Fujii Masao [Wed, 4 Jun 2014 03:09:45 +0000 (12:09 +0900)]
Save pg_stat_statements statistics file into $PGDATA/pg_stat directory at shutdown.
187492b6c2e8cafc5b39063ca3b67846e8155d24 changed pgstat.c so that
the stats files were saved into $PGDATA/pg_stat directory when the server
was shutdowned. But it accidentally forgot to change the location of
pg_stat_statements permanent stats file. This commit fixes pg_stat_statements
so that its stats file is also saved into $PGDATA/pg_stat at shutdown.
Since this fix changes the file layout, we don't back-patch it to 9.3
where this oversight was introduced.
Andrew Dunstan [Tue, 3 Jun 2014 22:26:47 +0000 (18:26 -0400)]
Use EncodeDateTime instead of to_char to render JSON timestamps.
Per gripe from Peter Eisentraut and Tom Lane.
The output is slightly different, but still ISO 8601 compliant: to_char
doesn't output the minutes when time zone offset is an integer number of
hours, while EncodeDateTime outputs ":00".
Andrew Dunstan [Tue, 3 Jun 2014 20:11:31 +0000 (16:11 -0400)]
Do not escape a unicode sequence when escaping JSON text.
Previously, any backslash in text being escaped for JSON was doubled so
that the result was still valid JSON. However, this led to some perverse
results in the case of Unicode sequences, These are now detected and the
initial backslash is no longer escaped. All other backslashes are
still escaped. No validity check is performed, all that is looked for is
\uXXXX where X is a hexidecimal digit.
This is a change from the 9.2 and 9.3 behaviour as noted in the Release
notes.
Andrew Dunstan [Tue, 3 Jun 2014 17:56:53 +0000 (13:56 -0400)]
Output timestamps in ISO 8601 format when rendering JSON.
Many JSON processors require timestamp strings in ISO 8601 format in
order to convert the strings. When converting a timestamp, with or
without timezone, to a JSON datum we therefore now use such a format
rather than the type's default text output, in functions such as
to_json().
This is a change in behaviour from 9.2 and 9.3, as noted in the release
notes.
Tom Lane [Tue, 3 Jun 2014 16:01:27 +0000 (12:01 -0400)]
Make plpython_unicode regression test work in more database encodings.
This test previously used a data value containing U+0080, and would
therefore fail if the database encoding didn't have an equivalent to
that; which only about half of our supported server encodings do.
We could fall back to using some plain-ASCII character, but that seems
like it's losing most of the point of the test. Instead switch to using
U+00A0 (no-break space), which translates into all our supported encodings
except the four in the EUC_xx family.
Per buildfarm testing. Back-patch to 9.1, which is as far back as this
test is expected to succeed everywhere. (9.0 has the test, but without
back-patching some 9.1 code changes we could not expect to get consistent
results across platforms anyway.)
Andres Freund [Tue, 3 Jun 2014 12:02:54 +0000 (14:02 +0200)]
Set the process latch when processing recovery conflict interrupts.
Because RecoveryConflictInterrupt() didn't set the process latch
anything using the latter to wait for events didn't get notified about
recovery conflicts. Most latch users are never the target of recovery
conflicts, which explains the lack of reports about this until
now.
Since 9.3 two possible affected users exist though: The sql callable
pg_sleep() now uses latches to wait and background workers are
expected to use latches in their main loop. Both would currently wait
until the end of WaitLatch's timeout.
Fix by adding a SetLatch() to RecoveryConflictInterrupt(). It'd also
be possible to fix the issue by having each latch user set
set_latch_on_sigusr1. That seems failure prone and though, as most of
these callsites won't often receive recovery conflicts and thus will
likely only be tested against normal query cancels et al. It'd also be
unnecessarily verbose.
Backpatch to 9.1 where latches were introduced. Arguably 9.3 would be
sufficient, because that's where pg_sleep() was converted to waiting
on the latch and background workers got introduced; but there could be
user level code making use of the latch pre 9.3.
