Tom Lane [Sun, 30 Oct 2016 21:35:42 +0000 (17:35 -0400)]
Fix nasty performance problem in tsquery_rewrite().
tsquery_rewrite() tries to find matches to subsets of AND/OR conditions;
for example, in the query 'a | b | c' the substitution subquery 'a | c'
should match and lead to replacement of the first and third items.
That's fine, but the matching algorithm apparently takes about O(2^N)
for an N-clause query (I say "apparently" because the code is also both
unintelligible and uncommented). We could probably do better than that
even without any extra assumptions --- but actually, we know that the
subclauses are sorted, indeed are depending on that elsewhere in this very
same function. So we can just scan the two lists a single time to detect
matches, as though we were doing a merge join.
Also do a re-flattening call (QTNTernary()) in tsquery_rewrite_query, just
to make sure that the tree fits the expectations of the next search cycle.
I didn't try to devise a test case for this, but I'm pretty sure that the
oversight could have led to failure to match in some cases where a match
would be expected.
Improve comments, and also stick a CHECK_FOR_INTERRUPTS into
dofindsubquery, just in case it's still too slow for somebody.
Per report from Andreas Seltenreich. Back-patch to all supported branches.
Tom Lane [Sun, 30 Oct 2016 19:24:40 +0000 (15:24 -0400)]
Fix bogus tree-flattening logic in QTNTernary().
QTNTernary() contains logic to flatten, eg, '(a & b) & c' into 'a & b & c',
which is all well and good, but it tries to do that to NOT nodes as well,
so that '!!a' gets changed to '!a'. Explicitly restrict the conversion to
be done only on AND and OR nodes, and add a test case illustrating the bug.
In passing, provide some comments for the sadly naked functions in
tsquery_util.c, and simplify some baroque logic in QTNFree(), which
I think may have been leaking some items it intended to free.
Noted while investigating a complaint from Andreas Seltenreich.
Back-patch to all supported versions.
Tom Lane [Sun, 30 Oct 2016 16:27:41 +0000 (12:27 -0400)]
Improve speed of aggregates that use array_append as transition function.
In the previous coding, if an aggregate's transition function returned an
expanded array, nodeAgg.c and nodeWindowAgg.c would always copy it and thus
force it into the flat representation. This led to ping-ponging between
flat and expanded formats, which costs a lot. For an aggregate using
array_append as transition function, I measured about a 15X slowdown
compared to the pre-9.5 code, when working on simple int[] arrays.
Of course, the old code was already O(N^2) in this usage due to copying
flat arrays all the time, but it wasn't quite this inefficient.
To fix, teach nodeAgg.c and nodeWindowAgg.c to allow expanded transition
values without copying, so long as the transition function takes care to
return the transition value already properly parented under the aggcontext.
That puts a bit of extra responsibility on the transition function, but
doing it this way allows us to not need any extra logic in the fast path
of advance_transition_function (ie, with a pass-by-value transition value,
or with a modified-in-place pass-by-reference value). We already know
that that's a hot spot so I'm loath to add any cycles at all there. Also,
while only array_append currently knows how to follow this convention,
this solution allows other transition functions to opt-in without needing
to have a whitelist in the core aggregation code.
(The reason we would need a whitelist is that currently, if you pass a
R/W expanded-object pointer to an arbitrary function, it's allowed to do
anything with it including deleting it; that breaks the core agg code's
assumption that it should free discarded values. Returning a value under
aggcontext is the transition function's signal that it knows it is an
aggregate transition function and will play nice. Possibly the API rules
for expanded objects should be refined, but that would not be a
back-patchable change.)
With this fix, an aggregate using array_append is no longer O(N^2), so it's
much faster than pre-9.5 code rather than much slower. It's still a bit
slower than the bespoke infrastructure for array_agg, but the differential
seems to be only about 10%-20% rather than orders of magnitude.
Robert Haas [Fri, 28 Oct 2016 16:21:15 +0000 (12:21 -0400)]
pgstattuple: Don't take heavyweight locks when examining a hash index.
It's currently necessary to take a heavyweight lock when scanning a
hash bucket, but pgstattuple only examines individual pages, so it
doesn't need to do this. If, for some hypothetical reason, it did
need to do any heavyweight locking here, this logic would probably
still be incorrect, because most of the locks that it is taking are
meaningless. Only a heavyweight lock on a primary bucket page has any
meaning, but this takes heavyweight locks on all pages regardless of
function - and in particular overflow pages, where you might imagine
that we'd want to lock the primary bucket page if we needed to lock
anything at all.
This is arguably a bug that has existed since this code was added in
commit dab42382f483c3070bdce14a4d93c5d0cf61e82b, but I'm not going to
bother back-patching it because in most cases the only consequence is
that running pgstattuple() on a hash index is a little slower than it
otherwise might be, which is no big deal.
Extracted from a vastly larger patch by Amit Kapila which heavyweight
locking for hash indexes entirely; analysis of why this can be done
independently of the rest by me.
