Fujii Masao [Tue, 17 Jan 2017 08:27:32 +0000 (17:27 +0900)]
Fix an assertion failure related to an exclusive backup.
Previously multiple sessions could execute pg_start_backup() and
pg_stop_backup() to start and stop an exclusive backup at the same time.
This could trigger the assertion failure of
"FailedAssertion("!(XLogCtl->Insert.exclusiveBackup)".
This happend because, even while pg_start_backup() was starting
an exclusive backup, other session could run pg_stop_backup()
concurrently and mark the backup as not-in-progress unconditionally.
This patch introduces ExclusiveBackupState indicating the state of
an exclusive backup. This state is used to ensure that there is only
one session running pg_start_backup() or pg_stop_backup() at
the same time, to avoid the assertion failure.
Back-patch to all supported versions.
Author: Michael Paquier Reviewed-By: Kyotaro Horiguchi and me Reported-By: Andreas Seltenreich
Discussion: <87mvktojme.fsf@credativ.de>
Tom Lane [Mon, 16 Jan 2017 20:23:11 +0000 (15:23 -0500)]
Fix check_srf_call_placement() to handle VALUES cases correctly.
INSERT ... VALUES with a single VALUES row is implemented quite differently
from the general VALUES case. A user-visible implication of that is that
we accept SRFs in the single-row case, but not in the multi-row case.
That's a historical artifact no doubt, but in view of the lack of field
complaints, I'm not excited about fixing it right now.
However, check_srf_call_placement() needs to know about this, first because
it should throw an error in the unsupported case, and second because it
should set p_hasTargetSRFs in the single-row case (because we treat that
like a SELECT tlist). That's an oversight in commit a4c35ea1c.
To fix, split EXPR_KIND_VALUES into two values. So far as I can see,
this is the only place where we need to distinguish the two cases at
present; but there might be more later.
Tom Lane [Mon, 16 Jan 2017 18:53:40 +0000 (13:53 -0500)]
Fix NULL pointer dereference in tuplesort.c.
Oversight in commit e94568ecc. This could cause a crash when an external
datum tuplesort of a pass-by-value type required multiple passes.
Per report from Mithun Cy.
Magnus Hagander [Mon, 16 Jan 2017 12:56:43 +0000 (13:56 +0100)]
Make pg_basebackup use temporary replication slots
Temporary replication slots will be used by default when wal streaming
is used and no slot name is specified with -S. If a slot name is
specified, then a permanent slot with that name is used. If --no-slot is
specified, then no permanent or temporary slot will be used.
Temporary slots are only used on 10.0 and newer, of course.
Tom Lane [Sun, 15 Jan 2017 19:09:35 +0000 (14:09 -0500)]
Fix matching of boolean index columns to sort ordering.
Normally, if we have a WHERE clause like "indexcol = constant",
the planner will figure out that that index column can be ignored
when determining whether the index has a desired sort ordering.
But this failed to work for boolean index columns, because a
condition like "boolcol = true" is canonicalized to just "boolcol"
which does not give rise to an EquivalenceClass. Add a check to
allow the same type of deduction to be made in this case too.
Per a complaint from Dima Pavlov. Arguably this is a bug, but given the
limited impact and the small number of complaints so far, I won't risk
destabilizing plans in stable branches by back-patching.
Tom Lane [Sat, 14 Jan 2017 21:17:30 +0000 (16:17 -0500)]
Teach contrib/pg_stat_statements to handle multi-statement commands better.
Make use of the statement boundary info added by commit ab1f0c822
to let pg_stat_statements behave more sanely when multiple SQL queries
are jammed into one query string. It now records just the relevant
part of the source string, not the whole thing, for each individual
query.
Even when no multi-statement strings are involved, users may notice small
changes in the output: leading and trailing whitespace and semicolons will
be stripped from statements, which did not happen before.
Also, significantly expand pg_stat_statements' regression test script.
Fabien Coelho, reviewed by Craig Ringer and Kyotaro Horiguchi,
some mods by me
Tom Lane [Sat, 14 Jan 2017 21:02:35 +0000 (16:02 -0500)]
Change representation of statement lists, and add statement location info.
This patch makes several changes that improve the consistency of
representation of lists of statements. It's always been the case
that the output of parse analysis is a list of Query nodes, whatever
the types of the individual statements in the list. This patch brings
similar consistency to the outputs of raw parsing and planning steps:
* The output of raw parsing is now always a list of RawStmt nodes;
the statement-type-dependent nodes are one level down from that.
* The output of pg_plan_queries() is now always a list of PlannedStmt
nodes, even for utility statements. In the case of a utility statement,
"planning" just consists of wrapping a CMD_UTILITY PlannedStmt around
the utility node. This list representation is now used in Portal and
CachedPlan plan lists, replacing the former convention of intermixing
PlannedStmts with bare utility-statement nodes.
Now, every list of statements has a consistent head-node type depending
on how far along it is in processing. This allows changing many places
that formerly used generic "Node *" pointers to use a more specific
pointer type, thus reducing the number of IsA() tests and casts needed,
as well as improving code clarity.
Also, the post-parse-analysis representation of DECLARE CURSOR is changed
so that it looks more like EXPLAIN, PREPARE, etc. That is, the contained
SELECT remains a child of the DeclareCursorStmt rather than getting flipped
around to be the other way. It's now true for both Query and PlannedStmt
that utilityStmt is non-null if and only if commandType is CMD_UTILITY.
That allows simplifying a lot of places that were testing both fields.
(I think some of those were just defensive programming, but in many places,
it was actually necessary to avoid confusing DECLARE CURSOR with SELECT.)
Because PlannedStmt carries a canSetTag field, we're also able to get rid
of some ad-hoc rules about how to reconstruct canSetTag for a bare utility
statement; specifically, the assumption that a utility is canSetTag if and
only if it's the only one in its list. While I see no near-term need for
relaxing that restriction, it's nice to get rid of the ad-hocery.