Andres Freund [Tue, 3 Jun 2014 10:19:18 +0000 (12:19 +0200)]
Use unaligned output in another regression test query to reduce diff noise.
Use the unaligned/no rowcount output mode in a regression tests that
shows all built-in leakproof functions. Currently a new leakproof
function will often change the alignment of all existing functions,
making it hard to see the actual difference and creating unnecessary
patch conflicts.
Noticed while looking over a patch introducing new leakproof functions.
Andrew Dunstan [Sun, 1 Jun 2014 23:04:02 +0000 (19:04 -0400)]
Improve the efficiency of certain jsonb get operations.
Instead of iterating over jsonb structures, use the inbuilt functions
findJsonbValueFromContainerLen() and getIthJsonbValueFromContainer() to
extract values directly. These functions use algorithms that are O(n log
n) and O(1) respectively, whereas iterating is O(n), so we should see
considerable speedup here.
Andres Freund [Sat, 31 May 2014 13:58:04 +0000 (15:58 +0200)]
Improvements to the replication protocol documentation.
Document the CREATE_REPLICATION_SLOT's output_plugin parameter; that
START_REPLICATION ... LOGICAL takes parameters; that START_REPLICATION
... LOGICAL uses the same messages as ... PHYSICAL; and be more
consistent with the usage of <literal/>.
Michael Paquier, with some additional changes by me.
Tom Lane [Fri, 30 May 2014 22:18:11 +0000 (18:18 -0400)]
On OS X, link libpython normally, ignoring the "framework" framework.
As of Xcode 5.0, Apple isn't including the Python framework as part of the
SDK-level files, which means that linking to it might fail depending on
whether Xcode thinks you've selected a specific SDK version. According to
their Tech Note 2328, they've basically deprecated the framework method of
linking to libpython and are telling people to link to the shared library
normally. (I'm pretty sure this is in direct contradiction to the advice
they were giving a few years ago, but whatever.) Testing says that this
approach works fine at least as far back as OS X 10.4.11, so let's just
rip out the framework special case entirely. We do still need a special
case to decide that OS X provides a shared library at all, unfortunately
(I wonder why the distutils check doesn't work ...). But this is still
less of a special case than before, so it's fine.
Back-patch to all supported branches, since we'll doubtless be hearing
about this more as more people update to recent Xcode.
Tom Lane [Thu, 29 May 2014 17:51:02 +0000 (13:51 -0400)]
When using the OSSP UUID library, cache its uuid_t state object.
The original coding in contrib/uuid-ossp created and destroyed a uuid_t
object (or, in some cases, even two of them) each time it was called.
This is not the intended usage: you're supposed to keep the uuid_t object
around so that the library can cache its state across uses. (Other UUID
libraries seem to keep equivalent state behind-the-scenes in static
variables, but OSSP chose differently.) Aside from being quite inefficient,
creating a new uuid_t loses knowledge of the previously generated UUID,
which in theory could result in duplicate V1-style UUIDs being created
on sufficiently fast machines.
On at least some platforms, creating a new uuid_t also draws some entropy
from /dev/urandom, leaving less for the rest of the system. This seems
sufficiently unpleasant to justify back-patching this change.
Tom Lane [Thu, 29 May 2014 03:15:51 +0000 (23:15 -0400)]
Fix uuid-ossp regression tests based on buildfarm feedback.
The previous version of these tests expected uuid_generate_v1() to always
emit MAC addresses with the local-admin and multicast address bits zero.
However, several of the buildfarm critters are reporting values with the
local-admin bit set. (Perhaps they're running inside VMs or jails.)
And a couple are reporting values with the multicast bit set, probably
meaning that the UUID library couldn't read the system MAC address.
Also, it emerges that if OSSP UUID can't read the system MAC address, it
falls back to V1MC behavior wherein the whole node field gets randomized
each time, breaking the test that expected the node field to remain stable
in V1 output. (It looks like e2fs doesn't behave that way, though.)