Peter Eisentraut [Thu, 27 Oct 2016 16:00:00 +0000 (12:00 -0400)]
Remove invitation to report a bug about unknown encoding
The error message when we couldn't determine the encoding from a locale
said to report a bug about that. That might have been appropriate when
this code was first added, but by now this works pretty solidly and any
encodings we don't recognize we probably just don't support. We still
print the warning, but no longer invite the bug report.
Peter Eisentraut [Thu, 27 Oct 2016 16:00:00 +0000 (12:00 -0400)]
Add function name to PyArg_ParseTuple()
This causes the supplied function name to appear in any error message,
making the error message friendlier and relieving us from having to
provide our own in some cases.
Robert Haas [Thu, 27 Oct 2016 15:19:51 +0000 (11:19 -0400)]
Fix possible pg_basebackup failure on standby with "include WAL".
If a restartpoint flushed no dirty buffers, it could fail to update
the minimum recovery point, leading to a minimum recovery point prior
to the starting REDO location. perform_base_backup() would interpret
that as meaning that no WAL files at all needed to be included in the
backup, failing an internal sanity check. To fix, have restartpoints
always update the minimum recovery point to just after the checkpoint
record itself, so that the file (or files) containing the checkpoint
record will always be included in the backup.
Code by Amit Kapila, per a design suggestion by me, with some
additional work on the code comment by me. Test case by Michael
Paquier. Report by Kyotaro Horiguchi.
Tom Lane [Wed, 26 Oct 2016 21:05:06 +0000 (17:05 -0400)]
Fix incorrect trigger-property updating in ALTER CONSTRAINT.
The code to change the deferrability properties of a foreign-key constraint
updated all the associated triggers to match; but a moment's examination of
the code that creates those triggers in the first place shows that only
some of them should track the constraint's deferrability properties. This
leads to odd failures in subsequent exercise of the foreign key, as the
triggers are fired at the wrong times. Fix that, and add a regression test
comparing the trigger properties produced by ALTER CONSTRAINT with those
you get by creating the constraint as-intended to begin with.
Per report from James Parks. Back-patch to 9.4 where this ALTER
functionality was introduced.
Tom Lane [Wed, 26 Oct 2016 17:40:41 +0000 (13:40 -0400)]
Fix not-HAVE_SYMLINK code in zic.c.
I broke this in commit f3094920a. Apparently it's dead code anyway,
at least as far as our buildfarm is concerned (and the upstream IANA
code doesn't worry at all about symlink() not being present).
But as long as the rest of our code is willing to guard against not
having symlink(), this should too. Noted while investigating a
tangentially-related complaint from Sandeep Thakkar.
Tom Lane [Wed, 26 Oct 2016 15:46:25 +0000 (11:46 -0400)]
Doc: improve documentation about inheritance.
Clarify documentation about inheritance of check constraints, in
particular mentioning the NO INHERIT option, which didn't exist when
this text was written.
Document that in an inherited query, the applicable row security policies
are those of the explicitly-named table, not its children. This is the
intended behavior (per off-list discussion with Stephen Frost), and there
are regression tests for it, but it wasn't documented anywhere user-facing
as far as I could find.
Do a bit of wordsmithing on the description of inherited access-privilege
checks.
Turns out that the output format of Python Decimal isn't totally platform-
independent either. There are other tests for multi-dimensional arrays,
so rather than try to fix this test case, just remove it.
Instead of treating all python sequence types as array dimensions, except
for tuples and various kinds of strings, only treat Python lists as
dimensions. The PyBytes_Check() function used previously is only available
on Python 2.6 and newer, and it was a bit fiddly anyway. The list of
exceptions would require adjustment if Python got a new kind of a sequence
similar to bytes/unicodes/strings, so only checking for Lists seems more
future-proof. The documentation only mentioned using Lists, so this is
closer to what was documented, anyway.
This should fix the buildfarm failures on systems building with Python 2.5,
although I don't have Python 2.5 installed myself to test with.
Avoid using platform-dependent floats in test case.
The number of decimals printed for floats varied in this test case, as
noted by several buildfarm members. There's nothing special about floats
and arrays in the code being tested, so replace the floats with numerics to
make the output platform-independent.
Multi-dimensional arrays can now be used as arguments to a PL/python function
(used to throw an error), and they can be returned as nested Python lists.
This makes a backwards-incompatible change to the handling of composite
types in arrays. Previously, you could return an array of composite types
as "[[col1, col2], [col1, col2]]", but now that is interpreted as a two-
dimensional array. Composite types in arrays must now be returned as
Python tuples, not lists, to resolve the ambiguity. I.e. "[(col1, col2),
(col1, col2)]".
To avoid breaking backwards-compatibility, when not necessary, () is still
accepted for arrays at the top-level, but it is always treated as a
single-dimensional array. Likewise, [] is still accepted for composite types,
when they are not in an array. Update the documentation to recommend using []
for arrays, and () for composite types, with a mention that those other things
are also accepted in some contexts.
This needs to be mentioned in the release notes.