The API of ProcessUtility() is changed so that what it's passed is the
wrapper PlannedStmt not just the bare utility statement. This will affect
all users of ProcessUtility_hook, but the changes are pretty trivial; see
the affected contrib modules for examples of the minimum change needed.
(Most compilers should give pointer-type-mismatch warnings for uncorrected
code.)
There's also a change in the API of ExplainOneQuery_hook, to pass through
cursorOptions instead of expecting hook functions to know what to pick.
This is needed because of the DECLARE CURSOR changes, but really should
have been done in 9.6; it's unlikely that any extant hook functions
know about using CURSOR_OPT_PARALLEL_OK.
Finally, teach gram.y to save statement boundary locations in RawStmt
nodes, and pass those through to Query and PlannedStmt nodes. This allows
more intelligent handling of cases where a source query string contains
multiple statements. This patch doesn't actually do anything with the
information, but a follow-on patch will. (Passing this information through
cleanly is the true motivation for these changes; while I think this is all
good cleanup, it's unlikely we'd have bothered without this end goal.)
catversion bump because addition of location fields to struct Query
affects stored rules.
This patch is by me, but it owes a good deal to Fabien Coelho who did
a lot of preliminary work on the problem, and also reviewed the patch.
Tom Lane [Sat, 14 Jan 2017 18:27:47 +0000 (13:27 -0500)]
Throw suitable error for COPY TO STDOUT/FROM STDIN in a SQL function.
A client copy can't work inside a function because the FE/BE wire protocol
doesn't support nesting of a COPY operation within query results. (Maybe
it could, but the protocol spec doesn't suggest that clients should support
this, and libpq for one certainly doesn't.)
In most PLs, this prohibition is enforced by spi.c, but SQL functions don't
use SPI. A comparison of _SPI_execute_plan() and init_execution_state()
shows that rejecting client COPY is the only discrepancy in what they
allow, so there's no other similar bugs.
This is an astonishingly ancient oversight, so back-patch to all supported
branches.
Peter Eisentraut [Fri, 13 Jan 2017 17:00:00 +0000 (12:00 -0500)]
pg_ctl: Change default to wait for all actions
The different actions in pg_ctl had different defaults for -w and -W,
mostly for historical reasons. Most users will want the -w behavior, so
make that the default.
Remove the -w option in most example and test code, so avoid confusion
and reduce verbosity. pg_upgrade is not touched, so it can continue to
work with older installations.
Reviewed-by: Beena Emerson <memissemerson@gmail.com> Reviewed-by: Ryan Murphy <ryanfmurphy@gmail.com>
Tom Lane [Fri, 13 Jan 2017 22:32:37 +0000 (17:32 -0500)]
Fix some more regression test row-order-instability issues.
Commit 0563a3a8b just introduced another instance of the same unsafe
testing methodology that appeared in 2ac3ef7a0, which I corrected in 257d81572. Robert/Amit, please stop doing that.
Also look through the rest of f0e44751d's test cases, and correct some
other queries with underdetermined ordering of results from the system
catalogs. These haven't failed in the buildfarm yet, but I don't
have any confidence in that staying true.
Tom Lane [Fri, 13 Jan 2017 21:59:52 +0000 (16:59 -0500)]
In PL/Tcl tests, don't choke if optional error fields are missing.
This fixes a portability issue introduced by commit 961bed020: with a
compiler that doesn't support PG_FUNCNAME_MACRO, the "funcname" field of
errorCode won't be provided, leading to a failure of the unset command.
I added -nocomplain to the unset commands for filename and lineno too, just
in case, though I know of no platform that wouldn't populate those fields.
(BTW, -nocomplain is new in Tcl 8.4, but fortunately we dropped support
for pre-8.4 Tcl some time ago.)
Peter Eisentraut [Fri, 13 Jan 2017 17:00:00 +0000 (12:00 -0500)]
pg_upgrade: Fix for changed pg_ctl default stop mode
In 9.5, the default pg_ctl stop mode was changed from "smart" to "fast".
pg_upgrade still thought the default mode was "smart" and only specified
the mode when "fast" was asked for. This results in using "fast" all
the time. It's not clear what the effect in practice is, but fix it
nonetheless to restore the previous behavior.
Robert Haas [Fri, 13 Jan 2017 19:03:52 +0000 (14:03 -0500)]
Fix a bug in how we generate partition constraints.
Move the code for doing parent attnos to child attnos mapping for Vars
in partition constraint expressions to a separate function
map_partition_varattnos() and call it from the appropriate places.
Doing it in get_qual_from_partbound(), as is now, would produce wrong
result in certain multi-level partitioning cases, because it only
considers the current pair of parent-child relations. In certain
multi-level partitioning cases, attnums for the same key attribute(s)
might differ between various levels causing the same attribute to be
numbered differently in different instances of the Var corresponding
to a given attribute.
With this commit, in generate_partition_qual(), we first generate the
the whole partition constraint (considering all levels of partitioning)
and then do the mapping, so that Vars in the final expression are
numbered according the leaf relation (to which it is supposed to apply).
Robert Haas [Fri, 13 Jan 2017 18:29:31 +0000 (13:29 -0500)]
Fix cardinality estimates for parallel joins.
For a partial path, the cardinality estimate needs to reflect the
number of rows we think each worker will see, rather than the total
number of rows; otherwise, costing will go wrong. The previous coding
got this completely wrong for parallel joins.
Unfortunately, this change may destabilize plans for users of 9.6 who
have enabled parallel query, but since 9.6 is still fairly new I'm
hoping expectations won't be too settled yet. Also, this is really a
brown-paper-bag bug, so leaving it unfixed for the entire lifetime of
9.6 seems unwise.
Related reports (whose import I initially failed to recognize) by
Tomas Vondra and Tom Lane.
