This patch makes two closely related sets of changes:
1. For UPDATE, the subplan of the ModifyTable node now only delivers
the new values of the changed columns (i.e., the expressions computed
in the query's SET clause) plus row identity information such as CTID.
ModifyTable must re-fetch the original tuple to merge in the old
values of any unchanged columns. The core advantage of this is that
the changed columns are uniform across all tables of an inherited or
partitioned target relation, whereas the other columns might not be.
A secondary advantage, when the UPDATE involves joins, is that less
data needs to pass through the plan tree. The disadvantage of course
is an extra fetch of each tuple to be updated. However, that seems to
be very nearly free in context; even worst-case tests don't show it to
add more than a couple percent to the total query cost. At some point
it might be interesting to combine the re-fetch with the tuple access
that ModifyTable must do anyway to mark the old tuple dead; but that
would require a good deal of refactoring and it seems it wouldn't buy
all that much, so this patch doesn't attempt it.
2. For inherited UPDATE/DELETE, instead of generating a separate
subplan for each target relation, we now generate a single subplan
that is just exactly like a SELECT's plan, then stick ModifyTable
on top of that. To let ModifyTable know which target relation a
given incoming row refers to, a tableoid junk column is added to
the row identity information. This gets rid of the horrid hack
that was inheritance_planner(), eliminating O(N^2) planning cost
and memory consumption in cases where there were many unprunable
target relations.
Point 2 of course requires point 1, so that there is a uniform
definition of the non-junk columns to be returned by the subplan.
We can't insist on uniform definition of the row identity junk
columns however, if we want to keep the ability to have both
plain and foreign tables in a partitioning hierarchy. Since
it wouldn't scale very far to have every child table have its
own row identity column, this patch includes provisions to merge
similar row identity columns into one column of the subplan result.
In particular, we can merge the whole-row Vars typically used as
row identity by FDWs into one column by pretending they are type
RECORD. (It's still okay for the actual composite Datums to be
labeled with the table's rowtype OID, though.)
There is more that can be done to file down residual inefficiencies
in this patch, but it seems to be committable now.
FDW authors should note several API changes:
* The argument list for AddForeignUpdateTargets() has changed, and so
has the method it must use for adding junk columns to the query. Call
add_row_identity_var() instead of manipulating the parse tree directly.
You might want to reconsider exactly what you're adding, too.
* PlanDirectModify() must now work a little harder to find the
ForeignScan plan node; if the foreign table is part of a partitioning
hierarchy then the ForeignScan might not be the direct child of
ModifyTable. See postgres_fdw for sample code.
* To check whether a relation is a target relation, it's no
longer sufficient to compare its relid to root->parse->resultRelation.
Instead, check it against all_result_relids or leaf_result_relids,
as appropriate.
Amit Langote and Tom Lane
Discussion: https://postgr.es/m/CA+HiwqHpHdqdDn48yCEhynnniahH78rwcrv1rEX65-fsZGBOLQ@mail.gmail.com
This implements asynchronous execution, which runs multiple parts of a
non-parallel-aware Append concurrently rather than serially to improve
performance when possible. Currently, the only node type that can be
run concurrently is a ForeignScan that is an immediate child of such an
Append. In the case where such ForeignScans access data on different
remote servers, this would run those ForeignScans concurrently, and
overlap the remote operations to be performed simultaneously, so it'll
improve the performance especially when the operations involve
time-consuming ones such as remote join and remote aggregation.
We may extend this to other node types such as joins or aggregates over
ForeignScans in the future.
This also adds the support for postgres_fdw, which is enabled by the
table-level/server-level option "async_capable". The default is false.
Robert Haas, Kyotaro Horiguchi, Thomas Munro, and myself. This commit
is mostly based on the patch proposed by Robert Haas, but also uses
stuff from the patch proposed by Kyotaro Horiguchi and from the patch
proposed by Thomas Munro. Reviewed by Kyotaro Horiguchi, Konstantin
Knizhnik, Andrey Lepikhov, Movead Li, Thomas Munro, Justin Pryzby, and
others.
Discussion: https://postgr.es/m/CA%2BTgmoaXQEt4tZ03FtQhnzeDEMzBck%2BLrni0UWHVVgOTnA6C1w%40mail.gmail.com
Discussion: https://postgr.es/m/CA%2BhUKGLBRyu0rHrDCMC4%3DRn3252gogyp1SjOgG8SEKKZv%3DFwfQ%40mail.gmail.com
Discussion: https://postgr.es/m/20200228.170650.667613673625155850.horikyota.ntt%40gmail.com
Commit 3048898e dropped -ING from some wait event names that correspond
to barrier phases. Update the phases' names to match.
While we're here making cosmetic changes, also rename "DONE" to "FREE".
That pairs better with "ALLOCATE", and describes the activity that
actually happens in that phase (as we do for the other phases) rather
than describing a state. The distinction is clearer after bugfix commit
3b8981b6 split the phase into two. As for the growth barriers, rename
their "ALLOCATE" phase to "REALLOCATE", which is probably a better
description of what happens then. Also improve the comments about
the phases a bit.
Discussion: https://postgr.es/m/CA%2BhUKG%2BMDpwF2Eo2LAvzd%3DpOh81wUTsrwU1uAwR-v6OGBB6%2B7g%40mail.gmail.com
With very unlucky timing and parallel_leader_participation off, PHJ
could attempt to access per-batch state just as it was being freed.