It's not entirely clear why we can't get a system MAC address, since the
buildfarm scripts would not work without internet access. Nonetheless,
the regression tests had better cope with the case, so adjust the tests
to expect these behaviors.
It turns out that the %name-prefix syntax without "=" does not work
at all in pre-2.4 Bison. We are not prepared to make such a large
jump in minimum required Bison version just to suppress a warning
message in a version hardly any developers are using yet.
When 3.0 gets more popular, we'll figure out a way to deal with this.
In the meantime, BISONFLAGS=-Wno-deprecated is recommendable for
anyone using 3.0 who doesn't want to see the warning.
Andres Freund [Wed, 28 May 2014 22:32:09 +0000 (00:32 +0200)]
Don't pay heed to wal_sender_timeout while creating a decoding slot.
Sometimes CREATE_REPLICATION_SLOT ... LOGICAL ... needs to wait for
further WAL using WalSndWaitForWal(). That used to always respect
wal_sender_timeout and kill the session when waiting long enough
because no feedback/ping messages can be sent while the slot is still
being created.
Introduce the notion that last_reply_timestamp = 0 means that the
walsender currently doesn't need timeout processing to avoid that
problem. Use that notion for CREATE_REPLICATION_SLOT ... LOGICAL.
Bugreport and initial patch by Steve Singer, revised by me.
The only caller of compareJsonbScalarValue that needed locale-sensitive
comparison of strings was also the only caller that didn't just check for
equality. Separate the two cases for clarity: compareJsonbScalarValue now
does locale-sensitive comparison, and a new function,
equalsJsonbScalarValue, just checks for equality.
Tom Lane [Wed, 28 May 2014 19:41:53 +0000 (15:41 -0400)]
Fix bogus %name-prefix option syntax in all our Bison files.
%name-prefix doesn't use an "=" sign according to the Bison docs, but it
silently accepted one anyway, until Bison 3.0. This was originally a
typo of mine in commit 012abebab1bc72043f3f670bf32e91ae4ee04bd2, and we
seem to have slavishly copied the error into all the other grammar files.
Per report from Vik Fearing; analysis by Peter Eisentraut.
Back-patch to all active branches, since somebody might try to build
a back branch with up-to-date tools.
Tom Lane [Wed, 28 May 2014 18:21:17 +0000 (14:21 -0400)]
Improve regression tests for uuid-ossp.
On reflection, the timestamp-advances test might fail if we're unlucky
enough for the time_mid field to change between two calls, since uuid_cmp
is just bytewise comparison and the field ordering has more significant
fields later. Build some field extraction functions so we can do a more
honest test of that. Also check that the version and reserved fields
contain what they should.
Tom Lane [Wed, 28 May 2014 15:50:41 +0000 (11:50 -0400)]
Fix stack clobber in new uuid-ossp code.
The V5 (SHA1 hashing) code wrote 20 bytes into a 16-byte local variable.
This had accidentally failed to fail in my testing and Matteo's, but
buildfarm results exposed the problem.
Magnus Hagander [Wed, 28 May 2014 10:40:45 +0000 (12:40 +0200)]
Ensure cleanup in case of early errors in streaming base backups
Move the code that sends the initial status information as well as the
calculation of paths inside the ENSURE_ERROR_CLEANUP block. If this code
failed, we would "leak" a counter of number of concurrent backups, thereby
making the system always believe it was in backup mode. This could happen
if the sending failed (which it probably never did given that the small
amount of data to send would never cause a flush) or if the psprintf calls
ran out of memory. Both are very low risk, but all operations after
do_pg_start_backup should be protected.
Tom Lane [Wed, 28 May 2014 04:26:46 +0000 (00:26 -0400)]
pg_lsn should not be marked typispreferred.
In general it's not a good idea for built-in types in the 'U' category
to be marked preferred; they could draw behavior away from user-defined
types with similarly-named operators. pg_lsn is probably at low risk
of that right now given the lack of casts between it and other types,
but that doesn't make this marking OK.