Tom Lane [Thu, 12 Jan 2017 23:59:46 +0000 (18:59 -0500)]
Fix field order in struct catcache.
Somebody failed to grasp the point of having the #ifdef CATCACHE_STATS
fields at the end of the struct. Put that back the way it should be,
and add a comment making it more explicit why it should be that way.
Stephen Frost [Wed, 11 Jan 2017 20:45:50 +0000 (15:45 -0500)]
pg_restore: Don't allow non-positive number of jobs
pg_restore will currently accept invalid values for the number of
parallel jobs to run (eg: -1), unlike pg_dump which does check that the
value provided is reasonable.
Worse, '-1' is actually a valid, independent, parameter (as an alias for
--single-transaction), leading to potentially completely unexpected
results from a command line such as:
-> pg_restore -j -1
Where a user would get neither parallel jobs nor a single-transaction.
Add in validity checking of the parallel jobs option, as we already have
in pg_dump, before we try to open up the archive. Also move the check
that we haven't been asked to run more parallel jobs than possible on
Windows to the same place, so we do all the option validity checking
before opening the archive.
Back-patch all the way, though for 9.2 we're adding the Windows-specific
check against MAXIMUM_WAIT_OBJECTS as that check wasn't back-patched
originally.
Stephen Frost [Tue, 10 Jan 2017 16:34:51 +0000 (11:34 -0500)]
pg_dump: Strict names with no matching schema
When using pg_dump --strict-names and a schema pattern which doesn't
match any schemas (eg: --schema='nonexistant*'), we were incorrectly
throwing an error claiming no tables were found when, really, there
were no schemas found:
-> pg_dump --strict-names --schema='nonexistant*'
pg_dump: no matching tables were found for pattern "nonexistant*"
Fix that by changing the error message to say 'schemas' instead, since
that is what we are actually complaining about.
Noticed while testing pg_dump error cases.
Back-patch to 9.6 where --strict-names and this error message were
introduced.
Tom Lane [Mon, 9 Jan 2017 22:47:02 +0000 (17:47 -0500)]
Fix error handling in pltcl_returnnext.
We can't throw elog(ERROR) out of a Tcl command procedure; we have
to catch the error and return TCL_ERROR to the Tcl interpreter.
pltcl_returnnext failed to meet this requirement, so that errors
detected by pltcl_build_tuple_result or other functions called here
led to longjmp'ing out of the Tcl interpreter and thereby leaving it
in a bad state. Use the existing subtransaction support to prevent
that. Oversight in commit 26abb50c4, found more or less accidentally
by the buildfarm thanks to the tests added in 961bed020.
Alvaro Herrera [Mon, 9 Jan 2017 22:26:58 +0000 (19:26 -0300)]
Fix ALTER TABLE / SET TYPE for irregular inheritance
If inherited tables don't have exactly the same schema, the USING clause
in an ALTER TABLE / SET DATA TYPE misbehaves when applied to the
children tables since commit 9550e8348b79. Starting with that commit,
the attribute numbers in the USING expression are fixed during parse
analysis. This can lead to bogus errors being reported during
execution, such as:
ERROR: attribute 2 has wrong type
DETAIL: Table has type smallint, but query expects integer.
Since it wouldn't do to revert to the original coding, we now apply a
transformation to map the attribute numbers to the correct ones for each
child.
Reported by Justin Pryzby
Analysis by Tom Lane; patch by me.
Discussion: https://postgr.es/m/20170102225618.GA10071@telsasoft.com
Alvaro Herrera [Mon, 9 Jan 2017 21:19:29 +0000 (18:19 -0300)]
BRIN revmap pages are not standard pages ...
... and therefore we ought not to tell XLogRegisterBuffer the opposite,
when writing XLog for a brin update that moves the index tuple to a
different page. Otherwise, xlog insertion would try to "compress the
hole" when producing a full-page image for it; but since we don't update
pd_lower/upper, the hole covers the whole page. On WAL replay, the
revmap page becomes empty and so the entire portion of the index is
useless and needs to be recomputed.
This is low-probability: a BRIN update only moves an index tuple to a
different page when the summary tuple is larger than the existing one,
which doesn't happen with fixed-width datatypes. Also, the revmap
page must be first after a checkpoint.
Report and patch: Kuntal Ghosh
Bug is alleged to have detected by a WAL-consistency-checking tool.
Discussion: https://postgr.es/m/CAGz5QCJ=00UQjScSEFbV=0qO5ShTZB9WWz_Fm7+Wd83zPs9Geg@mail.gmail.com
I posted a test case demonstrating the problem, but I'm refraining from
adding it to the test suite; if the WAL consistency tool makes it in,
that will be a better way to catch this from regressing. (We should
definitely have someting that causes not-same-page updates, though.)
Tom Lane [Sat, 7 Jan 2017 21:02:16 +0000 (16:02 -0500)]
Get rid of ParseState.p_value_substitute; use a columnref hook instead.
I noticed that p_value_substitute, which is a single-purpose kluge I added
in 2002 (commit b0422b215), could be replaced by having domainAddConstraint
install a parser hook that looks for the name "value". The parser hook
code only dates back to 2009, so it's not surprising that we had to kluge
this in 2002, but we can do it more cleanly now.
Tom Lane [Sat, 7 Jan 2017 20:34:28 +0000 (15:34 -0500)]
Improve documentation of struct ParseState.
I got annoyed about how some fields of ParseState were documented in the
struct's block comment and some weren't; not all of the latter are trivial.
Fix that. Also reorder a couple of fields that seem to have been placed
rather randomly, or maybe with an idea of avoiding padding space; but there
are never so many ParseStates in existence at one time that we ought to
value pad space over readability.