There was code intended to prevent that by checking for a cleared
pointer, but it was buggy.
Fix, by introducing an extra barrier phase. The new phase
PHJ_BUILD_RUNNING means that it's safe to access the per-batch state to
find a batch to help with, and PHJ_BUILD_DONE means that it is too late.
The last to detach will free the array of per-batch state as before, but
now it will also atomically advance the phase at the same time, so that
late attachers can avoid the hazard, without the data race. This
mirrors the way per-batch hash tables are freed (see phases
PHJ_BATCH_PROBING and PHJ_BATCH_DONE).
Revealed by a one-off build farm failure, where BarrierAttach() failed a
sanity check assertion, because the memory had been clobbered by
dsa_free().
Back-patch to 11, where the code arrived.
Reported-by: Michael Paquier <michael@paquier.xyz>
Discussion: https://postgr.es/m/20200929061142.GA29096%40paquier.xyz
This adds a new executor node named TID Range Scan. The query planner
will generate paths for TID Range scans when quals are discovered on base
relations which search for ranges on the table's ctid column. These
ranges may be open at either end. For example, WHERE ctid >= '(10,0)';
will return all tuples on page 10 and over.
To support this, two new optional callback functions have been added to
table AM. scan_set_tidrange is used to set the scan range to just the
given range of TIDs. scan_getnextslot_tidrange fetches the next tuple
in the given range.
For AMs were scanning ranges of TIDs would not make sense, these functions
can be set to NULL in the TableAmRoutine. The query planner won't
generate TID Range Scan Paths in that case.
Author: Edmund Horner, David Rowley
Reviewed-by: David Rowley, Tomas Vondra, Tom Lane, Andres Freund, Zhihong Yu
Discussion: https://postgr.es/m/CAMyN-kB-nFTkF=VA_JPwFNo08S0d-Yk0F741S2B7LDmYAi8eyA@mail.gmail.com
If a cross-partition UPDATE violates a constraint on the target partition,
and the columns in the new partition are in different physical order than
in the parent, the error message can reveal columns that the user does not
have SELECT permission on. A similar bug was fixed earlier in commit
804b6b6db4.
The cause of the bug is that the callers of the
ExecBuildSlotValueDescription() function got confused when constructing
the list of modified columns. If the tuple was routed from a parent, we
converted the tuple to the parent's format, but the list of modified
columns was grabbed directly from the child's RTE entry.
ExecUpdateLockMode() had a similar issue. That lead to confusion on which
columns are key columns, leading to wrong tuple lock being taken on tables
referenced by foreign keys, when a row is updated with INSERT ON CONFLICT
UPDATE. A new isolation test is added for that corner case.
With this patch, the ri_RangeTableIndex field is no longer set for
partitions that don't have an entry in the range table. Previously, it was
set to the RTE entry of the parent relation, but that was confusing.
NOTE: This modifies the ResultRelInfo struct, replacing the
ri_PartitionRoot field with ri_RootResultRelInfo. That's a bit risky to
backpatch, because it breaks any extensions accessing the field. The
change that ri_RangeTableIndex is not set for partitions could potentially
break extensions, too. The ResultRelInfos are visible to FDWs at least,
and this patch required small changes to postgres_fdw. Nevertheless, this
seem like the least bad option. I don't think these fields widely used in
extensions; I don't think there are FDWs out there that uses the FDW
"direct update" API, other than postgres_fdw. If there is, you will get a
compilation error, so hopefully it is caught quickly.
Backpatch to 11, where support for both cross-partition UPDATEs, and unique
indexes on partitioned tables, were added.
Reviewed-by: Amit Langote
Security: CVE-2021-3393
SPI_execute_with_receiver and SPI_cursor_parse_open_with_paramlist are
new in v14 (cf. commit 2f48ede08). Before they can get out the door,
let's change their APIs to follow the practice recently established by
SPI_prepare_extended etc: shove all optional arguments into a struct
that callers are supposed to pre-zero. The hope is to allow future
addition of more options without either API breakage or a continuing
proliferation of new SPI entry points. With that in mind, choose
slightly more generic names for them: SPI_execute_extended and
SPI_cursor_parse_open respectively.
Discussion: https://postgr.es/m/CAFj8pRCLPdDAETvR7Po7gC5y_ibkn_-bOzbeJb39WHms01194Q@mail.gmail.com
This patch essentially is cleaning up technical debt left behind
by the original implementation of plpgsql procedures, particularly
commit d92bc83c4. That patch (or more precisely, follow-on patches
fixing its worst bugs) forced us to re-plan CALL and DO statements
each time through, if we're in a non-atomic context. That wasn't
for any fundamental reason, but just because use of a saved plan
requires having a ResourceOwner to hold a reference count for the
plan, and we had no suitable resowner at hand, nor would the
available APIs support using one if we did. While it's not that
expensive to create a "plan" for CALL/DO, the cycles do add up
in repeated executions.
This patch therefore makes the following API changes:
* GetCachedPlan/ReleaseCachedPlan are modified to let the caller
specify which resowner to use to pin the plan, rather than forcing
use of CurrentResourceOwner.
* spi.c gains a "SPI_execute_plan_extended" entry point that lets
callers say which resowner to use to pin the plan. This borrows the
idea of an options struct from the recently added SPI_prepare_extended,
hopefully allowing future options to be added without more API breaks.
