While beneficial, both for throughput and average/worst case latency, in
a significant number of workloads, there are other workloads in which
backend_flush_after can cause significant performance regressions in
comparison to < 9.6 releases. The regression is most likely when the hot
data set is bigger than shared buffers, but significantly smaller than
the operating system's page cache.
I personally think that the benefit of enabling backend flush control is
considerably bigger than the potential downsides, but a fair argument
can be made that not regressing is more important than improving
performance/latency. As the latter is the consensus, change the default
to 0.
The other settings introduced in 428b1d6b2 do not have the same
potential for regressions, so leave them enabled.
Benchmarks leading up to changing the default have been performed by
Mithun Cy, Ashutosh Sharma and Robert Haas.
Discussion: CAD__OuhPmc6XH=wYRm_+Q657yQE88DakN4=Ybh2oveFasHkoeA@mail.gmail.com
Commit b12fd41c6 added a "reltarget_has_non_vars" field to RelOptInfo,
but failed to maintain it accurately. Since its only purpose was to skip
calls to has_parallel_hazard() in the simple case where a rel's targetlist
is all Vars, and that call is really pretty cheap in that case anyway, it
seems like this is just a case of premature optimization. Let's drop the
flag and do the calls unconditionally until it's proven that we need more
smarts here.
As noted by Andres Freund, we'd accumulated quite a few similar functions
in clauses.c that examine all functions in an expression tree to see if
they satisfy some boolean test. Reduce the duplication by inventing a
function check_functions_in_node() that applies a simple callback function
to each SQL function OID appearing in a given expression node. This also
fixes some arguable oversights; for example, contain_mutable_functions()
did not check aggregate or window functions for mutability. I doubt that
that represents a live bug at the moment, because we don't really consider
mutability for aggregates; but it might someday be one.
I chose to put check_functions_in_node() in nodeFuncs.c because it seemed
like other modules might wish to use it in future. That in turn forced
moving set_opfuncid() et al into nodeFuncs.c, as the alternative was for
nodeFuncs.c to depend on optimizer/setrefs.c which didn't seem very clean.
In passing, teach contain_leaked_vars_walker() about a few more expression
node types it can safely look through, and improve the rather messy and
undercommented code in has_parallel_hazard_walker().
Discussion: <20160527185853.ziol2os2zskahl7v@alap3.anarazel.de>
Since indexes are created without valid LSNs, an index created
while a snapshot older than old_snapshot_threshold existed could
cause queries to return incorrect results when those old snapshots
were used, if any relevant rows had been subject to early pruning
before the index was built. Prevent usage of a newly created index
until all such snapshots are released, for relations where this can
happen.
Questions about the interaction of "snapshot too old" with index
creation were initially raised by Andres Freund.
Reviewed by Robert Haas.
Transmit the leader's temp-namespace state to workers. This is important
because without it, the workers do not really have the same search path
as the leader. For example, there is no good reason (and no extant code
either) to prevent a worker from executing a temp function that the
leader created previously; but as things stood it would fail to find the
temp function, and then either fail or execute the wrong function entirely.
We still prohibit a worker from creating a temp namespace on its own.
In effect, a worker can only see the session's temp namespace if the leader
had created it before starting the worker, which seems like the right
semantics.
Also, transmit the leader's BackendId to workers, and arrange for workers
to use that when determining the physical file path of a temp relation
belonging to their session. While the original intent was to prevent such
accesses entirely, there were a number of holes in that, notably in places
like dbsize.c which assume they can safely access temp rels of other
sessions anyway. We might as well get this right, as a small down payment
on someday allowing workers to access the leader's temp tables. (With
this change, directly using "MyBackendId" as a relation or buffer backend
ID is deprecated; you should use BackendIdForTempRelations() instead.
I left a couple of such uses alone though, as they're not going to be
reachable in parallel workers until we do something about localbuf.c.)
Move the thou-shalt-not-access-thy-leader's-temp-tables prohibition down
into localbuf.c, which is where it actually matters, instead of having it
in relation_open(). This amounts to recognizing that access to temp
tables' catalog entries is perfectly safe in a worker, it's only the data
in local buffers that is problematic.
Having done all that, we can get rid of the test in has_parallel_hazard()
that says that use of a temp table's rowtype is unsafe in parallel workers.
That test was unduly expensive, and if we really did need such a
prohibition, that was not even close to being a bulletproof guard for it.
(For example, any user-defined function executed in a parallel worker
might have attempted such access.)
Such paths are unsafe. To make it cheaper to detect when this case
applies, track whether a relation's default PathTarget contains any
non-Vars. In most cases, the answer will be no, which enables us to
determine cheaply that the target list for a proposed path is
parallel-safe. However, subquery pull-up can create cases that
require us to inspect the target list more carefully.