Alexey Grishchenko, Dave Cramer and me. Reviewed by Pavel Stehule.
Peter Eisentraut [Tue, 25 Oct 2016 16:00:00 +0000 (12:00 -0400)]
pg_dump: Simplify internal archive version handling
The ArchiveHandle structure contained the archive format version number
twice, once as a single field and once split into components. Simplify
that by just keeping the single field and adding some macros to extract
the components. Introduce some macros for composing version numbers, to
eliminate the repeated use of magic formulas. Drop the unused trailing
zero byte from the run-time composite version representation.
Robert Haas [Tue, 25 Oct 2016 02:36:24 +0000 (22:36 -0400)]
postgres_fdw: Try again to stabilize aggregate pushdown regression tests.
A query that only aggregates one row isn't a great argument for pushdown,
and buildfarm member brolga decides against it. Adjust the query a bit
in the hopes of getting remote aggregation to win consistently.
Alvaro Herrera [Mon, 24 Oct 2016 12:27:24 +0000 (09:27 -0300)]
Preserve commit timestamps across clean restart
An oversight in setting the boundaries of known commit timestamps during
startup caused old commit timestamps to become inaccessible after a
server restart.
Author and reporter: Julien Rouhaud
Review, test code: Craig Ringer
Tom Lane [Sun, 23 Oct 2016 23:14:32 +0000 (19:14 -0400)]
Avoid testing tuple visibility without buffer lock.
INSERT ... ON CONFLICT (specifically ExecCheckHeapTupleVisible) contains
another example of this unsafe coding practice. It is much harder to get
a failure out of it than the case fixed in commit 6292c2339, because in
most scenarios any hint bits that could be set would have already been set
earlier in the command. However, Konstantin Knizhnik reported a failure
with a custom transaction manager, and it's clearly possible to get a
failure via a race condition in async-commit mode.
For lack of a reproducible example, no regression test case in this
commit.
I did some testing with Asserts added to tqual.c's functions, and can say
that running "make check-world" exposed these two bugs and no others.
The Asserts are messy enough that I've not added them to the code for now.
Tom Lane [Sun, 23 Oct 2016 22:36:13 +0000 (18:36 -0400)]
Don't throw serialization errors for self-conflicts in INSERT ON CONFLICT.
A transaction that conflicts against itself, for example
INSERT INTO t(pk) VALUES (1),(1) ON CONFLICT DO NOTHING;
should behave the same regardless of isolation level. It certainly
shouldn't throw a serialization error, as retrying will not help.
We got this wrong due to the ON CONFLICT logic not considering the case,
as reported by Jason Dusek.
Core of this patch is by Peter Geoghegan (based on an earlier patch by
Thomas Munro), though I didn't take his proposed code refactoring for fear
that it might have unexpected side-effects. Test cases by Thomas Munro
and myself.
Tom Lane [Sun, 23 Oct 2016 19:01:24 +0000 (15:01 -0400)]
Avoid testing tuple visibility without buffer lock in RI_FKey_check().
Despite the argumentation I wrote in commit 7a2fe85b0, it's unsafe to do
this, because in corner cases it's possible for HeapTupleSatisfiesSelf
to try to set hint bits on the target tuple; and at least since 8.2 we
have required the buffer content lock to be held while setting hint bits.
The added regression test exercises one such corner case. Unpatched, it
causes an assertion failure in assert-enabled builds, or otherwise would
cause a hint bit change in a buffer we don't hold lock on, which given
the right race condition could result in checksum failures or other data
consistency problems. The odds of a problem in the field are probably
pretty small, but nonetheless back-patch to all supported branches.
Magnus Hagander [Sun, 23 Oct 2016 13:16:31 +0000 (15:16 +0200)]
Allow pg_basebackup to stream transaction log in tar mode
This will write the received transaction log into a file called
pg_wal.tar(.gz) next to the other tarfiles instead of writing it to
base.tar. When using fetch mode, the transaction log is still written to
base.tar like before, and when used against a pre-10 server, the file
is named pg_xlog.tar.
To do this, implement a new concept of a "walmethod", which is
responsible for writing the WAL. Two implementations exist, one that
writes to a plain directory (which is also used by pg_receivexlog) and
one that writes to a tar file with optional compression.
Tom Lane [Sat, 22 Oct 2016 18:04:51 +0000 (14:04 -0400)]
Improve documentation about use of Linux huge pages.
Show how to get the system's huge page size, rather than misleadingly
referring to PAGE_SIZE (which is usually understood to be the regular
page size). Show how to confirm whether huge pages have been allocated.
Minor wordsmithing. Back-patch to 9.4 where this section appeared.
Robert Haas [Fri, 21 Oct 2016 15:27:32 +0000 (11:27 -0400)]
postgres_fdw: Attempt to stabilize regression results.
Set enable_hashagg to false for tests involving least_agg(), so that
we get the same plan regardless of local costing variances. Also,
remove a test involving sqrt(); it's there to test deparsing of
HAVING clauses containing expressions, but that's tested elsewhere
anyway, and sqrt(2) deparses with different amounts of precision on
different machines.
