public inbox for [email protected]
help / color / mirror / Atom feedFrom: Amit Langote <[email protected]>
To: Chao Li <[email protected]>
Cc: Tom Lane <[email protected]>
Cc: Tender Wang <[email protected]>
Cc: Alexander Lakhin <[email protected]>
Cc: Tomas Vondra <[email protected]>
Cc: Robert Haas <[email protected]>
Cc: Alvaro Herrera <[email protected]>
Cc: Andres Freund <[email protected]>
Cc: Daniel Gustafsson <[email protected]>
Cc: David Rowley <[email protected]>
Cc: PostgreSQL Hackers <[email protected]>
Cc: Thom Brown <[email protected]>
Subject: Re: generic plans and "initial" pruning
Date: Fri, 20 Mar 2026 02:20:49 +0900
Message-ID: <CA+HiwqHBxDL=3qQa1f-sBOBZqB88EiVAiagXF3X8Kagpr6Yhpw@mail.gmail.com> (raw)
In-Reply-To: <CA+HiwqELAcgVg_3Gb4VTOpC6wcNhHP0m-8OJFG0MeGRo0M=_4Q@mail.gmail.com>
References: <CA+HiwqFpZ80UJKr4tZus4Omgg7YESzFXKSwSHRW2Ap2=XSVyUA@mail.gmail.com>
<[email protected]>
<CA+HiwqH8N-SxEB6SysEBsYNgV_KJs66k9Z2SNmqVzbBP-60yWg@mail.gmail.com>
<[email protected]>
<CA+HiwqEmG9YCQvG6uux7sO=jKFSAW6hA4Ea-ymfD+JhJAe4PWQ@mail.gmail.com>
<CA+HiwqE2FfJfH=siLiR3kJ13tmXZORAGTWsZc2r52o1_5BDv+g@mail.gmail.com>
<[email protected]>
<CA+HiwqFhkpXHAA=4NY5SqYXX08uq=nYtXcSByNZF=2MAy1UA7A@mail.gmail.com>
<CA+HiwqHCcSoYfpMjFshaU1bj6NjreiDvMSDpVSeBmqk-kbWrPw@mail.gmail.com>
<CA+HiwqHOejJk0_qMuM5g38h70hY_JvHMAKwnH3k=urfTXauPQA@mail.gmail.com>
<CA+HiwqFsGKM82oaMby3VWYXf_XFpDAMeT+6SXgj-45HpTrS1dA@mail.gmail.com>
<CA+HiwqFA5hUWYktt3VMh4zQOYMxqH-MpdX8eemfM+o-9dY-zbQ@mail.gmail.com>
<CA+HiwqEn7bbUXaXO=SmUujBjJSHfS31cwQroHRBwT0sR=66bgg@mail.gmail.com>
<CA+HiwqGGLDTd1ZTK1c0zv4La7XOVSVMqBuNtscJeh6FyUQvFvA@mail.gmail.com>
<CA+HiwqE2JFiqqrXdyJVQWY-fMGwzDkLqjXQdUKbPaCpDpxd_2g@mail.gmail.com>
<CA+HiwqFp3jZGSz==QjeuV62_62F6+V6b62=Uqvy99sW_gsgWBA@mail.gmail.com>
<CAHewXNkUz9XGG8nnoxZaw35e+5bQVVP=eeJE4cW4V2e+P9ndFw@mail.gmail.com>
<CA+HiwqFKSpfYruzcVz-5CcFxg7gMa+ycXjMa2aPz_J_P4LGXTg@mail.gmail.com>
<[email protected]>
<CA+HiwqEQ1oME-hcDXwC9rGQb=u7MdUFG3Sc=Qg27uH480v10FA@mail.gmail.com>
<[email protected]>
<CA+HiwqGXMLSQyJvynWF40yNwBAx-pXtxemReP8L+C+kaUa5v5A@mail.gmail.com>
<CA+HiwqGBfMgcxokEH_mg6s=ttLFm54dj4hT6yXydU2t0g6oQ3g@mail.gmail.com>
<CA+HiwqEEkGfMc_LSJhfz96o-czVS4B59Vhw6i1_t58ZGqhP8VA@mail.gmail.com>
<CA+HiwqHAd+9nptjxP6=KrcKA1BMsS6pbB3B2oaojwdyH_wBWCA@mail.gmail.com>
<CA+HiwqE7_YpU--EsrhvNqcZ+10+92EGFaX5609AUJb9ENLntnQ@mail.gmail.com>
<CA+HiwqEF9SgKyQ1HrYOURpv8DGRGHDNwBT9Y6yEBVCW+=kh_=w@mail.gmail.com>
<CA+HiwqFpEHBjosRackQhm6yKKnHgqm8Ewkn=qsctT1N0PqVSrg@mail.gmail.com>
<CA+HiwqGJP91Qed0EjuB72Lv4_QAiVOMYjya7GA0aas8K6NZUZA@mail.gmail.com>
<[email protected]>
<CA+HiwqE7LDSoaF024Mt9v1Gt-uE-WoT9GawC5ds45SaPczV8Qw@mail.gmail.com>
<CA+HiwqGn38DsKgMYKWZ6jyv3_oqCSB0j+XucTjNM0S+BFsQpVA@mail.gmail.com>
<CA+HiwqGFNe7kBkKZm0KtG_CFfw-ciK659SJMGP0CWVaa2q8rmw@mail.gmail.com>
<CA+HiwqELAcgVg_3Gb4VTOpC6wcNhHP0m-8OJFG0MeGRo0M=_4Q@mail.gmail.com>
On Mon, Mar 9, 2026 at 1:41 PM Amit Langote <[email protected]> wrote:
> On Sat, Mar 7, 2026 at 6:54 PM Amit Langote <[email protected]> wrote:
> > Attached is v6 of the patch series. I've been working toward
> > committing this, so I wanted to lay out the ExecutorPrep() design and
> > the key trade-offs before doing so.
> >
> > When a cached generic plan references a partitioned table,
> > GetCachedPlan() locks all partitions upfront via
> > AcquireExecutorLocks(), even those that initial pruning will
> > eliminate. But initial partition pruning only runs later during
> > ExecutorStart(). Moving pruning earlier requires some executor setup
> > (range table, permissions, pruning state), and ExecutorPrep() is the
> > vehicle for that. Unlike the approach reverted in last May, this
> > keeps the CachedPlan itself unchanged -- all per-execution state flows
> > through a separate CachedPlanPrepData that the caller provides.
> >
> > The approach also keeps GetCachedPlan()'s interface
> > backward-compatible: the new CachedPlanPrepData argument is optional.
> > If a caller passes NULL, all partitions are locked as before and
> > nothing changes. This means existing callers and any new code that
> > calls GetCachedPlan() without caring about pruning-aware locking just
> > works.
> >
> > The risk is on the other side: if a caller does pass a
> > CachedPlanPrepData, GetCachedPlan() will lock only the surviving
> > partitions and populate prep_estates with the EStates that
> > ExecutorPrep() created. The caller then must make those EStates
> > available to ExecutorStart() -- via QueryDesc->estate,
> > portal->prep_estates, or the equivalent path for SPI and SQL
> > functions. If it fails to do so, ExecutorStart() will call
> > ExecutorPrep() again, which may compute different pruning results than
> > the original call, potentially expecting locks on relations that were
> > never acquired. The executor would then operate on relations it
> > doesn't hold locks on.
> >
> > So the contract is: if you opt in to pruning-aware locking by passing
> > CachedPlanPrepData, you must complete the pipeline by delivering the
> > prep EStates to the executor. In the current patch, all the call sites
> > that pass a CachedPlanPrepData (portals, SPI, EXECUTE, SQL functions,
> > EXPLAIN) do thread the EStates through correctly, and I've tried to
> > make the plumbing straightforward enough that it's hard to get wrong.
> > But it is a new invariant that didn't exist before, and a caller that
> > gets it wrong would fail silently rather than with an obvious error.
> >
> > To catch such violations, I've added a debug-only check in
> > standard_ExecutorStart() that fires when no prep EState was provided.
> > It iterates over the plan's rtable and verifies that every lockable
> > relation is actually locked. It should always be true if
> > AcquireExecutorLocks() locked everything, but would fail if
> > pruning-aware locking happened upstream and the caller dropped the
> > prep EState. The check is skipped in parallel workers, which acquire
> > relation locks lazily in ExecGetRangeTableRelation().