Ordinarily we'd bump catversion when changing any predefined catalog
contents like this, but since we're past beta1, the costs of a forced
initdb seem to outweigh the benefits of guaranteed behavioral consistency.
There's not any known behavioral impact today anyway --- this is more
in the nature of being sure there's not problems in future.
Tom Lane [Wed, 28 May 2014 02:31:21 +0000 (22:31 -0400)]
Fix obsolete config-module-exclusion logic in vcregress.pl.
The recent addition of regression tests to uuid-ossp exposed the fact
that the MSVC build system wasn't being consistent about whether it was
building/testing that contrib module, ie, it would try to test the module
even when it hadn't built it. The same hazard was latent for sslinfo.
For the moment I just copied the more up-to-date logic from point A to
point B, but this is screaming for refactoring.
Tom Lane [Wed, 28 May 2014 02:01:13 +0000 (22:01 -0400)]
Propagate system identifier generation improvement into pg_resetxlog.
Commit 5035701e07e8bd395aa878465a102afd7b74e8c3 improved xlog.c's method
for creating a database system identifier, but I neglected to fix the
copy of that code appearing in pg_resetxlog.c. Spotted by Andres Freund.
Tom Lane [Tue, 27 May 2014 23:42:08 +0000 (19:42 -0400)]
Support BSD and e2fsprogs UUID libraries alongside OSSP UUID library.
Allow the contrib/uuid-ossp extension to be built atop any one of these
three popular UUID libraries. (The extension's name is now arguably a
misnomer, but we'll keep it the same so as not to cause unnecessary
compatibility issues for users.)
We would not normally consider a change like this post-beta1, but the issue
has been forced by our upgrade to autoconf 2.69, whose more rigorous header
checks are causing OSSP's header files to be rejected on some platforms.
It's been foreseen for some time that we'd have to move away from depending
on OSSP UUID due to lack of upstream maintenance, so this is a down payment
on that problem.
While at it, add some simple regression tests, in hopes of catching any
major incompatibilities between the three implementations.
Robert Haas [Tue, 27 May 2014 02:56:28 +0000 (22:56 -0400)]
worker_spi: Initialize bgw_notify_pid in all cases.
Commit 090d0f2050647958865cb495dff74af7257d2bb4 added new code showing
how it can be useful to set bgw_notify_pid to a non-zero value, but it
failed to make sure that the existing call to RegisterBackgroundWorker
initialized the new field at all.
Tom Lane [Tue, 27 May 2014 02:23:29 +0000 (22:23 -0400)]
Avoid unportable usage of sscanf(UINT64_FORMAT).
On Mingw, it seems that scanf() doesn't necessarily accept the same format
codes that printf() does, and in particular it may fail to recognize %llu
even though printf() does. Since configure only probes printf() behavior
while setting up the INT64_FORMAT macros, this means it's unsafe to use
those macros with scanf(). We had only one instance of such a coding
pattern, in contrib/pg_stat_statements, so change that code to avoid
the problem.
Per buildfarm warnings. Back-patch to 9.0 where the troublesome code
was introduced.
Andres Freund [Sun, 25 May 2014 16:47:05 +0000 (18:47 +0200)]
Fix pg_recvlogical to accept the documented -I instead only --startpos.
The bug was caused by omitting 'I:' from the short argument list to
getopt_long(). To make similar bugs in the future less likely reorder
options in --help, long and short option lists to be in the same,
alphabetical within groups, order.
Report and fix by Michael Paquier, some additional reordering by me.
Fix error when trying to delete page with half-dead left sibling.
The new page deletion code didn't cope with the case the target page's
right sibling was marked half-dead. It failed a sanity check which checked
that the downlinks in the parent page match the lower level, because a
half-dead page has no downlink. To cope, check for that condition, and
just give up on the deletion if it happens. The vacuum will finish the
deletion of the half-dead page when it gets there, and on the next vacuum
after that the empty can be deleted.