This supersedes SPI_execute_plan_with_paramlist (which I've marked
deprecated) as well as SPI_execute_plan_with_receiver (which is new
in v14, so I just took it out altogether).
* I also took the opportunity to remove the crude hack of letting
plpgsql reach into SPI private data structures to mark SPI plans as
"no_snapshot". It's better to treat that as an option of
SPI_prepare_extended.
Now, when running a non-atomic procedure or DO block that contains
any CALL or DO commands, plpgsql creates a ResourceOwner that
will be used to pin the plans of the CALL/DO commands. (In an
atomic context, we just use CurrentResourceOwner, as before.)
Having done this, we can just save CALL/DO plans normally,
whether or not they are used across transaction boundaries.
This seems to be good for something like 2X speedup of a CALL
of a trivial procedure with a few simple argument expressions.
By restricting the creation of an extra ResourceOwner like this,
there's essentially zero penalty in cases that can't benefit.
Pavel Stehule, with some further hacking by me
Discussion: https://postgr.es/m/CAFj8pRCLPdDAETvR7Po7gC5y_ibkn_-bOzbeJb39WHms01194Q@mail.gmail.com
Add an executor aminsert() hint mechanism that informs index AMs that
the incoming index tuple (the tuple that accompanies the hint) is not
being inserted by execution of an SQL statement that logically modifies
any of the index's key columns.
The hint is received by indexes when an UPDATE takes place that does not
apply an optimization like heapam's HOT (though only for indexes where
all key columns are logically unchanged). Any index tuple that receives
the hint on insert is expected to be a duplicate of at least one
existing older version that is needed for the same logical row. Related
versions will typically be stored on the same index page, at least
within index AMs that apply the hint.
Recognizing the difference between MVCC version churn duplicates and
true logical row duplicates at the index AM level can help with cleanup
of garbage index tuples. Cleanup can intelligently target tuples that
are likely to be garbage, without wasting too many cycles on less
promising tuples/pages (index pages with little or no version churn).
This is infrastructure for an upcoming commit that will teach nbtree to
perform bottom-up index deletion. No index AM actually applies the hint
just yet.
Author: Peter Geoghegan <pg@bowt.ie>
Reviewed-By: Victor Yegorov <vyegorov@gmail.com>
Discussion: https://postgr.es/m/CAH2-Wz=CEKFa74EScx_hFVshCOn6AA5T-ajFASTdzipdkLTNQQ@mail.gmail.com
This patch essentially allows gram.y to implement a family of related
syntax trees, rather than necessarily always parsing a list of SQL
statements. raw_parser() gains a new argument, enum RawParseMode,
to say what to do. As proof of concept, add a mode that just parses
a TypeName without any other decoration, and use that to greatly
simplify typeStringToTypeName().
In addition, invent a new SPI entry point SPI_prepare_extended() to
allow SPI users (particularly plpgsql) to get at this new functionality.
In hopes of making this the last variant of SPI_prepare(), set up its
additional arguments as a struct rather than direct arguments, and
promise that future additions to the struct can default to zero.
SPI_prepare_cursor() and SPI_prepare_params() can perhaps go away at
some point.
Discussion: https://postgr.es/m/4165684.1607707277@sss.pgh.pa.us
This patch generalizes the subscripting infrastructure so that any
data type can be subscripted, if it provides a handler function to
define what that means. Traditional variable-length (varlena) arrays
all use array_subscript_handler(), while the existing fixed-length
types that support subscripting use raw_array_subscript_handler().
It's expected that other types that want to use subscripting notation
will define their own handlers. (This patch provides no such new
features, though; it only lays the foundation for them.)
To do this, move the parser's semantic processing of subscripts
(including coercion to whatever data type is required) into a
method callback supplied by the handler. On the execution side,
replace the ExecEvalSubscriptingRef* layer of functions with direct
calls to callback-supplied execution routines. (Thus, essentially
no new run-time overhead should be caused by this patch. Indeed,
there is room to remove some overhead by supplying specialized
execution routines. This patch does a little bit in that line,
but more could be done.)
Additional work is required here and there to remove formerly
hard-wired assumptions about the result type, collation, etc
of a SubscriptingRef expression node; and to remove assumptions
that the subscript values must be integers.
One useful side-effect of this is that we now have a less squishy
mechanism for identifying whether a data type is a "true" array:
instead of wiring in weird rules about typlen, we can look to see
if pg_type.typsubscript == F_ARRAY_SUBSCRIPT_HANDLER. For this
to be bulletproof, we have to forbid user-defined types from using
that handler directly; but there seems no good reason for them to
do so.
This patch also removes assumptions that the number of subscripts
is limited to MAXDIM (6), or indeed has any hard-wired limit.
That limit still applies to types handled by array_subscript_handler
or raw_array_subscript_handler, but to discourage other dependencies
on this constant, I've moved it from c.h to utils/array.h.