Amit Kapila, reviewed by me.
Document these as "nearest integer >= argument" and "nearest integer <=
argument", which will hopefully be less confusing than the old formulation.
New wording is from Matlab via Dean Rasheed.
I changed the pg_description entries as well as the SGML docs. In the
back branches, this will only affect installations initdb'd in the future,
but it should be harmless otherwise.
Discussion: <CAEZATCW3yzJo-NMSiQs5jXNFbTsCEftZS-Og8=FvFdiU+kYuSA@mail.gmail.com>
Fix a couple of overlooked uses of "degree" terminology. Make the parallel
worker count selection logic in create_plain_partial_paths more robust (in
particular, it failed with max_parallel_workers_per_gather set to zero).
This terminology provoked widespread complaints. So, instead, rename
the GUC max_parallel_degree to max_parallel_workers_per_gather
(leaving room for a possible future GUC max_parallel_workers that acts
as a system-wide limit), and rename the parallel_degree reloption to
parallel_workers. Rename structure members to match.
These changes create a dump/restore hazard for users of PostgreSQL
9.6beta1 who have set the reloption (or applied the GUC using ALTER
USER or ALTER DATABASE).
To achieve this, ANALYZE the data table before querying it, as suggested
by Tom Lane. On my system, this enables the test to pass with 128 kB of
work_mem (a value with which other tests fail -- so it seems good
enough).
Reported by Michaël Paquier.
In VPATH builds, the build directory was not being searched for files in
GETTEXT_FILES, leading to failure to construct the .pot files. This has
bit me all along, but never hard enough to get it fixed; I suppose not a
lot of people uses VPATH and NLS-enabled builds, and those that do,
don't do "make update-po" often.
This is a longstanding problem, so backpatch all the way back.
This commit reverts 137805f89 as well as the associated commits 015e88942,
5306df283, and 68d704edb. We found multiple bugs in this feature, and
there was concern about possible planner slowdown (though to be fair,
exhibiting a very large slowdown proved difficult). The way forward
requires a considerable rewrite, which may or may not be possible to
accomplish in time for beta2. In my judgment reviewing the rewrite will
be easier to accomplish starting from a clean slate, so let's temporarily
revert what's there now. This also leaves us in a safe state if it turns
out to be necessary to postpone the rewrite to the next development cycle.
Discussion: <20160429102531.GA13701@huehner.biz>
crosstabview.c was not added to nls.mk when it was added. Also remove
redundant gettext markers, since psql_error() is already registered as a
gettext keyword.
dumpAccessMethod() didn't get the memo that we now have a bitfield for
the components which should be dumped instead of a simple boolean.
Correct that by checking if the relevant bit is set for each component
being dumped out (and not dumping it out if it isn't set).
This corrects an issue where CREATE ACCESS METHOD commands were being
included in non-binary-upgrades when an extension included an access
method (as the bloom extensions does).
Also add a regression test to make sure that we only dump out the
ACCESS METHOD commands, when they are part of an extension, when doing
a binary upgrade.
Pointed out by Thom Brown.
In commit 5c3c3cd0a3046339597a03bc708cb5530dc07059, the new tests were
apparently just dumped into the first convenient file. Move them to a
separate file dedicated to testing that functionality and leave the
plpython_test test to test basic functionality, as it did before.
If we ANALYZE only selected columns of a table, we should not postpone
auto-analyze because of that; other columns may well still need stats
updates. As committed, the counter is left alone if a column list is
given, whether or not it includes all analyzable columns of the table.
Per complaint from Tomasz Ostrowski.
It's been like this a long time, so back-patch to all supported branches.
Report: <ef99c1bd-ff60-5f32-2733-c7b504eb960c@ato.waw.pl>
If a Gather node has read as many tuples as it needs (for example, due
to Limit) it may detach the queue connecting it to the worker before
reading all of the worker's tuples. Rather than let the worker
continue to generate and send all of the results, have it stop after
sending the next tuple.
More could be done here to stop the worker even quicker, but this is
about as well as we can hope to do for 9.6.
This is in response to a problem report from Andreas Seltenreich.
Commit 44339b892a04e94bbb472235882dc6f7023bdc65 should be actually be
sufficient to fix that example even without this change, but it seems
better to do this, too, since we might otherwise waste quite a large
amount of effort in one or more workers.
