has_parallel_hazard() was ignoring the proparallel markings for
aggregates, which is no good. Fix that. There was no way to mark
an aggregate as actually being parallel-safe, either, so add a
PARALLEL option to CREATE AGGREGATE.
Patch by me, reviewed by David Rowley.
Experimentation shows this only costs about 6kB, which seems well
worth it given the major performance effects that can be caused
by insufficient alignment, especially on larger systems.
Discussion: 14166.1458924422@sss.pgh.pa.us
PL/Python failed if a PL/Python function was invoked recursively via SPI,
since arguments are passed to the function in its global dictionary
(a horrible decision that's far too ancient to undo) and it would delete
those dictionary entries on function exit, leaving the outer recursion
level(s) without any arguments. Not deleting them would be little better,
since the outer levels would then see the innermost level's arguments.
Since PL/Python uses ValuePerCall mode for evaluating set-returning
functions, it's possible for multiple executions of the same SRF to be
interleaved within a query. PL/Python failed in such a case, because
it stored only one iterator per function, directly in the function's
PLyProcedure struct. Moreover, one interleaved instance of the SRF
would see argument values that should belong to another.
Hence, invent code for saving and restoring the argument entries. To fix
the recursion case, we only need to save at recursive entry and restore
at recursive exit, so the overhead in non-recursive cases is negligible.
To fix the SRF case, we have to save when suspending a SRF and restore
when resuming it, which is potentially not negligible; but fortunately
this is mostly a matter of manipulating Python object refcounts and
should not involve much physical data copying.
Also, store the Python iterator and saved argument values in a structure
associated with the SRF call site rather than the function itself. This
requires adding a memory context deletion callback to ensure that the SRF
state is cleaned up if the calling query exits before running the SRF to
completion. Without that we'd leak a refcount to the iterator object in
such a case, resulting in session-lifespan memory leakage. (In the
pre-existing code, there was no memory leak because there was only one
iterator pointer, but what would happen is that the previous iterator
would be resumed by the next query attempting to use the SRF. Hardly the
semantics we want.)
We can buy back some of whatever overhead we've added by getting rid of
PLy_function_delete_args(), which seems a useless activity: there is no
need to delete argument entries from the global dictionary on exit,
since the next time anyone would see the global dict is on the next
fresh call of the PL/Python function, at which time we'd overwrite those
entries with new arg values anyway.
Also clean up some really ugly coding in the SRF implementation, including
such gems as returning directly out of a PG_TRY block. (The only reason
that failed to crash hard was that all existing call sites immediately
exited their own PG_TRY blocks, popping the dangling longjmp pointer before
there was any chance of it being used.)
In principle this is a bug fix; but it seems a bit too invasive relative to
its value for a back-patch, and besides the fix depends on memory context
callbacks so it could not go back further than 9.5 anyway.
Alexey Grishchenko and Tom Lane
This lets us use parallel aggregate for a variety of useful cases
that didn't work before, like sum(int8), sum(numeric), several
versions of avg(), and various other functions.
Add some regression tests, as well, testing the general sanity of
these and future catalog entries.
David Rowley, reviewed by Tomas Vondra, with a few further changes
by me.
Previously non-exclusive backups had to be done using the replication protocol
and pg_basebackup. With this commit it's now possible to make them using
pg_start_backup/pg_stop_backup as well, as long as the backup program can
maintain a persistent connection to the database.
Doing this, backup_label and tablespace_map are returned as results from
pg_stop_backup() instead of being written to the data directory. This makes
the server safe from a crash during an ongoing backup, which can be a problem
with exclusive backups.
The old syntax of the functions remain and work exactly as before, but since the
new syntax is safer this should eventually be deprecated and removed.
Only reference documentation is included. The main section on backup still needs
to be rewritten to cover this, but since that is already scheduled for a separate
large rewrite, it's not included in this patch.
Reviewed by David Steele and Amit Kapila
As noted by Julian Schauder in bug #14063, the configuration-file parser
doesn't support embedded newlines in string literals. While there might
someday be a good reason to remove that restriction, there doesn't seem
to be one right now. However, ALTER SYSTEM SET could accept strings
containing newlines, since many of the variable-specific value-checking
routines would just see a newline as whitespace. This led to writing a
postgresql.auto.conf file that was broken and had to be removed manually.