Tom Lane [Sun, 25 May 2014 20:33:29 +0000 (16:33 -0400)]
Allow total number of transactions in pgbench to exceed INT_MAX.
Change the total-transactions counters from int32 to int64 to accommodate
cases where we do more than 2^31 transactions during a run. This patch
does not change the INT_MAX limit on explicit "-t" parameters, but it
does allow the product of the -t and -c parameters to exceed INT_MAX, or
allow a -T limit that is large enough that more than 2^31 transactions
can be completed. While pgbench did not actually fail in such cases,
it did print an incorrect total-transactions count, and some of the
derived numbers such as TPS would have been wrong as well.
Andres Freund [Sat, 24 May 2014 23:37:52 +0000 (01:37 +0200)]
Don't allocate memory inside an Assert() iff in a critical section.
HeapTupleHeaderGetCmax() asserts that it is only used if the tuple has
been updated by the current transaction. That check is correct and
sensible but requires allocating memory if xmax is a multixact. When
wal_level is set to logical cmax needs to be included in a wal record
, generated inside a critical section, which can trigger the assertion
added in 4a170ee9e.
Andres Freund [Wed, 7 May 2014 20:30:05 +0000 (22:30 +0200)]
Silence a couple of spurious valgrind warnings in inval.c.
Define padding bytes in SharedInvalidationMessage structs to be
defined. Otherwise the sinvaladt.c ringbuffer, which is accessed by
multiple processes, will cause spurious valgrind warnings about
undefined memory being used. That's because valgrind remembers the
undefined bytes from the last local process's store, not realizing
that another process has written since, filling the previously
uninitialized bytes.
Tom Lane [Tue, 20 May 2014 16:20:47 +0000 (12:20 -0400)]
Prevent auto_explain from changing the output of a user's EXPLAIN.
Commit af7914c6627bcf0b0ca614e9ce95d3f8056602bf, which introduced the
EXPLAIN (TIMING) option, for some reason coded explain.c to look at
planstate->instrument->need_timer rather than es->timing to decide
whether to print timing info. However, the former flag might get set
as a result of contrib/auto_explain wanting timing information. We
certainly don't want activation of auto_explain to change user-visible
statement behavior, so fix that.
Also fix an independent bug introduced in the same patch: in the code
path for a never-executed node with a machine-friendly output format,
if timing was selected, it would fail to print the Actual Rows and Actual
Loops items.
Per bug #10404 from Tomonari Katsumata. Back-patch to 9.2 where the
faulty code was introduced.
Fix backup-block numbering in redo of b-tree split.
I got the backup block numbers off-by-one in the commit that changed the
way incomplete-splits are handled. I blame the comments, which said
"backup block 1" and "backup block 2", even though the backup blocks
are numbered starting from 0, in the macros and functions used in replay.
Fix the comments and the code.
Per Jeff Janes' bug report about corruption caused by torn page writes.
The incorrect code is new in git master, but backpatch the comment change
down to 9.0, where the numbering in the redo-side macros was changed.
Tom Lane [Mon, 19 May 2014 04:06:26 +0000 (00:06 -0400)]
Fix non-C89-compatible coding in pgbench.
C89 says that compound initializers may only contain constant expressions;
a restriction violated by commit 89d00cbe. While we've had no actual field
complaints about this, C89 is still the project standard, and it's not
saving all that much code to break compatibility here. So let's adhere to
the old restriction.
In passing, replace a bunch of hardwired constants "256" with
sizeof(target-variable), just because the latter is more readable and
less breakable. And const-ify where possible.
Back-patch to 9.3 where the nonportable code was added.
Tom Lane [Sun, 18 May 2014 22:17:55 +0000 (18:17 -0400)]
Ooops, I broke initdb with that last patch.
That's what I get for not fully retesting the final version of the patch.
The replace_allowed cross-check needs an additional special case for
bootstrapping.
Tom Lane [Sun, 18 May 2014 20:51:46 +0000 (16:51 -0400)]
Fix two ancient memory-leak bugs in relcache.c.