Tom Lane [Fri, 21 Oct 2016 15:01:35 +0000 (11:01 -0400)]
Doc: wording tweak for PERL, PYTHON, TCLSH configuration variables.
Replace "Full path to ..." with "Full path name of ...". At least one
user has misinterpreted the existing wording as meaning "Directory
containing ...".
Robert Haas [Fri, 21 Oct 2016 13:54:29 +0000 (09:54 -0400)]
postgres_fdw: Push down aggregates to remote servers.
Now that the upper planner uses paths, and now that we have proper hooks
to inject paths into the upper planning process, it's possible for
foreign data wrappers to arrange to push aggregates to the remote side
instead of fetching all of the rows and aggregating them locally. This
figures to be a massive win for performance, so teach postgres_fdw to
do it.
Jeevan Chalke and Ashutosh Bapat. Reviewed by Ashutosh Bapat with
additional testing by Prabhat Sahu. Various mostly cosmetic changes
by me.
Tom Lane [Thu, 20 Oct 2016 21:17:50 +0000 (17:17 -0400)]
Fix EXPLAIN so that it doesn't emit invalid XML in corner cases.
With track_io_timing = on, EXPLAIN (ANALYZE, BUFFERS) will emit fields
named like "I/O Read Time". The slash makes that invalid as an XML
element name, so that adding FORMAT XML would produce invalid XML.
We already have code in there to translate spaces to dashes, so let's
generalize that to convert anything that isn't a valid XML name character,
viz letters, digits, hyphens, underscores, and periods. We could just
reject slashes, which would run a bit faster. But the fact that this went
unnoticed for so long doesn't give me a warm feeling that we'd notice the
next creative violation, so let's make it a permanent fix.
Reported by Markus Winand, though this isn't his initial patch proposal.
Back-patch to 9.2 where track_io_timing was added. The problem is only
latent in 9.1, so I don't feel a need to fix it there.
Tom Lane [Thu, 20 Oct 2016 19:40:07 +0000 (15:40 -0400)]
Sync our copy of the timezone library with IANA release tzcode2016h.
This absorbs a fix for a symlink-manipulation bug in zic that was
introduced in 2016g. It probably isn't interesting for our use-case,
but I'm not quite sure, so let's update while we're at it.
Robert Haas [Thu, 20 Oct 2016 15:24:37 +0000 (11:24 -0400)]
Rename "pg_xlog" directory to "pg_wal".
"xlog" is not a particularly clear abbreviation for "write-ahead log",
and it sometimes confuses users into believe that the contents of the
"pg_xlog" directory are not critical data, leading to unpleasant
consequences. So, rename the directory to "pg_wal".
This patch modifies pg_upgrade and pg_basebackup to understand both
the old and new directory layouts; the former is necessary given the
purpose of the tool, while the latter merely avoids an unnecessary
backward-compatibility break.
We may wish to consider renaming other programs, switches, and
functions which still use the old "xlog" naming to also refer to
"wal". However, that's still under discussion, so let's do just this
much for now.
Robert Haas [Thu, 20 Oct 2016 14:23:39 +0000 (10:23 -0400)]
Remove a comment which is now incorrect.
Before 5d305d86bd917723f09ab4f15c075d90586a210a, this comment was
correct, but now it says we do something which we don't actually do.
Accordingly, remove the comment.
Tom Lane [Wed, 19 Oct 2016 22:55:52 +0000 (18:55 -0400)]
Sync our copy of the timezone library with IANA release tzcode2016g.
This is mostly to absorb some corner-case fixes in zic for year-2037
timestamps. The other changes that have been made are unlikely to affect
our usage, but nonetheless we may as well take 'em.
Tom Lane [Wed, 19 Oct 2016 22:11:49 +0000 (18:11 -0400)]
Suppress "Factory" zone in pg_timezone_names view for tzdata >= 2016g.
IANA got rid of the really silly "abbreviation" and replaced it with one
that's only moderately silly. But it's still pointless, so keep on not
showing it.
Tom Lane [Wed, 19 Oct 2016 21:56:38 +0000 (17:56 -0400)]
Update time zone data files to tzdata release 2016g.
DST law changes in Turkey. Historical corrections for America/Los_Angeles,
Europe/Kirov, Europe/Moscow, Europe/Samara, and Europe/Ulyanovsk.
Rename Asia/Rangoon to Asia/Yangon, with a backward compatibility link.
The IANA crew continue their campaign to replace invented time zone
abbrevations with numeric GMT offsets. This update changes numerous zones
in Antarctica and the former Soviet Union, for instance Antarctica/Casey
now reports "+08" not "AWST" in the pg_timezone_names view. I kept these
abbreviations in the tznames/ data files, however, so that we will still
accept them for input. (We may want to start trimming those files someday,
but today is not that day.)
An exception is that since IANA no longer claims that "AMT" is in use
in Armenia for GMT+4, I replaced it in the Default file with GMT-4,
corresponding to Amazon Time which is in use in South America. It may be
that that meaning is also invented and IANA will drop it in a future
update; but for now, it seems silly to give pride of place to a meaning
not traceable to IANA over one that is.
