tables.
plpython tried to use a single cache entry for a trigger function, but it
needs a separate cache entry for each table the trigger is applied to,
because there is table-dependent data in there. This was done correctly
before 9.1, but commit 46211da1b84bc3537e799ee1126098e71c2428e8 broke it
by simplifying the lookup key from "function OID and triggered table OID"
to "function OID and is-trigger boolean". Go back to using both OIDs
as the lookup key. Per bug report from Sandro Santilli.
Tom Lane [Fri, 25 Jan 2013 19:14:41 +0000 (14:14 -0500)]
Change plan caching to honor, not resist, changes in search_path.
In the initial implementation of plan caching, we saved the active
search_path when a plan was first cached, then reinstalled that path
anytime we needed to reparse or replan. The idea of that was to try to
reselect the same referenced objects, in somewhat the same way that views
continue to refer to the same objects in the face of schema or name
changes. Of course, that analogy doesn't bear close inspection, since
holding the search_path fixed doesn't cope with object drops or renames.
Moreover sticking with the old path seems to create more surprises than
it avoids. So instead of doing that, consider that the cached plan depends
on search_path, and force reparse/replan if the active search_path is
different than it was when we last saved the plan.
This gets us fairly close to having "transparency" of plan caching, in the
sense that the cached statement acts the same as if you'd just resubmitted
the original query text for another execution. There are still some corner
cases where this fails though: a new object added in the search path
schema(s) might capture a reference in the query text, but we'd not realize
that and force a reparse. We might try to fix that in the future, but for
the moment it looks too expensive and complicated.
Add some randomness to the choice of which GiST page to insert to.
When descending the tree for an insert, and there are multiple equally good
pages we could insert to, make the choice in random. Previously, we would
always choose the tuple with lowest offset number. That meant that when two
non-leaf pages overlap - in the extreme case they might have exactly the same
key - all but the first such page went unused. That wasn't optimal for space
usage; if you deleted some tuples from the non-first pages, the space would
never be reused.
With this patch, the other pages are sometimes chosen too, although there's
still a heavy bias towards low-offset tuples, so that we don't lose cache
locality when doing a lot of inserts with similar keys.
Original idea by Alexander Korotkov, although this patch version was written
by me and copy-edited by Tom Lane.
Tom Lane [Fri, 25 Jan 2013 05:19:18 +0000 (00:19 -0500)]
Fix concat() and format() to handle VARIADIC-labeled arguments correctly.
Previously, the VARIADIC labeling was effectively ignored, but now these
functions act as though the array elements had all been given as separate
arguments.
Tom Lane [Thu, 24 Jan 2013 23:34:00 +0000 (18:34 -0500)]
Fix SPI documentation for new handling of ExecutorRun's count parameter.
Since 9.0, the count parameter has only limited the number of tuples
actually returned by the executor. It doesn't affect the behavior of
INSERT/UPDATE/DELETE unless RETURNING is specified, because without
RETURNING, the ModifyTable plan node doesn't return control to execMain.c
for each tuple. And we only check the limit at the top level.
While this behavioral change was unintentional at the time, discussion of
bug #6572 led us to the conclusion that we prefer the new behavior anyway,
and so we should just adjust the docs to match rather than change the code.
Accordingly, do that. Back-patch as far as 9.0 so that the docs match the
code in each branch.
Alvaro Herrera [Thu, 24 Jan 2013 19:04:48 +0000 (16:04 -0300)]
Redefine HEAP_XMAX_IS_LOCKED_ONLY
Tuples marked SELECT FOR UPDATE in a cluster that's later processed by
pg_upgrade would have a different infomask bit pattern than those
produced by 9.3dev; that bit pattern was being seen as "dead" by HEAD
(because they would fail the "is this tuple locked" test, and so the
visibility rules would thing they're updated, even though there's no
HEAP_UPDATED version of them). In other words, some rows could silently
disappear after pg_upgrade.
With this new definition, those tuples become visible again.
This is breakage resulting from my commit 0ac5ad5134.
Alvaro Herrera [Thu, 24 Jan 2013 15:25:48 +0000 (12:25 -0300)]
Use the catversion to distinguish old/new clusters
This makes 9.3 -> 9.3 upgrades work when they cross the commit that
added persistent multixacts; early 9.3 pg_controldata did not have the
required oldestMultiXact line, and so would fail to upgrade.
Simon Riggs [Thu, 24 Jan 2013 14:19:29 +0000 (14:19 +0000)]
Fix rare missing cancellations in Hot Standby.
The machinery around XLOG_HEAP2_CLEANUP_INFO failed
to correctly pass through the necessary information
on latestRemovedXid, avoiding cancellations in some
infrequent concurrent update/cleanup scenarios.
Backpatchable fix to 9.0
Detailed bug report and fix by Noah Misch,
backpatchable version by me.
Tom Lane [Thu, 24 Jan 2013 03:08:01 +0000 (22:08 -0500)]
Fix failure to rotate postmaster log file for size reasons on Windows.