Discussion: CAA4eK1KOKGqmz9bGu+Z42qhRwMbm4R5rfnqsLCNqFs9j14jzEA@mail.gmail.com
Amit Kapila
Prior to this patch, it was occasionally possible, after shm_mq_sendv
had previously returned SHM_MQ_DETACHED, for a later shm_mq_sendv
operation to fail an assertion instead of just again returning
SHM_MQ_ATTACHED. From the shm_mq code's point of view, it was
expecting to be called again with the same arguments, since the
previous operation had only partially completed. However, a caller
who isn't using non-blocking mode won't be prepared to repeat the call
with the same arguments, and this code shouldn't expect that they
will. Repair in such a way that we'll be OK whether the next call
uses the same arguments or not.
Found by Andreas Seltenreich. Analysis and sketch of fix by Amit
Kapila. Patch by me, reviewed by Amit Kapila.
For historical reasons, copyFile and rewriteVisibilityMap took a force
argument which was always passed as true, meaning that any existing
file should be overwritten. However, it seems much safer to instead
fail if a file we need to write already exists.
While we're at it, remove the "force" argument altogether, since it was
never passed as anything other than true (and now we would never pass
it as anything other than false, if we kept it).
Noted by Andres Freund during post-commit review of the patch that added
rewriteVisibilityMap, commit 7087166a88fe0c04fc6636d0d6d6bea1737fc1fb,
but this also changes the behavior when copying files without rewriting
them.
Patch by Masahiko Sawada.
In the old logic, if read() were to return an error, we'd silently stop
rewriting the visibility map at that point in the file. That's safe,
but reporting the error is better, so do that instead.
Report by Andres Freund. Patch by Masahiko Sawada, with one correction
by me.
Fix still another bug in commit 35fcb1b3d: it failed to fully initialize
the SortSupport states it introduced to allow the executor to re-check
ORDER BY expressions containing distance operators. That led to a null
pointer dereference if the sortsupport code tried to use ssup_cxt. The
problem only manifests in narrow cases, explaining the lack of previous
field reports. It requires a GiST-indexable distance operator that lacks
SortSupport and is on a pass-by-ref data type, which among core+contrib
seems to be only btree_gist's interval opclass; and it requires the scan
to be done as an IndexScan not an IndexOnlyScan, which explains how
btree_gist's regression test didn't catch it. Per bug #14134 from
Jihyun Yu.
Peter Geoghegan
Report: <20160511154904.2603.43889@wrigleys.postgresql.org>
It'd be good for "(x AND y) AND z" to produce a three-child AND node
whether or not operator_precedence_warning is on, but that failed to
happen when it's on because makeAndExpr() didn't look through the added
AEXPR_PAREN node. This has no effect on generated plans because prepqual.c
would flatten the AND nest anyway; but it does affect the number of parens
printed in ruleutils.c, for example. I'd already fixed some similar
hazards in parse_expr.c in commit abb164655, but didn't think to search
gram.y for problems of this ilk. Per gripe from Jean-Pierre Pelletier.
Report: <fa0535ec6d6428cfec40c7e8a6d11156@mail.gmail.com>
This attempts to buy back some of whatever performance we lost from fixing
bug #14174 by inlining the initial checks in MakeExpandedObjectReadOnly()
into the callers. We can do that in a macro without creating multiple-
evaluation hazards, so it's pretty much free notationally; and the amount
of code added to callers should be minimal as well. (Testing a value can't
take many more instructions than passing it to a subroutine.)
Might as well inline DatumIsReadWriteExpandedObject() while we're at it.
This is an ABI break for callers, so it doesn't seem safe to put into 9.5,
but I see no reason not to do it in HEAD.
Further thought about bug #14174 motivated me to try the case of a
R/W datum being returned from a VALUES list, and sure enough it was
broken. Fix that.
Also add a regression test case exercising the same scenario for
FunctionScan. That's not broken right now, because the function's
result will get shoved into a tuplestore between generation and use;
but it could easily become broken whenever we get around to optimizing
FunctionScan better.
There don't seem to be any other places where we put the result of
expression evaluation into a virtual tuple slot that could then be
the source for Vars of further expression evaluation, so I think
this is the end of this bug.
If a plan node output expression returns an "expanded" datum, and that
output column is referenced in more than one place in upper-level plan
nodes, we need to ensure that what is returned is a read-only reference
not a read/write reference. Otherwise one of the referencing sites could
scribble on or even delete the expanded datum before we have evaluated the
others. Commit 1dc5ebc9077ab742, which introduced this feature, supposed
that it'd be sufficient to make SubqueryScan nodes force their output
columns to read-only state. The folly of that was revealed by bug #14174
from Andrew Gierth, and really should have been immediately obvious
considering that the planner will happily optimize SubqueryScan nodes
out of the plan without any regard for this issue.