Dmitry Dolgov, reviewed at various times by Tom Lane, Arthur Zakirov,
Peter Eisentraut, Pavel Stehule
Discussion: https://postgr.es/m/CA+q6zcVDuGBv=M0FqBYX8DPebS3F_0KQ6OVFobGJPM507_SZ_w@mail.gmail.com
Discussion: https://postgr.es/m/CA+q6zcVovR+XY4mfk-7oNk-rF91gH0PebnNfuUjuuDsyHjOcVA@mail.gmail.com
Previously, ExecInitModifyTable relied on ExecInitJunkFilter,
and thence ExecCleanTypeFromTL, to build the target descriptor from
the query tlist. While we just checked (in ExecCheckPlanOutput)
that the tlist produces compatible output, this is not a great
substitute for the relation's actual tuple descriptor that's
available from the relcache. For one thing, dropped columns will
not be correctly marked attisdropped; it's a bit surprising that
we've gotten away with that this long. But the real reason for
being concerned with this is that using the table's descriptor means
that the slot will have correct attrmissing data, allowing us to
revert the klugy fix of commit ba9f18abd. (This commit undoes
that one's changes in trigger.c, but keeps the new test case.)
Thus we can solve the bogus-trigger-tuple problem with fewer cycles
rather than more.
No back-patch, since this doesn't fix any additional bug, and it
seems somewhat more likely to have unforeseen side effects than
ba9f18abd's narrow fix.
Discussion: https://postgr.es/m/16644-5da7ef98a7ac4545@postgresql.org
Since commit 913bbd88d, check_sql_fn_retval() can either insert type
coercion steps in-line in the Query that produces the SQL function's
results, or generate a new top-level Query to perform the coercions,
if modifying the Query's output in-place wouldn't be safe. However,
it appears that the latter case has never actually worked, because
the code tried to inject the new Query back into the query list it was
passed ... which is not the list that will be used for later processing
when we execute the SQL function "normally" (without inlining it).
So we ended up with no coercion happening at run-time, leading to
wrong results or crashes depending on the datatypes involved.
While the regression tests look like they cover this area well enough,
through a huge bit of bad luck all the test cases that exercise the
separate-Query path were checking either inline-able cases (which
accidentally didn't have the bug) or cases that are no-ops at runtime
(e.g., varchar to text), so that the failure to perform the coercion
wasn't obvious. The fact that the cases that don't work weren't
allowed at all before v13 probably contributed to not noticing the
problem sooner, too.
To fix, get rid of the separate "flat" list of Query nodes and instead
pass the real two-level list that is going to be used later. I chose
to make the same change in check_sql_fn_statements(), although that has
no actual bug, just so that we don't need that data structure at all.
This is an API change, as evidenced by the adjustments needed to
callers outside functions.c. That's a bit scary to be doing in a
released branch, but so far as I can tell from a quick search,
there are no outside callers of these functions (and they are
sufficiently specific to our semantics for SQL-language functions that
it's not apparent why any extension would need to call them). In any
case, v13 already changed the API of check_sql_fn_retval() compared to
prior branches.
Per report from pinker. Back-patch to v13 where this code came in.
Discussion: https://postgr.es/m/1603050466566-0.post@n3.nabble.com
Store the tuple conversion map to convert a tuple from a child table's
format to the root format in a new ri_ChildToRootMap field in
ResultRelInfo. It is initialized if transition tuple capture for FOR
STATEMENT triggers or INSERT tuple routing on a partitioned table is
needed. Previously, ModifyTable kept the maps in the per-subplan
ModifyTableState->mt_per_subplan_tupconv_maps array, or when tuple
routing was used, in
ResultRelInfo->ri_Partitioninfo->pi_PartitionToRootMap. The new field
replaces both of those.
Now that the child-to-root tuple conversion map is always available in
ResultRelInfo (when needed), remove the TransitionCaptureState.tcs_map
field. The callers of Exec*Trigger() functions no longer need to set or
save it, which is much less confusing and bug-prone. Also, as a future
optimization, this will allow us to delay creating the map for a given
result relation until the relation is actually processed during
execution.
Author: Amit Langote
Discussion: https://www.postgresql.org/message-id/CA%2BHiwqHtCWLdK-LO%3DNEsvOdHx%2B7yv4mE_zYK0i3BH7dXb-wxog%40mail.gmail.com
Maintaining 'es_result_relation_info' correctly at all times has become
cumbersome, especially with partitioning where each partition gets its
own result relation info. Having to set and reset it across arbitrary
operations has caused bugs in the past.
This changes all the places that used 'es_result_relation_info', to
receive the currently active ResultRelInfo via function parameters
instead.
Author: Amit Langote
Discussion: https://www.postgresql.org/message-id/CA%2BHiwqGEmiib8FLiHMhKB%2BCH5dRgHSLc5N5wnvc4kym%2BZYpQEQ%40mail.gmail.com
Instead of allocating all the ResultRelInfos upfront in one big array,
allocate them in ExecInitModifyTable(). es_result_relations is now an
array of ResultRelInfo pointers, rather than an array of structs, and it
is indexed by the RT index.
This simplifies things: we get rid of the separate concept of a "result
rel index", and don't need to set it in setrefs.c anymore. This also
allows follow-up optimizations (not included in this commit yet) to skip
initializing ResultRelInfos for target relations that were not needed at
runtime, and removal of the es_result_relation_info pointer.
The EState arrays of regular result rels and root result rels are merged
into one array. Similarly, the resultRelations and rootResultRelations
lists in PlannedStmt are merged into one. It's not actually clear to me
why they were kept separate in the first place, but now that the
es_result_relations array is indexed by RT index, it certainly seems
pointless.
The PlannedStmt->resultRelations list is now only needed for
ExecRelationIsTargetRelation(). One visible effect of this change is that
ExecRelationIsTargetRelation() will now return 'true' also for the
partition root, if a partitioned table is updated. That seems like a good
thing, although the function isn't used in core code, and I don't see any
reason for an FDW to call it on a partition root.