When we eliminated "unnecessary" wakeups of the syslogger process, we
broke size-based logfile rotation on Windows, because on that platform
data transfer is done in a separate thread. While non-Windows platforms
would recheck the output file size after every log message, Windows only
did so when the control thread woke up for some other reason, which might
be quite infrequent. Per bug #7814 from Tsunezumi. Back-patch to 9.2
where the problem was introduced.
Alvaro Herrera [Wed, 23 Jan 2013 15:04:59 +0000 (12:04 -0300)]
Improve concurrency of foreign key locking
This patch introduces two additional lock modes for tuples: "SELECT FOR
KEY SHARE" and "SELECT FOR NO KEY UPDATE". These don't block each
other, in contrast with already existing "SELECT FOR SHARE" and "SELECT
FOR UPDATE". UPDATE commands that do not modify the values stored in
the columns that are part of the key of the tuple now grab a SELECT FOR
NO KEY UPDATE lock on the tuple, allowing them to proceed concurrently
with tuple locks of the FOR KEY SHARE variety.
Foreign key triggers now use FOR KEY SHARE instead of FOR SHARE; this
means the concurrency improvement applies to them, which is the whole
point of this patch.
The added tuple lock semantics require some rejiggering of the multixact
module, so that the locking level that each transaction is holding can
be stored alongside its Xid. Also, multixacts now need to persist
across server restarts and crashes, because they can now represent not
only tuple locks, but also tuple updates. This means we need more
careful tracking of lifetime of pg_multixact SLRU files; since they now
persist longer, we require more infrastructure to figure out when they
can be removed. pg_upgrade also needs to be careful to copy
pg_multixact files over from the old server to the new, or at least part
of multixact.c state, depending on the versions of the old and new
servers.
Tuple time qualification rules (HeapTupleSatisfies routines) need to be
careful not to consider tuples with the "is multi" infomask bit set as
being only locked; they might need to look up MultiXact values (i.e.
possibly do pg_multixact I/O) to find out the Xid that updated a tuple,
whereas they previously were assured to only use information readily
available from the tuple header. This is considered acceptable, because
the extra I/O would involve cases that would previously cause some
commands to block waiting for concurrent transactions to finish.
Another important change is the fact that locking tuples that have
previously been updated causes the future versions to be marked as
locked, too; this is essential for correctness of foreign key checks.
This causes additional WAL-logging, also (there was previously a single
WAL record for a locked tuple; now there are as many as updated copies
of the tuple there exist.)
With all this in place, contention related to tuples being checked by
foreign key rules should be much reduced.
As a bonus, the old behavior that a subtransaction grabbing a stronger
tuple lock than the parent (sub)transaction held on a given tuple and
later aborting caused the weaker lock to be lost, has been fixed.
Many new spec files were added for isolation tester framework, to ensure
overall behavior is sane. There's probably room for several more tests.
There were several reviewers of this patch; in particular, Noah Misch
and Andres Freund spent considerable time in it. Original idea for the
patch came from Simon Riggs, after a problem report by Joel Jacobson.
Most code is from me, with contributions from Marti Raudsepp, Alexander
Shulgin, Noah Misch and Andres Freund.
This patch was discussed in several pgsql-hackers threads; the most
important start at the following message-ids:
AANLkTimo9XVcEzfiBR-ut3KVNDkjm2Vxh+t8kAmWjPuv@mail.gmail.com 1290721684-sup-3951@alvh.no-ip.org 1294953201-sup-2099@alvh.no-ip.org 1320343602-sup-2290@alvh.no-ip.org 1339690386-sup-8927@alvh.no-ip.org 4FE5FF020200002500048A3D@gw.wicourts.gov 4FEAB90A0200002500048B7D@gw.wicourts.gov
Fix more issues with cascading replication and timeline switches.
When a standby server follows the master using WAL archive, and it chooses
a new timeline (recovery_target_timeline='latest'), it only fetches the
timeline history file for the chosen target timeline, not any other history
files that might be missing from pg_xlog. For example, if the current
timeline is 2, and we choose 4 as the new recovery target timeline, the
history file for timeline 3 is not fetched, even if it's part of this
server's history. That's enough for the standby itself - the history file
for timeline 4 includes timeline 3 as well - but if a cascading standby
server wants to recover to timeline 3, it needs the history file. To fix,
when a new recovery target timeline is chosen, try to copy any missing
history files from the archive to pg_xlog between the old and new target
timeline.
A second similar issue was with the WAL files. When a standby recovers from
archive, and it reaches a segment that contains a switch to a new timeline,
recovery fetches only the WAL file labelled with the new timeline's ID. The
file from the new timeline contains a copy of the WAL from the old timeline
up to the point where the switch happened, and recovery recovers it from the
new file. But in streaming replication, walsender only tries to read it
from the old timeline's file. To fix, change walsender to read it from the
new file, so that it behaves the same as recovery in that sense, and doesn't
try to open the possibly nonexistent file with the old timeline's ID.
Bruce Momjian [Wed, 23 Jan 2013 03:27:13 +0000 (22:27 -0500)]
pg_upgrade: remove --single-transaction usage
With AtEOXact applied, --single-transaction makes pg_restore slower, and
has the potential to require lock table configuration, so remove the
argument.
projects / postgresql / log