Pending a reason to work harder, just throw an error if someone tries this.
In passing, fix several places in the ALTER SYSTEM logic that failed to
provide an errcode() for an ereport(), and thus would falsely log the
failure as an internal XX000 error.
Back-patch to 9.4 where ALTER SYSTEM was introduced.
Alex Shulgin complained that the underlying strategy wasn't all that
apparent, particularly not the fact that we intentionally have two
code paths depending on whether we think the column has a limited set
of possible values or not. Try to make it clearer.
On reflection, the pre-existing logic in ANALYZE is specifically meant to
compare the frequency of a candidate MCV against the estimated frequency of
a random distinct value across the whole table. The change to compare it
against the average frequency of values actually seen in the sample doesn't
seem very principled, and if anything it would make us less likely not more
likely to consider a value an MCV. So revert that, but keep the aspect of
considering only nonnull values, which definitely is correct.
In passing, rename the local variables in these stanzas to
"ndistinct_table", to avoid confusion with the "ndistinct" that appears at
an outer scope in compute_scalar_stats.
\gexec executes the just-entered query, like \g, but instead of printing
the results it takes each field as a SQL command to send to the server.
Computing a series of queries to be executed is a fairly common thing,
but up to now you always had to resort to kluges like writing the queries
to a file and then inputting the file. Now it can be done with no
intermediate step.
The implementation is fairly straightforward except for its interaction
with FETCH_COUNT. ExecQueryUsingCursor isn't capable of being called
recursively, and even if it were, its need to create a transaction
block interferes unpleasantly with the desired behavior of \gexec after
a failure of a generated query (i.e., that it can continue). Therefore,
disable use of ExecQueryUsingCursor when doing the master \gexec query.
We can still apply it to individual generated queries, however, and there
might be some value in doing so.
While testing this feature's interaction with single-step mode, I (tgl) was
led to conclude that SendQuery needs to recognize SIGINT (cancel_pressed)
as a negative response to the single-step prompt. Perhaps that's a
back-patchable bug fix, but for now I just included it here.
Corey Huinker, reviewed by Jim Nasby, Daniel Vérité, and myself
This elevel is useful for logging audit messages and similar information
that should not be passed to the client. It's equivalent to LOG in terms
of decisions about logging priority in the postmaster log, but messages
with this elevel will never be sent to the client.
In the current implementation, it's just an alias for the longstanding
COMMERROR elevel (or more accurately, we've made COMMERROR an alias for
this). At some point it might be interesting to allow a LOG_ONLY flag to
be attached to any elevel, but that would be considerably more complicated,
and it's not clear there's enough use-cases to justify the extra work.
For now, let's just take the easy 90% solution.
David Steele, reviewed by Fabien Coelho, Petr Jelínek, and myself
The first iteration of the signal-checking loop would compute sigmask(0)
which expands to 1<<(-1) which is undefined behavior according to the
C standard. The lack of field reports of trouble suggest that it
evaluates to 0 on all existing Windows compilers, but that's hardly
something to rely on. Since signal 0 isn't a queueable signal anyway,
we can just make the loop iterate from 1 instead, and save a few cycles
as well as avoiding the undefined behavior.
In passing, avoid evaluating the volatile expression UNBLOCKED_SIGNAL_QUEUE
twice in a row; there's no reason to waste cycles like that.
Noted by Aleksander Alekseev, though this isn't his proposed fix.
Back-patch to all supported branches.
When adjusting the estimate for the number of distinct values from a
rel in a grouped query to take into account the selectivity of the
rel's restrictions, use a formula that is less likely to produce
under-estimates.
The old formula simply multiplied the number of distinct values in the
rel by the restriction selectivity, which would be correct if the
restrictions were fully correlated with the grouping expressions, but
can produce significant under-estimates in cases where they are not
well correlated.
The new formula is based on the random selection probability, and so
assumes that the restrictions are not correlated with the grouping
expressions. This is guaranteed to produce larger estimates, and of
course risks over-estimating in cases where the restrictions are
correlated, but that has less severe consequences than
under-estimating, which might lead to a HashAgg that consumes an
excessive amount of memory.