Stephen Frost [Fri, 6 Jan 2017 21:29:31 +0000 (16:29 -0500)]
Add basic pg_dumpall/pg_restore TAP tests
For reasons unknown, pg_dumpall and pg_restore managed to escape the
basic set of TAP tests that were added for pg_dump in 6bd356c3, so
let's get them added now. A few minor adjustments are also made to the
dump/restore tests to improve code coverage for pg_restore/pg_dumpall.
Tom Lane [Fri, 6 Jan 2017 21:21:57 +0000 (16:21 -0500)]
Merge two copies of tuple-building code in pltcl.c.
Make pltcl_trigger_handler() construct modified tuples using
pltcl_build_tuple_result(), rather than its own copy of essentially
the same logic. This results in slightly different message wording for
the error cases, and in one case a different SQLSTATE, but it seems
unlikely that any existing applications are depending on any of those
details.
While at it, fix a typo in commit 26abb50c4: pltcl_build_tuple_result was
applying encoding conversion in the wrong direction. That would be a
back-patchable bug fix, except the code hasn't shipped yet.
Stephen Frost [Fri, 6 Jan 2017 20:27:47 +0000 (15:27 -0500)]
Protect against NULL-dereference in pg_dump
findTableByOid() is allowed to return NULL and we should therefore be
checking for that case. getOwnedSeqs() and dumpSequence() shouldn't
ever actually see this happen, but given odd circumstances it might and
commit f9e439b1 probably shouldn't have removed that check.
Pointed out by Coverity. Initial patch from Michael Paquier.
Back-patch to 9.6, where that commit had removed the check.
Tom Lane [Fri, 6 Jan 2017 19:12:52 +0000 (14:12 -0500)]
Invalidate cached plans on FDW option changes.
This fixes problems where a plan must change but fails to do so,
as seen in a bug report from Rajkumar Raghuwanshi.
For ALTER FOREIGN TABLE OPTIONS, do this through the standard method of
forcing a relcache flush on the table. For ALTER FOREIGN DATA WRAPPER
and ALTER SERVER, just flush the whole plan cache on any change in
pg_foreign_data_wrapper or pg_foreign_server. That matches the way
we handle some other low-probability cases such as opclass changes, and
it's unclear that the case arises often enough to be worth working harder.
Besides, that gives a patch that is simple enough to back-patch with
confidence.
Back-patch to 9.3. In principle we could apply the code change to 9.2 as
well, but (a) we lack postgres_fdw to test it with, (b) it's doubtful that
anyone is doing anything exciting enough with FDWs that far back to need
this desperately, and (c) the patch doesn't apply cleanly.
Patch originally by Amit Langote, reviewed by Etsuro Fujita and Ashutosh
Bapat, who each contributed substantial changes as well.
This commit purported to use a variable hash seed for Partial
HashAggregate, but actually did the opposite - it made us use a
variable seed for any HashAggregate that is NOT partial. Woops.
Robert Haas [Thu, 5 Jan 2017 18:12:16 +0000 (13:12 -0500)]
Fix possible leak of semaphore count.
Commit 4aec49899e5782247e134f94ce1c6ee926f88e1c reorganized the order
of operations here so that we no longer increment the number of "extra
waits" before locking the semaphore, but it did not change the
starting value of extraWaits from 0 to -1 to compensate. In the worst
case, this could leak a semaphore count, but that seems to be unlikely
in practice.
Robert Haas [Thu, 5 Jan 2017 17:27:09 +0000 (12:27 -0500)]
Fix possible crash reading pg_stat_activity.
With the old code, a backend that read pg_stat_activity without ever
having executed a parallel query might see a backend in the midst of
executing one waiting on a DSA LWLock, resulting in a crash. The
solution is for backends to register the tranche at startup time, not
the first time a parallel query is executed.
Report by Andreas Seltenreich. Patch by me, reviewed by Thomas Munro.
Tom Lane [Thu, 5 Jan 2017 16:33:51 +0000 (11:33 -0500)]
Fix handling of empty arrays in array_fill().
array_fill(..., array[0]) produced an empty array, which is probably
what users expect, but it was a one-dimensional zero-length array
which is not our standard representation of empty arrays. Also, for
no very good reason, it rejected empty input arrays; that case should
be allowed and produce an empty output array.
In passing, remove the restriction that the input array(s) have lower
bound 1. That seems rather pointless, and it would have needed extra
complexity to make the check deal with empty input arrays.
Per bug #14487 from Andrew Gierth. It's been broken all along, so
back-patch to all supported branches.
Tom Lane [Wed, 4 Jan 2017 23:00:11 +0000 (18:00 -0500)]
Handle OID column inheritance correctly in ALTER TABLE ... INHERIT.
Inheritance operations must treat the OID column, if any, much like
regular user columns. But MergeAttributesIntoExisting() neglected to
do that, leading to weird results after a table with OIDs is associated
to a parent with OIDs via ALTER TABLE ... INHERIT.
Report and patch by Amit Langote, reviewed by Ashutosh Bapat, some
adjustments by me. It's been broken all along, so back-patch to
all supported branches.
Robert Haas [Wed, 4 Jan 2017 21:30:16 +0000 (16:30 -0500)]
Improve documentation of timestamp internal representation.
Be more clear that we represent timestamps in microseconds when
integer timestamps are used, and in fractional seconds when
floating-point timestamps are used.
Robert Haas [Wed, 4 Jan 2017 19:36:34 +0000 (14:36 -0500)]
Fix reporting of constraint violations for table partitioning.
After a tuple is routed to a partition, it has been converted from the
root table's row type to the partition's row type. ExecConstraints
needs to report the failure using the original tuple and the parent's
tuple descriptor rather than the ones for the selected partition.
Tom Lane [Wed, 4 Jan 2017 18:36:44 +0000 (13:36 -0500)]
Prefer int-wide pg_atomic_flag over char-wide when using gcc intrinsics.
configure can only probe the existence of gcc intrinsics, not how well
they're implemented, and unfortunately the answer is sometimes "badly".