Peter Eisentraut [Wed, 19 Oct 2016 16:00:00 +0000 (12:00 -0400)]
Use pg_ctl promote -w in TAP tests
Switch TAP tests to use the new wait mode of pg_ctl promote. This
allows avoiding extra logic with poll_query_until() to be sure that a
promoted standby is ready for read-write queries.
Peter Eisentraut [Wed, 19 Oct 2016 16:00:00 +0000 (12:00 -0400)]
initdb pg_basebackup: Rename --noxxx options to --no-xxx
--noclean and --nosync were the only options spelled without a hyphen,
so change this for consistency with other options. The options in
pg_basebackup have not been in a release, so we just rename them. For
initdb, we retain the old variants.
When a relation is truncated, it is important that the FSM is truncated as
well. Otherwise, after recovery, the FSM can return a page that has been
truncated away, leading to errors like:
ERROR: could not read block 28991 in file "base/16390/572026": read only 0
of 8192 bytes
We were using MarkBufferDirtyHint() to dirty the buffer holding the last
remaining page of the FSM, but during recovery, that might in fact not
dirty the page, and the FSM update might be lost.
To fix, use the stronger MarkBufferDirty() function. MarkBufferDirty()
requires us to do WAL-logging ourselves, to protect from a torn page, if
checksumming is enabled.
Also fix an oversight in visibilitymap_truncate: it also needs to WAL-log
when checksumming is enabled.
Robert Haas [Tue, 18 Oct 2016 19:55:03 +0000 (15:55 -0400)]
Improve regression test coverage for hash indexes.
On my system, this improves coverage for src/backend/access/hash from
61.3% of lines to 88.2% of lines, and from 83.5% of functions to 97.5%
of functions, which is pretty good for 36 lines of tests.
Mithun Cy, reviewing by Amit Kapila and Álvaro Herrera
Tom Lane [Tue, 18 Oct 2016 16:24:46 +0000 (12:24 -0400)]
Fix cidin() to handle values above 2^31 platform-independently.
CommandId is declared as uint32, and values up to 4G are indeed legal.
cidout() handles them properly by treating the value as unsigned int.
But cidin() was just using atoi(), which has platform-dependent behavior
for values outside the range of signed int, as reported by Bart Lengkeek
in bug #14379. Use strtoul() instead, as xidin() does.
In passing, make some purely cosmetic changes to make xidin/xidout
look more like cidin/cidout; the former didn't have a monopoly on
best practice IMO.
Neither xidin nor cidin make any attempt to throw error for invalid input.
I didn't change that here, and am not sure it's worth worrying about
since neither is really a user-facing type. The point is just to ensure
that indubitably-valid inputs work as expected.
It's been like this for a long time, so back-patch to all supported
branches.
Robert Haas [Mon, 17 Oct 2016 20:31:13 +0000 (16:31 -0400)]
By default, set log_line_prefix = '%m [%p] '.
This value might not be to everyone's taste; in particular, some
people might prefer %t to %m, and others may want %u, %d, or other
fields. However, it's a vast improvement on the old default of ''.
Fix use-after-free around DISTINCT transition function calls.
Have tuplesort_gettupleslot() copy the contents of its current table slot
as needed. This is based on an approach taken by tuplestore_gettupleslot().
In the future, tuplesort_gettupleslot() may also be taught to avoid copying
the tuple where caller can determine that that is safe (the
tuplestore_gettupleslot() interface already offers this option to callers).
Patch by Peter Geoghegan. Fixes bug #14344, reported by Regina Obe.
Replace PostmasterRandom() with a stronger way of generating randomness.
This adds a new routine, pg_strong_random() for generating random bytes,
for use in both frontend and backend. At the moment, it's only used in
the backend, but the upcoming SCRAM authentication patches need strong
random numbers in libpq as well.
pg_strong_random() is based on, and replaces, the existing implementation
in pgcrypto. It can acquire strong random numbers from a number of sources,
depending on what's available:
- OpenSSL RAND_bytes(), if built with OpenSSL
- On Windows, the native cryptographic functions are used
- /dev/urandom
- /dev/random
Original patch by Magnus Hagander, with further work by Michael Paquier
and me.
Andres Freund [Sat, 15 Oct 2016 00:22:51 +0000 (17:22 -0700)]
Use more efficient hashtable for execGrouping.c to speed up hash aggregation.
The more efficient hashtable speeds up hash-aggregations with more than
a few hundred groups significantly. Improvements of over 120% have been
measured.
Due to the the different hash table queries that not fully
determined (e.g. GROUP BY without ORDER BY) may change their result
order.
The conversion is largely straight-forward, except that, due to the
static element types of simplehash.h type hashes, the additional data
some users store in elements (e.g. the per-group working data for hash
aggregaters) is now stored in TupleHashEntryData->additional. The
meaning of BuildTupleHashTable's entrysize (renamed to additionalsize)
has been changed to only be about the additionally stored size. That
size is only used for the initial sizing of the hash-table.