Author: Amit Langote
Discussion: https://www.postgresql.org/message-id/CA%2BHiwqGEmiib8FLiHMhKB%2BCH5dRgHSLc5N5wnvc4kym%2BZYpQEQ%40mail.gmail.com
When commit bd3daddaf introduced AlternativeSubPlans, I had some
ambitions towards allowing the choice of subplan to change during
execution. That has not happened, or even been thought about, in the
ensuing twelve years; so it seems like a failed experiment. So let's
rip that out and resolve the choice of subplan at the end of planning
(in setrefs.c) rather than during executor startup. This has a number
of positive benefits:
* Removal of a few hundred lines of executor code, since
AlternativeSubPlans need no longer be supported there.
* Removal of executor-startup overhead (particularly, initialization
of subplans that won't be used).
* Removal of incidental costs of having a larger plan tree, such as
tree-scanning and copying costs in the plancache; not to mention
setrefs.c's own costs of processing the discarded subplans.
* EXPLAIN no longer has to print a weird (and undocumented)
representation of an AlternativeSubPlan choice; it sees only the
subplan actually used. This should mean less confusion for users.
* Since setrefs.c knows which subexpression of a plan node it's
working on at any instant, it's possible to adjust the estimated
number of executions of the subplan based on that. For example,
we should usually estimate more executions of a qual expression
than a targetlist expression. The implementation used here is
pretty simplistic, because we don't want to expend a lot of cycles
on the issue; but it's better than ignoring the point entirely,
as the executor had to.
That last point might possibly result in shifting the choice
between hashed and non-hashed EXISTS subplans in a few cases,
but in general this patch isn't meant to change planner choices.
Since we're doing the resolution so late, it's really impossible
to change any plan choices outside the AlternativeSubPlan itself.
Patch by me; thanks to David Rowley for review.
Discussion: https://postgr.es/m/1992952.1592785225@sss.pgh.pa.us
Add a GUC that acts as a multiplier on work_mem. It gets applied when
sizing executor node hash tables that were previously size constrained
using work_mem alone.
The new GUC can be used to preferentially give hash-based nodes more
memory than the generic work_mem limit. It is intended to enable admin
tuning of the executor's memory usage. Overall system throughput and
system responsiveness can be improved by giving hash-based executor
nodes more memory (especially over sort-based alternatives, which are
often much less sensitive to being memory constrained).
The default value for hash_mem_multiplier is 1.0, which is also the
minimum valid value. This means that hash-based nodes continue to apply
work_mem in the traditional way by default.
hash_mem_multiplier is generally useful. However, it is being added now
due to concerns about hash aggregate performance stability for users
that upgrade to Postgres 13 (which added disk-based hash aggregation in
commit 1f39bce0). While the old hash aggregate behavior risked
out-of-memory errors, it is nevertheless likely that many users actually
benefited. Hash agg's previous indifference to work_mem during query
execution was not just faster; it also accidentally made aggregation
resilient to grouping estimate problems (at least in cases where this
didn't create destabilizing memory pressure).
hash_mem_multiplier can provide a certain kind of continuity with the
behavior of Postgres 12 hash aggregates in cases where the planner
incorrectly estimates that all groups (plus related allocations) will
fit in work_mem/hash_mem. This seems necessary because hash-based
aggregation is usually much slower when only a small fraction of all
groups can fit. Even when it isn't possible to totally avoid hash
aggregates that spill, giving hash aggregation more memory will reliably
improve performance (the same cannot be said for external sort
operations, which appear to be almost unaffected by memory availability
provided it's at least possible to get a single merge pass).
The PostgreSQL 13 release notes should advise users that increasing
hash_mem_multiplier can help with performance regressions associated
with hash aggregation. That can be taken care of by a later commit.
Author: Peter Geoghegan
Reviewed-By: Álvaro Herrera, Jeff Davis
Discussion: https://postgr.es/m/20200625203629.7m6yvut7eqblgmfo@alap3.anarazel.de
Discussion: https://postgr.es/m/CAH2-WzmD%2Bi1pG6rc1%2BCjc4V6EaFJ_qSuKCCHVnH%3DoruqD-zqow%40mail.gmail.com
Backpatch: 13-, where disk-based hash aggregation was introduced.
Use HyperLogLog to estimate the group cardinality in a spilled
partition. This estimate is used to choose the number of partitions if
we recurse.
The previous behavior was to use the number of tuples in a spilled
partition as the estimate for the number of groups, which lead to
overpartitioning. That could cause the number of batches to be much
higher than expected (with each batch being very small), which made it
harder to interpret EXPLAIN ANALYZE results.
Reviewed-by: Peter Geoghegan
Discussion: https://postgr.es/m/a856635f9284bc36f7a77d02f47bbb6aaf7b59b3.camel@j-davis.com
Backpatch-through: 13
This representation saves 8 bytes per tuple compared to HeapTuple, and
avoids the need to allocate, copy and free on the receiving side.
Gather can emit the returned MinimalTuple directly, but GatherMerge now
needs to make an explicit copy because it buffers multiple tuples at a
time. That should be no worse than before.