In particular we've found that multiple compilers fail to implement
char-width __sync_lock_test_and_set() correctly on PPC; and even a correct
implementation would necessarily be pretty inefficient, since that hardware
has only a word-wide primitive to work with.
Given the knowledge we've accumulated in s_lock.h, it appears that it's
best to rely on int-width TAS operations on most non-Intel architectures.
Hence, pick int not char when both are nominally available to us in
generic-gcc.h (note that that code is not used for x86[_64]).
Back-patch to fix regression test failures on FreeBSD/PPC. Ordinarily
back-patching a change like this would be verboten because of ABI breakage.
But since pg_atomic_flag is not yet used in any Postgres data structure,
there's no ABI to break. It seems safer to back-patch to avoid possible
gotchas, if someday we do back-patch something that uses pg_atomic_flag.
Robert Haas [Wed, 4 Jan 2017 18:05:29 +0000 (13:05 -0500)]
Move partition_tuple_slot out of EState.
Commit 2ac3ef7a01df859c62d0a02333b646d65eaec5ff added a TupleTapleSlot
for partition tuple slot to EState (es_partition_tuple_slot) but it's
more logical to have it as part of ModifyTableState
(mt_partition_tuple_slot) and CopyState (partition_tuple_slot).
Tom Lane [Wed, 4 Jan 2017 17:43:52 +0000 (12:43 -0500)]
Re-allow SSL passphrase prompt at server start, but not thereafter.
Leave OpenSSL's default passphrase collection callback in place during
the first call of secure_initialize() in server startup. Although that
doesn't work terribly well in daemon contexts, some people feel we should
not break it for anyone who was successfully using it before. We still
block passphrase demands during SIGHUP, meaning that you can't adjust SSL
configuration on-the-fly if you used a passphrase, but this is no worse
than what it was before commit de41869b6. And we block passphrase demands
during EXEC_BACKEND reloads; that behavior wasn't useful either, but at
least now it's documented.
Tweak some related log messages for more readability, and avoid issuing
essentially duplicate messages about reload failure caused by a passphrase.
Robert Haas [Wed, 4 Jan 2017 17:03:40 +0000 (12:03 -0500)]
Update obsolete comments in lwlock.h.
The typical size of an LWLock is now 16 bytes even on 64-bit platforms,
and the size of slock_t is now irrelevant. But pg_atomic_uint32 can
(perhaps surprisingly) still be larger than 4 bytes, so there's still
some marginal point to allowing LWLOCK_MINIMAL_SIZE == 64.
Simon Riggs [Wed, 4 Jan 2017 16:50:23 +0000 (16:50 +0000)]
Allow PostgresNode.pm tests to wait for catchup
Add methods to the core test framework PostgresNode.pm to allow us to
test that standby nodes have caught up with the master, as well as
basic LSN handling. Used in tests recovery/t/001_stream_rep.pl and
recovery/t/004_timeline_switch.pl
Craig Ringer, reviewed by Aleksander Alekseev and Simon Riggs
Better fix for sequence access in hot standby test
The purpose of the test was to check access to the sequence relation on
a hot standby, so change the test to read a different column from the
sequence, instead of just reading the catalog.
Magnus Hagander [Wed, 4 Jan 2017 09:48:30 +0000 (10:48 +0100)]
Attempt to handle pending-delete files on Windows
These files are deleted but not yet gone from the filesystem. Operations
on them will return ERROR_DELETE_PENDING.
With this we start treating that as ENOENT, meaning files does not
exist (which is the state it will soon reach). This should be safe in
every case except when we try to recreate a file with exactly the same
name. This is an operation that PostgreSQL does very seldom, so
hopefully that won't happen much -- and even if it does, this treatment
should be no worse than treating it as an unhandled error.
We've been un able to reproduce the bug reliably, so pushing this to
master to get buildfarm coverage and other testing. Once it's proven to
be stable, it should be considered for backpatching.
Tom Lane [Tue, 3 Jan 2017 17:33:29 +0000 (12:33 -0500)]
Disable prompting for passphrase while (re)loading SSL config files.
OpenSSL's default behavior when loading a passphrase-protected key file
is to open /dev/tty and demand the password from there. It was kinda
sorta okay to allow that to happen at server start, but really that was
never workable in standard daemon environments. And it was a complete
fail on Windows, where the same thing would happen at every backend launch.
Yesterday's commit de41869b6 put the final nail in the coffin by causing
that to happen at every SIGHUP; even if you've still got a terminal acting
as the server's TTY, having the postmaster freeze until you enter the
passphrase again isn't acceptable.
Hence, override the default behavior with a callback that returns an empty
string, ensuring failure. Change the documentation to say that you can't
have a passphrase-protected server key, period.
If we can think of a production-grade way of collecting a passphrase from
somewhere, we might do that once at server startup and use this callback
to feed it to OpenSSL, but it's far from clear that anyone cares enough
to invest that much work in the feature. The lack of complaints about
the existing fractionally-baked behavior suggests nobody's using it anyway.
Remove bogus notice that older clients might not work with MD5 passwords.
That was written when we still had "crypt" authentication, and it was
referring to the fact that an older client might support "crypt"
authentication but not "md5". But we haven't supported "crypt" for years.
(As soon as we add a new authentication mechanism that doesn't work with
MD5 hashes, we'll need a similar notice again. But this text as it's worded
now is just wrong.)
Tom Lane [Tue, 3 Jan 2017 02:37:12 +0000 (21:37 -0500)]
Allow SSL configuration to be updated at SIGHUP.
It is no longer necessary to restart the server to enable, disable,
or reconfigure SSL. Instead, we just create a new SSL_CTX struct
(by re-reading all relevant files) whenever we get SIGHUP. Testing
shows that this is fast enough that it shouldn't be a problem.
In conjunction with that, downgrade the logic that complains about
pg_hba.conf "hostssl" lines when SSL isn't active: now that's just
a warning condition not an error.