RelationCacheInsert() ignored the possibility that hash_search(HASH_ENTER)
might find a hashtable entry already present for the same OID. However,
that can in fact occur during recursive relcache load scenarios. When it
did happen, we overwrote the pointer to the pre-existing Relation, causing
a session-lifespan leakage of that entire structure. As far as is known,
the pre-existing Relation would always have reference count zero by the
time we arrive back at the outer insertion, so add code that deletes the
pre-existing Relation if so. If by some chance its refcount is positive,
elog a WARNING and allow the pre-existing Relation to be leaked as before.
Also, AttrDefaultFetch() was sloppy about leaking the cstring form of the
pg_attrdef.adbin value it's copying into the relcache structure. This is
only a query-lifespan leakage, and normally not very significant, but it
adds up during CLOBBER_CACHE testing.
These bugs are of very ancient vintage, but I'll refrain from back-patching
since there's no evidence that these leaks amount to anything in ordinary
usage.
Tom Lane [Sat, 17 May 2014 22:29:46 +0000 (18:29 -0400)]
Make fallback implementation of pg_memory_barrier() work.
The fallback implementation involves acquiring and releasing a spinlock
variable that is otherwise unreferenced --- not even to the extent of
initializing it. This accidentally fails to fail on platforms where
spinlocks should be initialized to zeroes, but elsewhere it results in
a "stuck spinlock" failure during startup.
I griped about this last July, and put in a hack that worked for gcc
on HPPA, but didn't get around to fixing the general case. Per the
discussion back then, the best thing to do seems to be to initialize
dummy_spinlock in main.c.
Tom Lane [Sat, 17 May 2014 19:53:21 +0000 (15:53 -0400)]
Fix unaligned accesses in DecodeUpdate().
The xl_heap_header_len structures in an XLOG_HEAP_UPDATE record aren't
necessarily aligned adequately. The regular replay function for these
records is aware of that, but decode.c didn't get the memo. I'm not
sure why the buildfarm failed to catch this; the test_decoding test
certainly blows up real good on my old HPPA box.
Also, I'm pretty sure that the address arithmetic was wrong for the
case of XLOG_HEAP_CONTAINS_OLD and not XLOG_HEAP_CONTAINS_NEW_TUPLE,
though this apparently can't happen when logical decoding is active.
Update README, we don't do post-recovery cleanup actions anymore.
transam/README explained how B-tree incomplete splits were tracked and
fixed after recovery, as an example of handling complex actions that need
multiple WAL records, but that's not how it works anymore. Explain the new
paradigm.
Tom Lane [Fri, 16 May 2014 20:51:28 +0000 (16:51 -0400)]
Make sure chr(int) can't create invalid UTF8 sequences.
Several years ago we changed chr(int) so that if the database encoding is
UTF8, it would interpret its argument as a Unicode code point and expand it
into the appropriate multibyte sequence. However, we weren't sufficiently
careful about checking validity of the input. According to RFC3629, UTF8
disallows code points above U+10FFFF (note that the predecessor standard
RFC2279 was more liberal). Also, both versions of the UTF8 spec agree
that Unicode surrogate-pair codes should never appear in UTF8. Because
our encoding validity checks follow RFC3629, our failure to enforce these
restrictions in chr() means it could be used to produce text strings that
will be rejected when the database is dumped and reloaded. To ensure
consistency with the input functions, let's actually apply
pg_utf8_islegal() to the proposed output of chr().
Per discussion, this seems like too much of a behavioral change to
back-patch, but it's not too late to squeeze it into 9.4.
Tom Lane [Fri, 16 May 2014 19:11:51 +0000 (15:11 -0400)]
Fix valgrind warning for btree_gist indexes on macaddr.
The macaddr opclass stores two macaddr structs (each of size 6) in an
index column that's declared as being of type gbtreekey16, ie 16 bytes.