This could possibly be improved upon in the future by identifying
correlated restrictions and using a hybrid of the old and new
formulae.
Author: Tomas Vondra, with some hacking be me
Reviewed-by: Mark Dilger, Alexander Korotkov, Dean Rasheed and Tom Lane
Discussion: http://www.postgresql.org/message-id/flat/56CD0381.5060502@2ndquadrant.com
If archive_timeout > 0 we should avoid logging XLOG_RUNNING_XACTS if idle.
Bug 13685 reported by Laurence Rowe, investigated in detail by Michael Paquier,
though this is not his proposed fix.
20151016203031.3019.72930@wrigleys.postgresql.org
Simple non-invasive patch to allow later backpatch to 9.4 and 9.5
The restore() function assumed that the result of sprintf() with %e format
would necessarily contain an 'e', which is false: what if the supplied
number is an infinity or NaN? If that did happen, we'd get a
null-pointer-dereference core dump. The case appears impossible currently,
because seg_in() does not accept such values, and there are no seg-creating
functions that would create one. But it seems unwise to rely on it never
happening in future.
Quite aside from that, the code was pretty ugly: it relied on modifying a
static format string when it could use a "*" precision argument, and it
used strtok() entirely gratuitously, and it stripped off trailing spaces
by hand instead of just not asking for them to begin with.
Coverity noticed the potential null pointer dereference (though I wonder
why it didn't complain years ago, since this code is ancient).
Since this is just code cleanup and forestalling a hypothetical future
bug, there seems no need for back-patching.
The code was supposing that rd_amcache wouldn't disappear from under it
during a scan; which is wrong. Copy the data out of the relcache rather
than trying to reference it there.
Coverity complained about implicit sign-extension in the
BloomPageGetFreeSpace macro, probably because sizeOfBloomTuple isn't wide
enough for size calculations. No overflow is really possible as long as
maxoff and sizeOfBloomTuple are small enough to represent a realistic
situation, but it seems like a good idea to declare sizeOfBloomTuple as
Size not int32.
Add missing check on BloomPageAddItem() result, again from Coverity.
Avoid core dump due to not allocating so->sign array when
scan->numberOfKeys is zero. Also thanks to Coverity.
Use FLEXIBLE_ARRAY_MEMBER rather than declaring an array as size 1
when it isn't necessarily.
Very minor beautification of related code.
Unfortunately, none of the Coverity-detected mistakes look like they
could account for the remaining buildfarm unhappiness with this
module. It's barely possible that the FLEXIBLE_ARRAY_MEMBER mistake
does account for that, if it's enabling bogus compiler optimizations;
but I'm not terribly optimistic. We probably still have bugs to
find here.
Replay of XLOG_BTREE_VACUUM during Hot Standby was previously thought to require
complex interlocking that matched the requirements on the master. This required
an O(N) operation that became a significant problem with large indexes, causing
replication delays of seconds or in some cases minutes while the
XLOG_BTREE_VACUUM was replayed.
This commit skips the pin scan that was previously required, by observing in
detail when and how it is safe to do so, with full documentation. The pin
scan is skipped only in replay; the VACUUM code path on master is not
touched here and WAL is identical.
The current commit applies in all cases, effectively replacing commit
687f2cd7a0.
Often, upon getting an unexpected error in psql, one's first wish is that
the verbosity setting had been higher; for example, to be able to see the
schema-name field or the server code location info. Up to now the only way
has been to adjust the VERBOSITY variable and repeat the failing query.
That's a pain, and it doesn't work if the error isn't reproducible.
This commit adds a psql feature that redisplays the most recent server
error at full verbosity, without needing to make any variable changes or
re-execute the failed command. We just need to hang onto the latest error
PGresult in case the user executes \errverbose, and then apply libpq's
new PQresultVerboseErrorMessage() function to it. This will consume
some trivial amount of psql memory, but otherwise the cost when the
feature isn't used should be negligible.