Reviewed-by: Soumyadeep Chakraborty <soumyadeep2007@gmail.com>
Discussion: https://postgr.es/m/CA%2BhUKG%2B8T_ggoUTAE-U%3DA%2BOcPc4%3DB0nPPHcSfffuQhvXXjML6w%40mail.gmail.com
Since 1f39bce02, HashAgg nodes have had the ability to spill to disk when
memory consumption exceeds work_mem. That commit added new properties to
EXPLAIN ANALYZE to show the maximum memory usage and disk usage, however,
it didn't quite go as far as showing that information for parallel
workers. Since workers may have experienced something very different from
the main process, we should show this information per worker, as is done
in Sort.
Reviewed-by: Justin Pryzby
Reviewed-by: Jeff Davis
Discussion: https://postgr.es/m/CAApHDvpEKbfZa18mM1TD7qV6PG+w97pwCWq5tVD0dX7e11gRJw@mail.gmail.com
Backpatch-through: 13, where the hashagg spilling code was added.
plpgsql has always executed the query given in a RETURN QUERY command
by opening it as a cursor and then fetching a few rows at a time,
which it turns around and dumps into the function's result tuplestore.
The point of this was to keep from blowing out memory with an oversized
SPITupleTable result (note that while a tuplestore can spill tuples
to disk, SPITupleTable cannot). However, it's rather inefficient, both
because of extra data copying and because of executor entry/exit
overhead. In recent versions, a new performance problem has emerged:
use of a cursor prevents use of a parallel plan for the executed query.
We can improve matters by skipping use of a cursor and having the
executor push result tuples directly into the function's result
tuplestore. However, a moderate amount of new infrastructure is needed
to make that idea work:
* We can use the existing tstoreReceiver.c DestReceiver code to funnel
executor output to the tuplestore, but it has to be extended to support
plpgsql's requirement for possibly applying a tuple conversion map.
* SPI needs to be extended to allow use of a caller-supplied
DestReceiver instead of its usual receiver that puts tuples into
a SPITupleTable. Two new API calls are needed to handle both the
RETURN QUERY and RETURN QUERY EXECUTE cases.
I also felt that I didn't want these new API calls to use the legacy
method of specifying query parameter values with "char" null flags
(the old ' '/'n' convention); rather they should accept ParamListInfo
objects containing the parameter type and value info. This required
a bit of additional new infrastructure since we didn't yet have any
parse analysis callback that would interpret $N parameter symbols
according to type data supplied in a ParamListInfo. There seems to be
no harm in letting makeParamList install that callback by default,
rather than leaving a new ParamListInfo's parserSetup hook as NULL.
(Indeed, as of HEAD, I couldn't find anyplace that was using the
parserSetup field at all; plpgsql was using parserSetupArg for its
own purposes, but parserSetup seemed to be write-only.)
We can actually get plpgsql out of the business of using legacy null
flags altogether, and using ParamListInfo instead of its ad-hoc
PreparedParamsData structure; but this requires inventing one more
SPI API call that can replace SPI_cursor_open_with_args. That seems
worth doing, though.
SPI_execute_with_args and SPI_cursor_open_with_args are now unused
anywhere in the core PG distribution. Perhaps someday we could
deprecate/remove them. But cleaning up the crufty bits of the SPI
API is a task for a different patch.
Per bug #16040 from Jeremy Smith. This is unfortunately too invasive to
consider back-patching. Patch by me; thanks to Hamid Akhtar for review.
Discussion: https://postgr.es/m/16040-eaacad11fecfb198@postgresql.org
Includes some manual cleanup of places that pgindent messed up,
most of which weren't per project style anyway.
Notably, it seems some people didn't absorb the style rules of
commit c9d297751, because there were a bunch of new occurrences
of function calls with a newline just after the left paren, all
with faulty expectations about how the rest of the call would get
indented.
In commit 33e05f89c5, we have added the option to display WAL usage
statistics in Explain and auto_explain. The display format used two spaces
between each field which is inconsistent with Buffer usage statistics which
is using one space between each field. Change the format to make WAL usage
statistics consistent with Buffer usage statistics.
This commit also changed the usage of "full page writes" to
"full page images" for WAL usage statistics to make it consistent with
other parts of code and docs.
Author: Julien Rouhaud, Amit Kapila
Reviewed-by: Justin Pryzby, Kyotaro Horiguchi and Amit Kapila
Discussion: https://postgr.es/m/CAB-hujrP8ZfUkvL5OYETipQwA=e3n7oqHFU=4ZLxWS_Cza3kQQ@mail.gmail.com
Before discarding the old hash table in ExecReScanHashJoin, capture
its statistics, ensuring that we report the maximum hashtable size
across repeated rescans of the hash input relation. We can repurpose
the existing code for reporting hashtable size in parallel workers
to help with this, making the patch pretty small. This also ensures
that if rescans happen within parallel workers, we get the correct
maximums across all instances.
Konstantin Knizhnik and Tom Lane, per diagnosis by Thomas Munro
of a trouble report from Alvaro Herrera.
Discussion: https://postgr.es/m/20200323165059.GA24950@alvherre.pgsql
If the memory context's maxBlockSize is too big, a single block
allocation can suddenly exceed work_mem. For Hash Aggregation, this
can mean spilling to disk too early or reporting a confusing memory
usage number for EXPLAN ANALYZE.
Introduce CreateWorkExprContext(), which is like CreateExprContext(),
except that it creates the AllocSet with a maxBlockSize that is
reasonable in proportion to work_mem.