Reviewed-By: Tomas Vondra
Discussion: <20160727004333.r3e2k2y6fvk2ntup@alap3.anarazel.de>
Andres Freund [Fri, 14 Oct 2016 23:05:30 +0000 (16:05 -0700)]
Use more efficient hashtable for tidbitmap.c to speed up bitmap scans.
Use the new simplehash.h to speed up tidbitmap.c uses. For bitmap scan
heavy queries speedups of over 100% have been measured. Both lossy and
exact scans benefit, but the wins are bigger for mostly exact scans.
The conversion is mostly trivial, except that tbm_lossify() now restarts
lossifying at the point it previously stopped. Otherwise the hash table
becomes unbalanced because the scan in done in hash-order, leaving the
end of the hashtable more densely filled then the beginning. That caused
performance issues with dynahash as well, but due to the open chaining
they were less pronounced than with the linear adressing from
simplehash.h.
Reviewed-By: Tomas Vondra
Discussion: <20160727004333.r3e2k2y6fvk2ntup@alap3.anarazel.de>
Andres Freund [Fri, 14 Oct 2016 23:05:30 +0000 (16:05 -0700)]
Add a macro templatized hashtable.
dynahash.c hash tables aren't quite fast enough for some
use-cases. There are several reasons for lacking performance:
- the use of chaining for collision handling makes them cache
inefficient, that's especially an issue when the tables get bigger.
- as the element sizes for dynahash are only determined at runtime,
offset computations are somewhat expensive
- hash and element comparisons are indirect function calls, causing
unnecessary pipeline stalls
- it's two level structure has some benefits (somewhat natural
partitioning), but increases the number of indirections
to fix several of these the hash tables have to be adjusted to the
individual use-case at compile-time. C unfortunately doesn't provide a
good way to do compile code generation (like e.g. c++'s templates for
all their weaknesses do). Thus the somewhat ugly approach taken here is
to allow for code generation using a macro-templatized header file,
which generates functions and types based on a prefix and other
parameters.
Later patches use this infrastructure to use such hash tables for
tidbitmap.c (bitmap scans) and execGrouping.c (hash aggregation,
...). In queries where these use up a large fraction of the time, this
has been measured to lead to performance improvements of over 100%.
There are other cases where this could be useful (e.g. catcache.c).
The hash table design chosen is a variant of linear open-addressing. The
biggest disadvantage of simple linear addressing schemes are highly
variable lookup times due to clustering, and deletions leaving a lot of
tombstones around. To address these issues a variant of "robin hood"
hashing is employed. Robin hood hashing optimizes chaining lengths by
moving elements close to their optimal bucket ("rich" elements), out of
the way if a to-be-inserted element is further away from its optimal
position (i.e. it's "poor"). While that can make insertions slower, the
average lookup performance is a lot better, and higher fill factors can
be used in a still performant manner. To avoid tombstones - which
normally solve the issue that a deleted node's presence is relevant to
determine whether a lookup needs to continue looking or is done -
buckets following a deleted element are shifted backwards, unless
they're empty or already at their optimal position.
There's further possible improvements that can be made to this
implementation. Amongst others:
- Use distance as a termination criteria during searches. This is
generally a good idea, but I've been able to see the overhead of
distance calculations in some cases.
- Consider combining the 'empty' status into the hashvalue, and enforce
storing the hashvalue. That could, in some cases, increase memory
density and remove a few instructions.
- Experiment further with the, very conservatively choosen, fillfactor.
- Make maximum size of hashtable configurable, to allow storing very
very large tables. That'd require 64bit hash values to be more common
than now, though.
- some smaller memcpy calls could be optimized to copy larger chunks
But since the new implementation is already considerably faster than
dynahash it seem sensible to start using it.
Reviewed-By: Tomas Vondra
Discussion: <20160727004333.r3e2k2y6fvk2ntup@alap3.anarazel.de>
Andres Freund [Fri, 14 Oct 2016 23:05:30 +0000 (16:05 -0700)]
Add likely/unlikely() branch hint macros.
These are useful for very hot code paths. Because it's easy to guess
wrongly about likelihood, and because such likelihoods change over time,
they should be used sparingly.
Past tests have shown it'd be a good idea to use them in some places,
e.g. in error checks around ereports that ERROR out, but that's work for
later.
Tom Lane [Fri, 14 Oct 2016 20:28:34 +0000 (16:28 -0400)]
Fix assorted integer-overflow hazards in varbit.c.
bitshiftright() and bitshiftleft() would recursively call each other
infinitely if the user passed INT_MIN for the shift amount, due to integer
overflow in negating the shift amount. To fix, clamp to -VARBITMAXLEN.
That doesn't change the results since any shift distance larger than the
input bit string's length produces an all-zeroes result.
Also fix some places that seemed inadequately paranoid about input typmods
exceeding VARBITMAXLEN. While a typmod accepted by anybit_typmodin() will
certainly be much less than that, at least some of these spots are
reachable with user-chosen integer values.