The safest fix seems to be to make ExecProject() force its results into
read-only state; that will cover every case where a plan node returns
expression results. Actually we can delegate this to ExecTargetList()
since we can recursively assume that plain Vars will not reference
read-write datums. That should keep the extra overhead down to something
minimal. We no longer need ExecMakeSlotContentsReadOnly(), which was
introduced only in support of the idea that just a few plan node types
would need to do this.
In the future it would be nice to have the planner account for this problem
and inject force-to-read-only expression evaluation nodes into only the
places where there's a risk of multiple evaluation. That's not a suitable
solution for 9.5 or even 9.6 at this point, though.
Report: <20160603124628.9932.41279@wrigleys.postgresql.org>
The partial paths that get modified may already have been used as
part of a GatherPath which appears in the path list, so modifying
them is not a good idea at this stage - especially because this
code has no check that the PathTarget is in fact parallel-safe.
When partial aggregation is being performed, this is actually
harmless because we'll end up replacing the pathtargets here with
the correct ones within create_grouping_paths(). But if we've got
a query tree containing only scan/join operations then this can
result in incorrectly pushing down parallel-restricted target
list entries. If those are, for example, references to subqueries,
that can crash the server; but it's wrong in any event.
Amit Kapila
The "snapshot too old" condition was not being recognized when
using a copied snapshot, since the original timestamp and lsn were
not being passed along. Noticed when testing the combination of
"snapshot too old" with parallel query execution.
Adopt the same solution as in commit aa90e148ca70a235, but this time
let's put the ugliness inside the write_stderr() macro, instead of
expecting each call site to deal with it. Back-port that decision
into psql/common.c where I got the macro from in the first place.
Per gripe from Peter Eisentraut.
Mostly these are just comments but there are a few in documentation
and a handful in code and tests. Hopefully this doesn't cause too much
unnecessary pain for backpatching. I relented from some of the most
common like "thru" for that reason. The rest don't seem numerous
enough to cause problems.
Thanks to Kevin Lyda's tool https://pypi.python.org/pypi/misspellings
Per post-commit review comments from Andres Freund, improve variable
names, comments, and in one place, slightly improve the code structure.
Masahiko Sawada
Use MAXALIGN size/alignment to guarantee that later uses of memory are
aligned correctly. E.g. epoll_event might need 8 byte alignment on some
platforms, but earlier allocations like WaitEventSet and WaitEvent might
not sized to guarantee that when purely using sizeof().
Found by myself while testing on an Sun Ultra 5 (Sparc IIi) with some
editorializing by Andres Freund.
In passing fix a couple typos in the area
Formerly, Unix builds of pg_dump/pg_restore would trap SIGINT and similar
signals and set a flag that was tested in various data-transfer loops.
This was prone to errors of omission (cf commit 3c8aa6654); and even if
the client-side response was prompt, we did nothing that would cause
long-running SQL commands (e.g. CREATE INDEX) to terminate early.
Also, the master process would effectively do nothing at all upon receipt
of SIGINT; the only reason it seemed to work was that in typical scenarios
the signal would also be delivered to the child processes. We should
support termination when a signal is delivered only to the master process,
though.
Windows builds had no console interrupt handler, so they would just fall
over immediately at control-C, again leaving long-running SQL commands to
finish unmolested.
To fix, remove the flag-checking approach altogether. Instead, allow the
Unix signal handler to send a cancel request directly and then exit(1).
In the master process, also have it forward the signal to the children.
On Windows, add a console interrupt handler that behaves approximately
the same. The main difference is that a single execution of the Windows
handler can send all the cancel requests since all the info is available
in one process, whereas on Unix each process sends a cancel only for its
own database connection.
In passing, fix an old problem that DisconnectDatabase tends to send a
cancel request before exiting a parallel worker, even if nothing went
wrong. This is at least a waste of cycles, and could lead to unexpected
log messages, or maybe even data loss if it happened in pg_restore (though
in the current code the problem seems to affect only pg_dump). The cause
was that after a COPY step, pg_dump was leaving libpq in PGASYNC_BUSY
state, causing PQtransactionStatus() to report PQTRANS_ACTIVE. That's
normally harmless because the next PQexec() will silently clear the
PGASYNC_BUSY state; but in a parallel worker we might exit without any
additional SQL commands after a COPY step. So add an extra PQgetResult()
call after a COPY to allow libpq to return to PGASYNC_IDLE state.
This is a bug fix, IMO, so back-patch to 9.3 where parallel dump/restore
were introduced.
Thanks to Kyotaro Horiguchi for Windows testing and code suggestions.
Original-Patch: <7005.1464657274@sss.pgh.pa.us>
Discussion: <20160602.174941.256342236.horiguchi.kyotaro@lab.ntt.co.jp>