Right now, CreateWorkExprContext() is only used by Hash Aggregation,
but it may be generally useful in the future.
Discussion: https://postgr.es/m/412a3fbf306f84d8d78c4009e11791867e62b87c.camel@j-davis.com
Incremental Sort is an optimized variant of multikey sort for cases when
the input is already sorted by a prefix of the requested sort keys. For
example when the relation is already sorted by (key1, key2) and we need
to sort it by (key1, key2, key3) we can simply split the input rows into
groups having equal values in (key1, key2), and only sort/compare the
remaining column key3.
This has a number of benefits:
- Reduced memory consumption, because only a single group (determined by
values in the sorted prefix) needs to be kept in memory. This may also
eliminate the need to spill to disk.
- Lower startup cost, because Incremental Sort produce results after each
prefix group, which is beneficial for plans where startup cost matters
(like for example queries with LIMIT clause).
We consider both Sort and Incremental Sort, and decide based on costing.
The implemented algorithm operates in two different modes:
- Fetching a minimum number of tuples without check of equality on the
prefix keys, and sorting on all columns when safe.
- Fetching all tuples for a single prefix group and then sorting by
comparing only the remaining (non-prefix) keys.
We always start in the first mode, and employ a heuristic to switch into
the second mode if we believe it's beneficial - the goal is to minimize
the number of unnecessary comparions while keeping memory consumption
below work_mem.
This is a very old patch series. The idea was originally proposed by
Alexander Korotkov back in 2013, and then revived in 2017. In 2018 the
patch was taken over by James Coleman, who wrote and rewrote most of the
current code.
There were many reviewers/contributors since 2013 - I've done my best to
pick the most active ones, and listed them in this commit message.
Author: James Coleman, Alexander Korotkov
Reviewed-by: Tomas Vondra, Andreas Karlsson, Marti Raudsepp, Peter Geoghegan, Robert Haas, Thomas Munro, Antonin Houska, Andres Freund, Alexander Kuzmenkov
Discussion: https://postgr.es/m/CAPpHfdscOX5an71nHd8WSUH6GNOCf=V7wgDaTXdDd9=goN-gfA@mail.gmail.com
Discussion: https://postgr.es/m/CAPpHfds1waRZ=NOmueYq0sx1ZSCnt+5QJvizT8ndT2=etZEeAQ@mail.gmail.com
This allows gathering the WAL generation statistics for each statement
execution. The three statistics that we collect are the number of WAL
records, the number of full page writes and the amount of WAL bytes
generated.
This helps the users who have write-intensive workload to see the impact
of I/O due to WAL. This further enables us to see approximately what
percentage of overall WAL is due to full page writes.
In the future, we can extend this functionality to allow us to compute the
the exact amount of WAL data due to full page writes.
This patch in itself is just an infrastructure to compute WAL usage data.
The upcoming patches will expose this data via explain, auto_explain,
pg_stat_statements and verbose (auto)vacuum output.
Author: Kirill Bychik, Julien Rouhaud
Reviewed-by: Dilip Kumar, Fujii Masao and Amit Kapila
Discussion: https://postgr.es/m/CAB-hujrP8ZfUkvL5OYETipQwA=e3n7oqHFU=4ZLxWS_Cza3kQQ@mail.gmail.com
Don't try to be precise about it, just use a constant 16 bytes of
chunk overhead. Being smarter would require knowing the memory context
where the chunk will be allocated, which is not known by all callers.
Discussion: https://postgr.es/m/20200325220936.il3ni2fj2j2b45y5@alap3.anarazel.de
Previously pg_stat_statements calculated the difference of buffer counters
by its own code even while BufferUsageAccumDiff() had the same code.
This commit expose BufferUsageAccumDiff() and makes pg_stat_statements
use it for the calculation, in order to simply the code.
This change also would be useful for the upcoming patch for the planning
counters in pg_stat_statements because the patch will add one more code
for the calculation of difference of buffer counters and that can easily be
done by using BufferUsageAccumDiff().
Author: Julien Rouhaud
Reviewed-by: Fujii Masao
Discussion: https://postgr.es/m/bdfee4e0-a304-2498-8da5-3cb52c0a193e@oss.nttdata.com
This reverts the parts of commit 17a28b0364
that changed ereport's auxiliary functions from returning dummy integer
values to returning void. It turns out that a minority of compilers
complain (not entirely unreasonably) about constructs such as
(condition) ? errdetail(...) : 0
if errdetail() returns void rather than int. We could update those
call sites to say "(void) 0" perhaps, but the expectation for this
patch set was that ereport callers would not have to change anything.
And this aspect of the patch set was already the most invasive and
least compelling part of it, so let's just drop it.
Per buildfarm.
Discussion: https://postgr.es/m/CA+fd4k6N8EjNvZpM8nme+y+05mz-SM8Z_BgkixzkA34R+ej0Kw@mail.gmail.com
Change all the auxiliary error-reporting routines to return void,
now that we no longer need to pretend they are passing something
useful to errfinish(). While this probably doesn't save anything
significant at the machine-code level, it allows detection of some
additional types of mistakes.
Pass the error location details (__FILE__, __LINE__, PG_FUNCNAME_MACRO)
to errfinish not errstart. This shaves a few cycles off the case where
errstart decides we're not going to emit anything.