In the original coding this led to passing a palloc'd value of size 12
to the index insertion code, so that data would be fetched past the
end of the allocated value during index tuple construction. This makes
valgrind unhappy. In principle it could result in a SIGSEGV, though
with the current implementation of palloc there's no risk since
the 12-byte request size would be rounded up to 16 bytes anyway.
To fix, add a field to struct gbtree_ninfo showing the declared size of
the index datums, and use that in the palloc requests; and use palloc0
to be sure that any wasted bytes are cleanly initialized.
Per report from Andres Freund. No back-patch since there's no current
risk of a real problem.
Fix test_decoding test case's check that slot has been dropped.
pg_stat_replication shows connected replication clients. The ddl test case
never has any replication clients connected, so querying pg_stat_replication
is pointless. To check that a slot has been dropped correctly, query
pg_replication_slots instead.
Fix thinko in logical decoding of commit-prepared records.
The decoding of prepared transaction commits accidentally used the XID of
the transaction performing the COMMIT PREPARED, not the XID of the prepared
transaction. Before bb38fb0d43c8d that lead to those transactions not being
decoded, afterwards to a assertion failure.
Open output file before sleeping in pg_recvlogical.
Let's complain about e.g an invalid path or permission problem sooner rather
than later. Before this patch, we would only try to open the output file
after receiving the first decoded message from the server.
Initialize tsId and dbId fields in WAL record of COMMIT PREPARED.
Commit dd428c79 added dbId and tsId to the xl_xact_commit struct but missed
that prepared transaction commits reuse that struct. Fix that.
Because those fields were left unitialized, replaying a commit prepared WAL
record in a hot standby node would fail to remove the relcache init file.
That can lead to "could not open file" errors on the standby. Relcache init
file only needs to be removed when a system table/index is rewritten in the
transaction using two phase commit, so that should be rare in practice. In
HEAD, the incorrect dbId/tsId values are also used for filtering in logical
replication code, causing the transaction to always be filtered out.
Analysis and fix by Andres Freund. Backpatch to 9.0 where hot standby was
introduced.
Tom Lane [Thu, 15 May 2014 19:57:54 +0000 (15:57 -0400)]
Fix unportable setvbuf() usage in initdb.
In yesterday's commit 2dc4f011fd61501cce507be78c39a2677690d44b, I tried
to force buffering of stdout/stderr in initdb to be what it is by
default when the program is run interactively on Unix (since that's how
most manual testing is done). This tripped over the fact that Windows
doesn't support _IOLBF mode. We dealt with that a long time ago in
syslogger.c by falling back to unbuffered mode on Windows. Export that
solution in port.h and use it in initdb.
Fix a couple of bugs in pg_recvlogical output to stdout.
Don't close stdout on SIGHUP. Also, when a SIGHUP is received, close the
file immediately, rather than only after receiving some more data from
the server. Rename a variable, to avoid mentally dealing with double
negatives (not unsynced means synced).
The proc array can contain duplicate XIDs, when a transaction is just being
prepared for two-phase commit. To cope, remove any duplicates in
txid_current_snapshot(). Also ignore duplicates in the input functions, so
that if e.g. you have an old pg_dump file that already contains duplicates,
it will be accepted.
Report and fix by Jan Wieck. Backpatch to all supported versions.
Fix race condition in preparing a transaction for two-phase commit.
To lock a prepared transaction's shared memory entry, we used to mark it
with the XID of the backend. When the XID was no longer active according
to the proc array, the entry was implicitly considered as not locked
anymore. However, when preparing a transaction, the backend's proc array
entry was cleared before transfering the locks (and some other state) to
the prepared transaction's dummy PGPROC entry, so there was a window where
another backend could finish the transaction before it was in fact fully
prepared.
To fix, rewrite the locking mechanism of global transaction entries. Instead
of an XID, just have simple locked-or-not flag in each entry (we store the
locking backend's backend id rather than a simple boolean, but that's just
for debugging purposes). The backend is responsible for explicitly unlocking
the entry, and to make sure that that happens, install a callback to unlock
it on abort or process exit.