Tom Lane [Thu, 13 Oct 2016 23:45:58 +0000 (19:45 -0400)]
Fix handling of pgstat counters for TRUNCATE in a prepared transaction.
pgstat_twophase_postcommit is supposed to duplicate the math in
AtEOXact_PgStat, but it had missed out the bit about clearing
t_delta_live_tuples/t_delta_dead_tuples for a TRUNCATE.
It's harder than you might think to replicate the issue here, because
those counters would only be nonzero when a previous transaction in
the same backend had added/deleted tuples in the truncated table,
and those counts hadn't been sent to the stats collector yet.
Evident oversight in commit d42358efb. I've not added a regression
test for this; we tried to add one in d42358efb, and had to revert it
because it was too timing-sensitive for the buildfarm.
Tom Lane [Thu, 13 Oct 2016 21:05:14 +0000 (17:05 -0400)]
Fix another bug in merging of inherited CHECK constraints.
It's not good for an inherited child constraint to be marked connoinherit;
that would result in the constraint not propagating to grandchild tables,
if any are created later. The code mostly prevented this from happening
but there was one case that was missed.
This is somewhat related to commit e55a946a8, which also tightened checks
on constraint merging. Hence, back-patch to 9.2 like that one. This isn't
so much because there's a concrete feature-related reason to stop there,
as to avoid having more distinct behaviors than we have to in this area.
Tom Lane [Thu, 13 Oct 2016 20:08:16 +0000 (16:08 -0400)]
Remove dead code in pg_dump.
I'm not sure if this provision for "pg_backup" behaving a bit differently
from "pg_dump" ever did anything useful in a released version. But it's
definitely dead code now.
Tom Lane [Thu, 13 Oct 2016 19:06:46 +0000 (15:06 -0400)]
Try to find out the actual hugepage size when making a MAP_HUGETLB request.
Even if Linux's mmap() is okay with a partial-hugepage request, munmap()
is not, as reported by Chris Richards. Therefore it behooves us to try
a bit harder to find out the actual hugepage size, instead of assuming
that we can skate by with a guess.
For the moment, just look into /proc/meminfo to find out the default
hugepage size, and use that. Later, on kernels that support requests
for nondefault sizes, we might try to consider other alternatives.
But that smells more like a new feature than a bug fix, especially if
we want to provide any way for the DBA to control it, so leave it for
another day.
I set this up to allow easy addition of platform-specific code for
non-Linux platforms, if needed; but right now there are no reports
suggesting that we need to work harder on other platforms.
Back-patch to 9.4 where hugepage support was introduced.
Tom Lane [Thu, 13 Oct 2016 17:59:56 +0000 (13:59 -0400)]
Clean up handling of anonymous mmap'd shared-memory segment.
Fix detaching of the mmap'd segment to have its own on_shmem_exit callback,
rather than piggybacking on the one for detaching from the SysV segment.
That was confusing, and given the distance between the two attach calls,
it was trouble waiting to happen.
Make the detaching calls idempotent by clearing AnonymousShmem to show
we've already unmapped. I spent quite a bit of time yesterday trying
to find a path that would allow the munmap()'s to be done twice, and
while I did not succeed, it seems silly that there's even a question.
Make the #ifdef logic less confusing by separating "do we want to use
anonymous shmem" from EXEC_BACKEND. Even though there's no current
scenario where those conditions are different, it is not helpful for
different places in the same file to be testing EXEC_BACKEND for what
are fundamentally different reasons.
Don't do on_exit_reset() in StartBackgroundWorker(). At best that's
useless (InitPostmasterChild would have done it already) and at worst
it could zap some callback that's unrelated to shared memory.
Improve comments, and simplify the huge_pages enablement logic slightly.
Back-patch to 9.4 where hugepage support was introduced.
Arguably this should go into 9.3 as well, but the code looks
significantly different there, and I doubt it's worth the
trouble of adapting the patch given I can't show a live bug.
Tom Lane [Thu, 13 Oct 2016 14:46:22 +0000 (10:46 -0400)]
Fix pg_dumpall regression test to be locale-independent.
The expected results in commit b4fc64578 seem to have been generated
in a non-C locale, which just points up the fact that the ORDER BY
clause was locale-sensitive.
Tom Lane [Thu, 13 Oct 2016 04:25:28 +0000 (00:25 -0400)]
Fix broken jsonb_set() logic for replacing array elements.
Commit 0b62fd036 did a fairly sloppy job of refactoring setPath()
to support jsonb_insert() along with jsonb_set(). In its defense,
though, there was no regression test case exercising the case of
replacing an existing element in a jsonb array.
Per bug #14366 from Peng Sun. Back-patch to 9.6 where bug was introduced.
Andres Freund [Thu, 13 Oct 2016 01:31:45 +0000 (18:31 -0700)]
Fix further hash table order dependent tests.
Similar to 0137caf273, this makes contrib and pl tests less dependant on
hash-table order. After this commit, at least some order affecting
changes to execGrouping.c don't result in regression test changes
anymore.