Re-implement elog() as a trivial wrapper around ereport(), removing
the separate support infrastructure it used to have. Aside from
getting rid of some now-surplus code, this means that elog() now
really does have exactly the same semantics as ereport(), in particular
that it can skip evaluation work if the message is not to be emitted.
Andres Freund and Tom Lane
Discussion: https://postgr.es/m/CA+fd4k6N8EjNvZpM8nme+y+05mz-SM8Z_BgkixzkA34R+ej0Kw@mail.gmail.com
While performing hash aggregation, track memory usage when adding new
groups to a hash table. If the memory usage exceeds work_mem, enter
"spill mode".
In spill mode, new groups are not created in the hash table(s), but
existing groups continue to be advanced if input tuples match. Tuples
that would cause a new group to be created are instead spilled to a
logical tape to be processed later.
The tuples are spilled in a partitioned fashion. When all tuples from
the outer plan are processed (either by advancing the group or
spilling the tuple), finalize and emit the groups from the hash
table. Then, create new batches of work from the spilled partitions,
and select one of the saved batches and process it (possibly spilling
recursively).
Author: Jeff Davis
Reviewed-by: Tomas Vondra, Adam Lee, Justin Pryzby, Taylor Vesely, Melanie Plageman
Discussion: https://postgr.es/m/507ac540ec7c20136364b5272acbcd4574aa76ef.camel@j-davis.com
Optionally push a step to check for a NULL pointer to the pergroup
state.
This will be important for disk-based hash aggregation in combination
with grouping sets. When memory limits are reached, a given tuple may
find its per-group state for some grouping sets but not others. For
the former, it advances the per-group state as normal; for the latter,
it skips evaluation and the calling code will have to spill the tuple
and reprocess it in a later batch.
Add the NULL check as a separate expression step because in some
common cases it's not needed.
Discussion: https://postgr.es/m/20200221202212.ssb2qpmdgrnx52sj%40alap3.anarazel.de
Do so by combining the various steps that are part of aggregate
transition function invocation into one larger step. As some of the
current steps are only necessary for some aggregates, have one variant
of the aggregate transition step for each possible combination.
To avoid further manual copies of code in the different transition
step implementations, move most of the code into helper functions
marked as "always inline".
The benefit of this change is an increase in performance when
aggregating lots of rows. This comes in part due to the reduced number
of indirect jumps due to the reduced number of steps, and in part by
reducing redundant setup code across steps. This mainly benefits
interpreted execution, but the code generated by JIT is also improved
a bit.
As a nice side-effect it also ends up making the code a bit simpler.
A small additional optimization is removing the need to set
aggstate->curaggcontext before calling ExecAggInitGroup, choosing to
instead passign curaggcontext as an argument. It was, in contrast to
other aggregate related functions, only needed to fetch a memory
context to copy the transition value into.
Author: Andres Freund
Discussion:
https://postgr.es/m/20191023163849.sosqbfs5yenocez3@alap3.anarazel.dehttps://postgr.es/m/5c371df7cee903e8cd4c685f90c6c72086d3a2dc.camel@j-davis.com
When updating a table row with generated columns, only recompute those
generated columns whose base columns have changed in this update and
keep the rest unchanged. This can result in a significant performance
benefit. The required information was already kept in
RangeTblEntry.extraUpdatedCols; we just have to make use of it.
Reviewed-by: Pavel Stehule <pavel.stehule@gmail.com>
Discussion: https://www.postgresql.org/message-id/flat/b05e781a-fa16-6b52-6738-761181204567@2ndquadrant.com
Commit 4eaea3db introduced TupleHashTableHash(), but the signature
didn't match the other exposed functions. Separate it into internal
and external versions. The external version hides the details behind
an API more consistent with the other external functions, and the
internal version is still suitable for simplehash.
Expose two new entry points: one for only calculating the hash value
of a tuple, and another for looking up a hash entry when the hash
value is already known. This will be useful for disk-based Hash
Aggregation to avoid recomputing the hash value for the same tuple
after saving and restoring it from disk.
Discussion: https://postgr.es/m/37091115219dd522fd9ed67333ee8ed1b7e09443.camel%40j-davis.com
It's already tracked via ExprState->parent, so we don't need to also
include it in ExprEvalStep. When that code originally was written
ExprState->parent didn't exist, but it since has been introduced in
6719b238e8.
Author: Andres Freund
Discussion: https://postgr.es/m/20191023163849.sosqbfs5yenocez3@alap3.anarazel.de
Use the parser's standard type coercion machinery to convert the
output column(s) of a SQL function's final SELECT or RETURNING
to the type(s) they should have according to the function's declared
result type. We'll allow any case where an assignment-level
coercion is available. Previously, we failed unless the required
coercion was a binary-compatible one (and the documentation ignored
this, falsely claiming that the types must match exactly).
Notably, the coercion now accounts for typmods, so that cases where
a SQL function is declared to return a composite type whose columns
are typmod-constrained now behave as one would expect. Arguably
this aspect is a bug fix, but the overall behavioral change here
seems too large to consider back-patching.
A nice side-effect is that functions can now be inlined in a
few cases where we previously failed to do so because of type
mismatches.
Discussion: https://postgr.es/m/18929.1574895430@sss.pgh.pa.us
This follows multiple complains from Peter Geoghegan, Andres Freund and
Alvaro Herrera that this issue ought to be dug more before actually
happening, if it happens.
Discussion: https://postgr.es/m/20191226144606.GA5659@alvherre.pgsql