Alex Shulgin, reviewed by Daniel Vérité, some improvements by me
Often, upon getting an unexpected error in psql, one's first wish is that
the verbosity setting had been higher; for example, to be able to see the
schema-name field or the server code location info. Up to now the only way
has been to adjust the VERBOSITY variable and repeat the failing query.
That's a pain, and it doesn't work if the error isn't reproducible.
This commit adds support in libpq for regenerating the error message for
an existing error PGresult at any desired verbosity level. This is almost
just a matter of refactoring the existing code into a subroutine, but there
is one bit of possibly-needed information that was not getting put into
PGresults: the text of the last query sent to the server. We must add that
string to the contents of an error PGresult. But we only need to save it
if it might be used, which with the existing error-formatting code only
happens if there is a PG_DIAG_STATEMENT_POSITION error field, which is
probably pretty rare for errors in production situations. So really the
overhead when the feature isn't used should be negligible.
Alex Shulgin, reviewed by Daniel Vérité, some improvements by me
Some buildfarm members are showing "comparison is always false due to
limited range of data type" complaints on this test, so #ifdef it out
on machines with 32-bit int.
Looking at result of buildfarm member jaguarundi it seems to me that
BloomOptions isn't inited sometime, but I don't see yet how it's possible.
Nevertheless, check of signature length's is missed, so, add
a limit of it. Also add missed GenericXLogAbort() in case of already
deleted page in vacuum + minor code refactoring.
As with the previous patch, large numbers of null rows could skew this
calculation unfavorably, causing us to discard values that have a
legitimate claim to be MCVs, since our definition of MCV is that it's
most common among the non-null population of the column. Hence, make
the numerator of avgcount be the number of non-null sample values not
the number of sample rows; likewise for maxmincount in the
compute_scalar_stats variant.
Also, make the denominator be the number of distinct values actually
observed in the sample, rather than reversing it back out of the computed
stadistinct. This avoids depending on the accuracy of the Haas-Stokes
approximation, and really it's what we want anyway; the threshold should
depend only on what we see in the sample, not on what we extrapolate
about the contents of the whole column.
Alex Shulgin, reviewed by Tomas Vondra and myself
Previously, we included null rows in the values of n and N that went
into the formula, which amounts to considering null as a value in its
own right; but the d and f1 values do not include nulls. This is
inconsistent, and it contributes to significant underestimation of
ndistinct when the column is mostly nulls. In any case stadistinct
is defined as the number of distinct non-null values, so we should
exclude nulls when doing this computation.
This is an aboriginal bug in our application of the Haas-Stokes formula,
but we'll refrain from back-patching for fear of destabilizing plan
choices in released branches.
While at it, make the code a bit more readable by omitting unnecessary
casts and intermediate variables.
Observation and original patch by Tomas Vondra, adjusted to fix both
uses of the formula by Alex Shulgin, cosmetic improvements by me
In the test_slot_timelines test module, we were abusing passing NULL
values which was received as zeroes in x86, but this breaks in ARM
(buildfarm member hamster) by crashing instead. Fix the breakage by
marking these functions as STRICT; the InvalidXid value that was
previously implicit in NULL values (on x86 at least) can now be passed
as 0. Failing to follow the fmgr protocol to check for NULLs beforehand
was causing ARM to fail, as evidenced by segmentation faults in
buildfarm member hamster.
In order to use the new functionality in the test script, use COALESCE
in the right spot to avoid forwarding NULL values.
This was diagnosed from the hamster crash by Craig Ringer, who also
proposed a different patch (checking for NULL values explicitely in the
C function code, and keeping the non-strictness in the C functions).
I decided to go with this approach instead.
Our actual convention, contrary to what I said in 59a2111b23, is not to
quote type names, as evidenced by unquoted use of format_type_be()
result value in error messages. Remove quotes from recently tweaked
messages accordingly.
Per note from Tom Lane
Commit acdf2a8b added a test case involving minus zero as a box endpoint.
This is not very portable, as evidenced by the several older buildfarm
members that are failing on the test because they print minus zero as
just "0". If there were any significant reason to test this behavior,
we could consider carrying a separate expected-file; but it doesn't look
to me like there's adequate justification to accept such a maintenance
burden. Just change the test to use plain zero, instead.