Tom Lane [Thu, 15 May 2014 01:13:54 +0000 (21:13 -0400)]
In initdb, ensure stdout/stderr buffering behavior is what we expect.
Since this program may print to either stdout or stderr, the relative
ordering of its messages depends on the buffering behavior of those files.
Force stdout to be line-buffered and stderr to be unbuffered, ensuring
that the behavior will match standard Unix interactive behavior, even
when stdout and stderr are rerouted to a file.
Per complaint from Tomas Vondra. The particular case he pointed out is
new in HEAD, but issues of the same sort could arise in any branch with
other error messages, so back-patch to all branches.
I'm unsure whether we might not want to do this in other client programs
as well. For the moment, just fix initdb.
Tom Lane [Wed, 14 May 2014 18:55:48 +0000 (14:55 -0400)]
Code review for recent changes in relcache.c.
rd_replidindex should be managed the same as rd_oidindex, and rd_keyattr
and rd_idattr should be managed like rd_indexattr. Omissions in this area
meant that the bitmapsets computed for rd_keyattr and rd_idattr would be
leaked during any relcache flush, resulting in a slow but permanent leak in
CacheMemoryContext. There was also a tiny probability of relcache entry
corruption if we ran out of memory at just the wrong point in
RelationGetIndexAttrBitmap. Otherwise, the fields were not zeroed where
expected, which would not bother the code any AFAICS but could greatly
confuse anyone examining the relcache entry while debugging.
Also, create an API function RelationGetReplicaIndex rather than letting
non-relcache code be intimate with the mechanisms underlying caching of
that value (we won't even mention the memory leak there).
Also, fix a relcache flush hazard identified by Andres Freund:
RelationGetIndexAttrBitmap must not assume that rd_replidindex stays valid
across index_open.
The aspects of this involving rd_keyattr date back to 9.3, so back-patch
those changes.
Tom Lane [Wed, 14 May 2014 15:51:10 +0000 (11:51 -0400)]
Make initdb throw error for bad locale values.
Historically we've printed a complaint for a bad locale setting, but then
fallen back to the environment default. Per discussion, this is not such
a great idea, because rectifying an erroneous locale choice post-initdb
(perhaps long after data has been loaded) could be enormously expensive.
Better to complain and give the user a chance to double-check things.
The behavior was particularly bad if the bad setting came from environment
variables rather than a bogus command-line switch: in that case not only
was there a fallback to C/SQL_ASCII, but the printed complaint was quite
unhelpful. It's hard to be entirely sure what variables setlocale looked
at, but we can at least give a hint where the problem might be.
When cache invalidations arrive while ri_LoadConstraintInfo() is busy
filling a new cache entry, InvalidateConstraintCacheCallBack() compares
the - not yet initialized - oidHashValue field with the to-be-invalidated
hash value. To fix, check whether the entry is already marked as invalid.
Initialize padding bytes in btree_gist varbit support.
The code expands a varbit gist leaf key to a node key by copying the bit
data twice in a varlen datum, as both the lower and upper key. The lower key
was expanded to INTALIGN size, but the padding bytes were not initialized.
That's a problem because when the lower/upper keys are compared, the padding
bytes are used compared too, when the values are otherwise equal. That could
lead to incorrect query results.
REINDEX is advised for any btree_gist indexes on bit or bit varying data
type, to fix any garbage padding bytes on disk.
Per Valgrind, reported by Andres Freund. Backpatch to all supported
versions.
Tom Lane [Tue, 13 May 2014 00:21:16 +0000 (20:21 -0400)]
Be more wary in choice of timezone names to test make_timestamptz with.
America/Metlakatla hasn't been in the IANA database all that long, so
some installations might not have it. It does seem worthwhile to test
with a fractional-minute GMT offset, but we can get that from almost
any pre-1900 date; I chose Europe/Paris, whose LMT offset from Greenwich
should be pretty darn well established.
Also, assuming that Mars/Mons_Olympus will never be in the IANA database
seems less than future-proof, so let's use a more fanciful location for
the bad-zone-name check.