> >
> > + if (queryDesc->estate == NULL)
> > + {
> > +#ifdef USE_ASSERT_CHECKING
> > + if (!IsParallelWorker())
> > + {
> > + ListCell *lc;
> > +
> > + foreach(lc, queryDesc->plannedstmt->rtable)
> > + {
> > + RangeTblEntry *rte = lfirst_node(RangeTblEntry, lc);
> > +
> > + if (rte->rtekind == RTE_RELATION ||
> > + (rte->rtekind == RTE_SUBQUERY && rte->relid != InvalidOid))
> > + Assert(CheckRelationOidLockedByMe(rte->relid,
> > + rte->rellockmode,
> > + true));
> > + }
> > + }
> > +#endif
> > + queryDesc->estate = ExecutorPrep(queryDesc->plannedstmt,
> > + queryDesc->params,
> > + CurrentResourceOwner,
> > + true,
> > + eflags);
> > + }
> > +#ifdef USE_ASSERT_CHECKING
> > + else
> > + {
> > + /*
> > + * A prep EState was provided, meaning pruning-aware locking
> > + * should have locked at least the unpruned relations.
> > + */
> > + if (!IsParallelWorker())
> > + {
> > + int rtindex = -1;
> > +
> > + while ((rtindex =
> > bms_next_member(queryDesc->estate->es_unpruned_relids,
> > + rtindex)) >= 0)
> > + {
> > + RangeTblEntry *rte = exec_rt_fetch(rtindex, queryDesc->estate);
> > +
> > + Assert(rte->rtekind == RTE_RELATION ||
> > + (rte->rtekind == RTE_SUBQUERY &&
> > + rte->relid != InvalidOid));
> > + Assert(CheckRelationOidLockedByMe(rte->relid,
> > + rte->rellockmode, true));
> > + }
> > + }
> > + }
> > +#endif
> >
> > So the invariant is: if no prep EState was provided, every relation in
> > the plan is locked; if one was provided, at least the unpruned
> > relations are locked. Both are checked in assert builds.
> >
> > I think this covers the main concerns, but I may be missing something.
> > If anyone sees a problem with this approach, I'd like to hear about
> > it.
>
> Here's v7. Some plancache.c changes that I'd made were in the wrong
> patch in v6; this version puts them where they belong.
Attached is an updated set. One more fix: I added an Assert in
SPI_cursor_open_internal()'s !plan->saved path to verify that
prep_estates is NIL. Unsaved plans always take the custom plan path,
so pruning-aware locking never applies, but it's worth guarding
explicitly since the copyObject/ReleaseCachedPlan sequence that
follows would not be safe otherwise. Also changed
SPI_plan_get_cached_plan() to pass NULL for cprep, since it only
returns the CachedPlan pointer and has no way to deliver prep_estates
to anyone.
Stepping back -- the core question is whether running executor logic
(pruning) inside GetCachedPlan() is acceptable at all. The plan cache
and executor have always had a clean boundary: plan cache locks
everything, executor runs. This optimization necessarily crosses that
line, because the information needed to decide which locks to skip
(pruning results) can only come from executor machinery.
The proposed approach has GetCachedPlan() call ExecutorPrep() to do a
limited subset of executor work (range table init, permissions,
pruning), carry the results out through CachedPlanPrepData, and leave
the CachedPlan itself untouched. The executor already has a multi-step
protocol: start/run/end. prep/start/run/end is just a finer
decomposition of what InitPlan() was already doing inside
ExecutorStart().
Of the attached patches, I'm targeting 0001-0003 for commit. 0004 (SQL
function support) and 0005 (parallel worker reuse) are useful
follow-ons but not essential. The optimization works without them for
most cases, and they can be reviewed and committed separately.
If there's a cleaner way to avoid locking pruned partitions without
the plumbing this patch adds, I haven't found it in the year since the
revert. I'd welcome a pointer if you see one. Failing that, I think
this is the right trade-off, but it's a judgment call about where to
hold your nose.
Tom, I'd value your opinion on whether this approach is something
you'd be comfortable seeing in the tree.