When getParamDescriptions was changed to handle out-of-memory better
by cribbing error recovery logic from getRowDescriptions/getAnotherTuple,
somebody omitted to copy the stanza about checking for excess data in
the message. But you need to do that, since continue'ing out of the
switch in pqParseInput3 means no such check gets applied there anymore.
Noted while looking at Michael Paquier's patch that made yet another
copy of this advance_and_error logic.
(This whole business desperately needs refactoring, because I sure don't
want to see a dozen copies of this code, but that's where we seem to be
headed. What's more, the "suspend parsing on EOF return" convention is a
holdover from protocol 2 and shouldn't exist at all in protocol 3, because
we don't process partial messages anymore. But for now, just fix the
obvious bug.)
Also, fix some wrong/missing comments about what the API spec is
for these three functions.
This doesn't seem worthy of back-patching, even though it's a bug;
the case shouldn't ever arise in the field.
Module provides new access method. It is actually a simple Bloom filter
implemented as pgsql's index. It could give some benefits on search
with large number of columns.
Module is a single way to test generic WAL interface committed earlier.
Author: Teodor Sigaev, Alexander Korotkov
Reviewers: Aleksander Alekseev, Michael Paquier, Jim Nasby
This interface is designed to give an access to WAL for extensions which
could implement new access method, for example. Previously it was
impossible because restoring from custom WAL would need to access system
catalog to find a redo custom function. This patch suggests generic way
to describe changes on page with standart layout.
Bump XLOG_PAGE_MAGIC because of new record type.
Author: Alexander Korotkov with a help of Petr Jelinek, Markus Nullmeier and
minor editorization by my
Reviewers: Petr Jelinek, Alvaro Herrera, Teodor Sigaev, Jim Nasby,
Michael Paquier
I should have remembered that we can't use INT64_MODIFIER with sscanf():
configure chooses that to work with snprintf(), but it might be for our
src/port/snprintf.c implementation and so not compatible with the
platform's sscanf(). This appears to be the explanation for buildfarm
member frogmouth's continuing unhappiness with the tzcode update.
Fortunately, in all of the places where zic is attempting to read into
an int64 variable, it's reading a year which certainly will fit just fine
into an int. So make it read into an int with %d, and then cast or copy
as necessary.
Previously this test was relying too much on WAL replay to occur in the
exact configured interval, which was unreliable on slow or overly busy
servers. Use a custom loop instead of poll_query_until, which is
hopefully more reliable.
Per continued failures on buildfarm member hamster (which is probably
the only one running this test suite)
Author: Michaël Paquier
Previously, the planner would reject an index-only scan if any restriction
clause for its table used a column not available from the index, even
if that restriction clause would later be dropped from the plan entirely
because it's implied by the index's predicate. This is a fairly common
situation for partial indexes because predicates using columns not included
in the index are often the most useful kind of predicate, and we have to
duplicate (or at least imply) the predicate in the WHERE clause in order
to get the index to be considered at all. So index-only scans were
essentially unavailable with such partial indexes.
To fix, we have to do detection of implied-by-predicate clauses much
earlier in the planner. This patch puts it in check_index_predicates
(nee check_partial_indexes), meaning it gets done for every partial index,
whereas we previously only considered this issue at createplan time,
so that the work was only done for an index actually selected for use.
That could result in a noticeable planning slowdown for queries against
tables with many partial indexes. However, testing suggested that there
isn't really a significant cost, especially not with reasonable numbers
of partial indexes. We do get a small additional benefit, which is that
cost_index is more accurate since it correctly discounts the evaluation
cost of clauses that will be removed. We can also avoid considering such
clauses as potential indexquals, which saves useless matching cycles in
the case where the predicate columns aren't in the index, and prevents
generating bogus plans that double-count the clause's selectivity when
the columns are in the index.
Tomas Vondra and Kyotaro Horiguchi, reviewed by Kevin Grittner and
Konstantin Knizhnik, and whacked around a little by me
Robert Haas
Tom Lane
Magnus Hagander
Peter Eisentraut
Alvaro Herrera
Teodor Sigaev
Dean Rasheed
Simon Riggs
Stephen Frost
Noah Misch