Tom Lane [Wed, 12 Oct 2016 22:01:43 +0000 (18:01 -0400)]
Revert addition of PGDLLEXPORT in PG_FUNCTION_INFO_V1 macro.
This turns out not to be as harmless as I thought: MSVC will complain
if it sees an "extern" declaration without PGDLLEXPORT and then one with.
(Seems fairly silly, given that this can be changed after the fact by the
linker, but there you have it.) Therefore, contrib modules that have
extern's for V1 functions in header files are falling over in the
buildfarm, since none of those externs are marked PGDLLEXPORT.
We might or might not conclude that we're willing to plaster those
declarations with PGDLLEXPORT in HEAD, but in any case there's no way we're
going to ship this change in the back branches. Third-party authors would
not thank us for breaking their code in a minor release. Hence, revert
the addition of PGDLLEXPORT (but let's keep the extra info in the comment).
If we do the other changes we can revert this commit in HEAD.
Tom Lane [Wed, 12 Oct 2016 19:11:31 +0000 (15:11 -0400)]
pg_dump's getTypes() needn't retrieve typinput or typoutput anymore.
Commit 64f3524e2 removed the stanza of code that examined these values.
I failed to notice they were unnecessary because my compiler didn't
warn about the un-read variables. Noted by Peter Eisentraut.
Tom Lane [Wed, 12 Oct 2016 18:54:08 +0000 (14:54 -0400)]
Remove unnecessary int2vector-specific hash function and equality operator.
These functions were originally added in commit d8cedf67a to support
use of int2vector columns as catcache lookup keys. However, there are
no catcaches that use such columns. (Indeed I now think it must always
have been dead code: a catcache with such a key column would need an
underlying unique index on the column, but we've never had an int2vector
btree opclass.)
Getting rid of the int2vector-specific operator and function does not
lose any functionality, because operations on int2vectors will now fall
back to the generic anyarray support. This avoids a wart that a btree
index on an int2vector column (made using anyarray_ops) would fail to
match equality searches, because int2vectoreq wasn't a member of the
opclass. We don't really care much about that, since int2vector is not
meant as a type for users to use, but it's silly to have extra code and
less functionality.
If we ever do want a catcache to be indexed by an int2vector column,
we'd need to put back full btree and hash opclasses for int2vector,
comparable to the support for oidvector. (The anyarray code can't be
used at such a low level, because it needs to do catcache lookups.)
But we'll deal with that if/when the need arises.
Also worth noting is that removal of the hash int2vector_ops opclass will
break any user-created hash indexes on int2vector columns. While hash
anyarray_ops would serve the same purpose, it would probably not compute
the same hash values and thus wouldn't be on-disk-compatible. Given that
int2vector isn't a user-facing type and we're planning other incompatible
changes in hash indexes for v10 anyway, this doesn't seem like something
to worry about, but it's probably worth mentioning here.
Tom Lane [Wed, 12 Oct 2016 16:45:50 +0000 (12:45 -0400)]
Provide DLLEXPORT markers for C functions via PG_FUNCTION_INFO_V1 macro.
This isn't really necessary for our own code, because we use a .DEF file
in MSVC builds (see gendef.pl), or --export-all-symbols in MinGW and
Cygwin builds, to ensure that all global symbols in loadable modules
will be exported on Windows. However, third-party authors might use
different build processes that need this marker, and it's harmless
enough for our own builds.
To some extent, this is an oversight in commit e7128e8db, so back-patch
to 9.4 where that was added.
Tom Lane [Wed, 12 Oct 2016 16:19:56 +0000 (12:19 -0400)]
Remove pg_dump/pg_dumpall support for dumping from pre-8.0 servers.
The need for dumping from such ancient servers has decreased to about nil
in the field, so let's remove all the code that catered to it. Aside
from removing a lot of boilerplate variant queries, this allows us to not
have to cope with servers that don't have (a) schemas or (b) pg_depend.
That means we can get rid of assorted squishy code around that. There
may be some nonobvious additional simplifications possible, but this patch
already removes about 1500 lines of code.
I did not remove the ability for pg_restore to read custom-format archives
generated by these old versions (and light testing says that that does
still work). If you have an old server, you probably also have a pg_dump
that will work with it; but you have an old custom-format backup file,
that might be all you have.
It'd be possible at this point to remove fmtQualifiedId()'s version
argument, but I refrained since that would affect code outside pg_dump.
Pass the buffer size as argument to LogicalTapeRewindForRead, rather than
setting it earlier with the separate LogicTapeAssignReadBufferSize call.
This way, the buffer size is set closer to where it's actually used, which
makes the code easier to understand.
This makes the calculation for how much memory to use for the buffers less
precise. We now use the same amount of memory for every tape, rounded down
to the nearest BLCKSZ boundary, instead of using one more block for some
tapes, to get the total up to exact amount of memory available. That should
be OK, merging isn't too sensitive to the exact amount of memory used.
projects / postgresql / log