An issue that still needs to be addressed is what shall we do with
passphrase-protected server keys? As this stands, the server would
demand the passphrase again on every SIGHUP, which is certainly
impractical. But the case was only barely supported before, so that
does not seem a sufficient reason to hold up committing this patch.
Andreas Karlsson, reviewed by Michael Banck and Michael Paquier
Tom Lane [Mon, 2 Jan 2017 18:41:51 +0000 (13:41 -0500)]
Use clock_gettime(), if available, in instr_time measurements.
The advantage of clock_gettime() is that the API allows the result to
be precise to nanoseconds, not just microseconds as in gettimeofday().
Now that it's routinely possible to do tens of plan node executions
in 1us, we really need more precision than gettimeofday() can offer
for EXPLAIN ANALYZE to accumulate statistics with.
Some research shows that clock_gettime() is available on pretty nearly
every modern Unix-ish platform, and as far as I have been able to test,
it has about the same execution time as gettimeofday(), so there's no
loss in switching over. (By the same token, this doesn't do anything
to fix the fact that we really wish clock readings were faster. But
there's enough win here to justify changing anyway.)
A small side benefit is that on most platforms, we can use CLOCK_MONOTONIC
instead of CLOCK_REALTIME and thereby render EXPLAIN impervious to
concurrent resets of the system clock. (This means that code must not
assume that the contents of struct instr_time have any well-defined
interpretation as timestamps, but really that was true before.)
Some platforms offer nonstandard clock IDs that might be of interest.
This patch knows we should use CLOCK_MONOTONIC_RAW on macOS, because it
provides more precision and is faster to read than their CLOCK_MONOTONIC.
If there turn out to be many more cases where we need special rules, it
might be appropriate to handle the selection of clock ID in configure,
but for the moment that doesn't seem worth the trouble.
Tom Lane [Mon, 2 Jan 2017 17:26:03 +0000 (12:26 -0500)]
In pgbench logging, avoid assuming that instr_times match Unix timestamps.
For aggregated logging, pg_bench supposed that printing the integer part of
INSTR_TIME_GET_DOUBLE() would produce a Unix timestamp. That was already
broken on Windows, and it's about to get broken on most other platforms as
well. As in commit 74baa1e3b, we can remove the entanglement at the price
of one extra syscall per transaction; though here it seems more convenient
to use time(NULL) instead of gettimeofday(), since we only need
integral-second precision.
I took the time to do some wordsmithing on the documentation about
pgbench's logging features, too.
Tom Lane [Sun, 1 Jan 2017 20:17:08 +0000 (15:17 -0500)]
Avoid assuming that instr_time == struct timeval in pgbench logging.
This code was presuming undue familiarity with the contents of the
instr_time struct. That was already broken on Windows, and it's about
to get broken on most other platforms as well. The simplest solution
that preserves the current output definition is to just do our own
gettimeofday() call here. Realistically, the extra cost is probably
negligible in comparison to everything else that's going on in a
pgbench transaction, so it's not worth sweating over.
On Windows, the precision delivered by gettimeofday() is lower than
one could wish, but this is still a big improvement over printing
zeroes, as the code did before.
Tom Lane [Sat, 31 Dec 2016 23:39:08 +0000 (18:39 -0500)]
Fix unstable regression test results.
Commit 2ac3ef7a0 added a query with an underdetermined output row order;
it has failed multiple times in the buildfarm since then. Add an ORDER BY
to fix. Also, don't rely on a DROP CASCADE to drop in a well-determined
order; that hasn't failed yet but I don't trust it much, and we're not
saving any typing by using CASCADE anyway.
Tom Lane [Thu, 29 Dec 2016 21:57:41 +0000 (16:57 -0500)]
Remove manual breaks in NodeTag assignments to fix duplicate tag numbers.
Commit f0e44751d added new node tags at a place in the tag numbering
where there was no daylight left before the next hard-coded number,
resulting in some duplicate tag assignments. This doesn't seem to have
caused any big problem so far, but it's surely trouble waiting to happen.
We could adjust the manually assigned breakpoints to make more room,
but that just leaves the same hazard waiting to strike again in future.
What seems like a better idea is to get rid of the manual assignments
and leave NodeTags to be automatically assigned, consecutively from one
on up. This means that any change in the tag list forces a backend-wide
recompile, but realistically that's usually needed anyway.
Peter Eisentraut [Wed, 28 Dec 2016 17:00:00 +0000 (12:00 -0500)]
Make more use of RoleSpec struct
Most code was casting this through a generic Node. By declaring
everything as RoleSpec appropriately, we can remove a bunch of casts and
ad-hoc node type checking.
Tom Lane [Tue, 27 Dec 2016 20:43:54 +0000 (15:43 -0500)]
Fix interval_transform so it doesn't throw away non-no-op casts.
interval_transform() contained two separate bugs that caused it to
sometimes mistakenly decide that a cast from interval to restricted
interval is a no-op and throw it away.
First, it was wrong to rely on dt.h's field type macros to have an
ordering consistent with the field's significance; in one case they do
not. This led to mistakenly treating YEAR as less significant than MONTH,
so that a cast from INTERVAL MONTH to INTERVAL YEAR was incorrectly
discarded.
Second, fls(1<<k) produces k+1 not k, so comparing its output directly
to SECOND was wrong. This led to supposing that a cast to INTERVAL
MINUTE was really a cast to INTERVAL SECOND and so could be discarded.
To fix, get rid of the use of fls(), and make a function based on
intervaltypmodout to produce a field ID code adapted to the need here.
Per bug #14479 from Piotr Stefaniak. Back-patch to 9.2 where transform
functions were introduced, because this code was born broken.
Andrew Dunstan [Tue, 27 Dec 2016 16:23:46 +0000 (11:23 -0500)]
Explain unaccounted for space in pgstattuple.
In addition to space accounted for by tuple_len, dead_tuple_len and
free_space, the table_len includes page overhead, the item pointers
table and padding bytes.
Magnus Hagander [Tue, 27 Dec 2016 09:37:11 +0000 (10:37 +0100)]
Don't rename .partial files in pg_receivexlog if an error occured
In 56c7d8d the behavior to keep .partial segments around
(considered corrupt) in case an connection failure occurs was
accidentally removed. This would lead to an incomplete segment
being considered complete.
Tom Lane [Mon, 26 Dec 2016 19:58:02 +0000 (14:58 -0500)]
Remove triggerable Assert in hashname().
hashname() asserted that the key string it is given is shorter than
NAMEDATALEN. That should surely always be true if the input is in fact a
regular value of type "name". However, for reasons of coding convenience,
we allow plain old C strings to be treated as "name" values in many places.
Some SQL functions accept arbitrary "text" inputs, convert them to C
strings, and pass them otherwise-untransformed to syscache lookups for name
columns, allowing an overlength input value to trigger hashname's Assert.
This would be a DOS problem, except that it only happens in assert-enabled
builds which aren't recommended for production. In a production build,
you'll just get a name lookup error, since regardless of the hash value
computed by hashname, the later equality comparison checks can't match.
Likewise, if the catalog lookup is done by seqscan or indexscan searches,
there will just be a lookup error, since the name comparison functions
don't contain any similar length checks, and will see an overlength input
as unequal to any stored entry.
After discussion we concluded that we should simply remove this Assert.
It's inessential to hashname's own functionality, and having such an
assertion in only some paths for name lookup is more of a foot-gun than
a useful check. There may or may not be a case for the affected callers
to do something other than let the name lookup fail, but we'll consider
that separately; in any case we probably don't want to change such
behavior in the back branches.
Per report from Tushar Ahuja. Back-patch to all supported branches.
Tom Lane [Sun, 25 Dec 2016 21:04:31 +0000 (16:04 -0500)]
Fix incorrect error reporting for duplicate data in \crosstabview.
\crosstabview's complaint about multiple entries for the same crosstab
cell quoted the wrong row and/or column values. It would accidentally
appear to work if the data had been in strcmp() order to start with,
which probably explains how we missed noticing this during development.
This could be fixed in more than one way, but the way I chose was to
hang onto both result pointers from bsearch() and use those to get at
the value names.
In passing, avoid casting away const in the bsearch comparison functions.
No bug there, just poor style.
Per bug #14476 from Tomonari Katsumata. Back-patch to 9.6 where
\crosstabview was introduced.
Stephen Frost [Sat, 24 Dec 2016 06:41:59 +0000 (01:41 -0500)]
pg_dumpall: Include --verbose option in --help output
The -v/--verbose option was not included in the output from --help for
pg_dumpall even though it's in the pg_dumpall documentation and has
apparently been around since pg_dumpall was reimplemented in C in 2002.
Stephen Frost [Sat, 24 Dec 2016 02:01:29 +0000 (21:01 -0500)]
Fix tab completion in psql for ALTER DEFAULT PRIVILEGES
When providing tab completion for ALTER DEFAULT PRIVILEGES, we are
including the list of roles as possible options for completion after the
GRANT or REVOKE. Further, we accept FOR ROLE/IN SCHEMA at the same time
and in either order, but the tab completion was only working for one or
the other. Lastly, we weren't using the actual list of allowed kinds of
objects for default privileges for completion after the 'GRANT X ON' but
instead were completeing to what 'GRANT X ON' supports, which isn't the
ssame at all.
Address these issues by improving the forward tab-completion for ALTER
DEFAULT PRIVILEGES and then constrain and correct how the tail
completion is done when it is for ALTER DEFAULT PRIVILEGES.
Back-patch the forward/tail tab-completion to 9.6, where we made it easy
to handle such cases.
For 9.5 and earlier, correct the initial tab-completion to at least be
correct as far as it goes and then add a check for GRANT/REVOKE to only
tab-complete when the GRANT/REVOKE is the start of the command, so we
don't try to do tab-completion after we get to the GRANT/REVOKE part of
the ALTER DEFAULT PRIVILEGES command, which is better than providing
incorrect completions.
Initial patch for master and 9.6 by Gilles Darold, though I cleaned it
up and added a few comments. All bugs in the 9.5 and earlier patch are
mine.
Tom Lane [Fri, 23 Dec 2016 18:35:11 +0000 (13:35 -0500)]
Replace enum InhOption with simple boolean.
Now that it has only INH_NO and INH_YES values, it's just weird that
it's not a plain bool, so make it that way.
Also rename RangeVar.inhOpt to "inh", to be like RangeTblEntry.inh.
My recollection is that we gave it a different name specifically because
it had a different representation than the derived bool value, but it
no longer does. And this is a good forcing function to be sure we
catch any places that are affected by the change.
Bump catversion because of possible effect on stored RangeVar nodes.
I'm not exactly convinced that we ever store RangeVar on disk, but
we have a readfuncs function for it, so be cautious. (If we do do so,
then commit e13486eba was in error not to bump catversion.)
Tom Lane [Fri, 23 Dec 2016 17:53:09 +0000 (12:53 -0500)]
Doc: improve index entry for "median".
We had an index entry for "median" attached to the percentile_cont function
entry, which was pretty useless because a person following the link would
never realize that that function was the one they were being hinted to use.
Instead, make the index entry point at the example in syntax-aggregates,
and add a <seealso> link to "percentile".
Also, since that example explicitly claims to be calculating the median,
make it use percentile_cont not percentile_disc. This makes no difference
in terms of the larger goals of that section, but so far as I can find,
nearly everyone thinks that "median" means the continuous not discrete
calculation.
Per gripe from Steven Winfield. Back-patch to 9.4 where we introduced
percentile_cont.
Tom Lane [Fri, 23 Dec 2016 16:53:35 +0000 (11:53 -0500)]
Spellcheck: s/descendent/descendant/g
I got a little annoyed by reading documentation paragraphs containing
both spellings within a few lines of each other. My dictionary says
"descendant" is the preferred spelling, and it's certainly the majority
usage in our tree, so standardize on that.
For one usage in parallel.sgml, I thought it better to rewrite to avoid
the term altogether.
Peter Eisentraut [Fri, 23 Dec 2016 17:00:00 +0000 (12:00 -0500)]
pg_dump: Remove obsolete handling of sequence names
There was code that attempted to check whether the sequence name stored
inside the sequence was the same as the name in pg_class. But that code
was already ifdef'ed out, and now that the sequence no longer stores its
own name, it's altogether obsolete, so remove it.
Robert Haas [Fri, 23 Dec 2016 12:14:37 +0000 (07:14 -0500)]
Remove _hash_chgbufaccess().
This is basically for the same reasons I got rid of _hash_wrtbuf()
in commit 25216c98938495fd741bf585dcbef45b3a9ffd40: it's not
convenient to have a function which encapsulates MarkBufferDirty(),
especially as we move towards having hash indexes be WAL-logged.
Patch by me, reviewed (but not entirely endorsed) by Amit Kapila.
Joe Conway [Fri, 23 Dec 2016 01:56:50 +0000 (17:56 -0800)]
Improve RLS documentation with respect to COPY
Documentation for pg_restore said COPY TO does not support row security
when in fact it should say COPY FROM. Fix that.
While at it, make it clear that "COPY FROM" does not allow RLS to be
enabled and INSERT should be used instead. Also that SELECT policies
will apply to COPY TO statements.
Back-patch to 9.5 where RLS first appeared.
Author: Joe Conway Reviewed-By: Dean Rasheed and Robert Haas
Discussion: https://postgr.es/m/5744FA24.3030008%40joeconway.com
Robert Haas [Thu, 22 Dec 2016 22:31:52 +0000 (17:31 -0500)]
Fix tuple routing in cases where tuple descriptors don't match.
The previous coding failed to work correctly when we have a
multi-level partitioned hierarchy where tables at successive levels
have different attribute numbers for the partition key attributes. To
fix, have each PartitionDispatch object store a standalone
TupleTableSlot initialized with the TupleDesc of the corresponding
partitioned table, along with a TupleConversionMap to map tuples from
the its parent's rowtype to own rowtype. After tuple routing chooses
a leaf partition, we must use the leaf partition's tuple descriptor,
not the root table's. To that end, a dedicated TupleTableSlot for
tuple routing is now allocated in EState.
Stephen Frost [Thu, 22 Dec 2016 22:08:43 +0000 (17:08 -0500)]
Use TSConfigRelationId in AlterTSConfiguration()
When we are altering a text search configuration, we are getting the
tuple from pg_ts_config and using its OID, so use TSConfigRelationId
when invoking any post-alter hooks and setting the object address.
Further, in the functions called from AlterTSConfiguration(), we're
saving information about the command via
EventTriggerCollectAlterTSConfig(), so we should be setting
commandCollected to true. Also add a regression test to
test_ddl_deparse for ALTER TEXT SEARCH CONFIGURATION.
Author: Artur Zakirov, a few additional comments by me
Discussion: https://www.postgresql.org/message-id/57a71eba-f2c7-e7fd-6fc0-2126ec0b39bd%40postgrespro.ru
Back-patch the fix for the InvokeObjectPostAlterHook() call to 9.3 where
it was introduced, and the fix for the ObjectAddressSet() call and
setting commandCollected to true to 9.5 where those changes to
ProcessUtilitySlow() were introduced.
Tom Lane [Thu, 22 Dec 2016 21:23:33 +0000 (16:23 -0500)]
Fix CREATE TABLE ... LIKE ... WITH OIDS.
Having a WITH OIDS specification should result in the creation of an OID
column, but commit b943f502b broke that in the case that there were LIKE
tables without OIDS. Commentary in that patch makes it look like this was
intentional, but if so it was based on a faulty reading of what inheritance
does: the parent tables can add an OID column, but they can't subtract one.
AFAICS, the behavior ought to be that you get an OID column if any of the
inherited tables, LIKE tables, or WITH clause ask for one.
Also, revert that patch's unnecessary split of transformCreateStmt's loop
over the tableElts list into two passes. That seems to have been based on
a misunderstanding as well: we already have two-pass processing here,
we don't need three passes.
Per bug #14474 from Jeff Dafoe. Back-patch to 9.6 where the misbehavior
was introduced.
Tom Lane [Thu, 22 Dec 2016 20:01:27 +0000 (15:01 -0500)]
Fix handling of expanded objects in CoerceToDomain and CASE execution.
When the input value to a CoerceToDomain expression node is a read-write
expanded datum, we should pass a read-only pointer to any domain CHECK
expressions and then return the original read-write pointer as the
expression result. Previously we were blindly passing the same pointer to
all the consumers of the value, making it possible for a function in CHECK
to modify or even delete the expanded value. (Since a plpgsql function
will absorb a passed-in read-write expanded array as a local variable
value, it will in fact delete the value on exit.)
A similar hazard of passing the same read-write pointer to multiple
consumers exists in domain_check() and in ExecEvalCase, so fix those too.
The fix requires adding MakeExpandedObjectReadOnly calls at the appropriate
places, which is simple enough except that we need to get the data type's
typlen from somewhere. For the domain cases, solve this by redefining
DomainConstraintRef.tcache as okay for callers to access; there wasn't any
reason for the original convention against that, other than not wanting the
API of typcache.c to be any wider than it had to be. For CASE, there's
no good solution except to add a syscache lookup during executor start.
Per bug #14472 from Marcos Castedo. Back-patch to 9.5 where expanded
values were introduced.