--
Thanks, Amit Langote
Attachments:
[application/octet-stream] v8-0005-Reuse-partition-pruning-results-in-parallel-worke.patch (11.0K, 2-v8-0005-Reuse-partition-pruning-results-in-parallel-worke.patch)
download | inline diff:
From 4c12c380b75b8684e9c41c80d0c77027cf592e17 Mon Sep 17 00:00:00 2001
From: Amit Langote <[email protected]>
Date: Thu, 19 Mar 2026 20:03:58 +0900
Subject: [PATCH v8 5/5] Reuse partition pruning results in parallel workers
Pass the leader's initial partition pruning results and unpruned
relids to parallel workers and reuse them via ExecutorPrep(). This
avoids repeating pruning logic in workers, which is not only
redundant but also risks divergence due to nondeterminism in pruning
steps or parameter evaluation timing.
Introduce CheckInitialPruningResultsInWorker() (debug-builds only)
to verify that the results match what the worker would compute. This
check helps catch inconsistencies across leader and worker pruning
logic.
---
src/backend/executor/execMain.c | 10 +--
src/backend/executor/execParallel.c | 108 +++++++++++++++++++++++++++-
src/backend/utils/cache/plancache.c | 2 +-
src/include/executor/executor.h | 3 +-
4 files changed, 116 insertions(+), 7 deletions(-)
diff --git a/src/backend/executor/execMain.c b/src/backend/executor/execMain.c
index 0f95ad88497..9a3700e672f 100644
--- a/src/backend/executor/execMain.c
+++ b/src/backend/executor/execMain.c
@@ -207,7 +207,7 @@ standard_ExecutorStart(QueryDesc *queryDesc, int eflags)
queryDesc->estate = ExecutorPrep(queryDesc->plannedstmt,
queryDesc->params,
CurrentResourceOwner,
- eflags);
+ eflags, true);
}
#ifdef USE_ASSERT_CHECKING
else
@@ -330,7 +330,8 @@ standard_ExecutorStart(QueryDesc *queryDesc, int eflags)
* ExecutorPrep: prepare executor state for a PlannedStmt outside ExecutorStart.
*
* Performs range table initialization, permission checks, and initial
- * partition pruning if partPruneInfos are present.
+ * partition pruning if partPruneInfos are present and do_initial_pruning is
+ * true; false in a parallel worker.
*
* Returns an EState that the caller must either pass to ExecutorStart()
* for reuse or free via FreeExecutorState() if execution will not proceed.
@@ -340,7 +341,7 @@ standard_ExecutorStart(QueryDesc *queryDesc, int eflags)
*/
EState *
ExecutorPrep(PlannedStmt *pstmt, ParamListInfo params, ResourceOwner owner,
- int eflags)
+ int eflags, bool do_initial_pruning)
{
ResourceOwner oldowner;
EState *estate;
@@ -386,7 +387,8 @@ ExecutorPrep(PlannedStmt *pstmt, ParamListInfo params, ResourceOwner owner,
* to es_part_prune_infos.
*/
ExecCreatePartitionPruneStates(estate);
- ExecDoInitialPruning(estate);
+ if (do_initial_pruning)
+ ExecDoInitialPruning(estate);
CurrentResourceOwner = oldowner;
diff --git a/src/backend/executor/execParallel.c b/src/backend/executor/execParallel.c
index 024780d3516..2de4b35a16e 100644
--- a/src/backend/executor/execParallel.c
+++ b/src/backend/executor/execParallel.c
@@ -24,6 +24,7 @@
#include "postgres.h"
#include "executor/execParallel.h"
+#include "executor/execPartition.h"
#include "executor/executor.h"
#include "executor/nodeAgg.h"
#include "executor/nodeAppend.h"
@@ -67,6 +68,8 @@
#define PARALLEL_KEY_QUERY_TEXT UINT64CONST(0xE000000000000008)
#define PARALLEL_KEY_JIT_INSTRUMENTATION UINT64CONST(0xE000000000000009)
#define PARALLEL_KEY_WAL_USAGE UINT64CONST(0xE00000000000000A)
+#define PARALLEL_KEY_PARTITION_PRUNE_RESULTS UINT64CONST(0xE00000000000000B)
+#define PARALLEL_KEY_UNPRUNED_RELIDS UINT64CONST(0xE00000000000000C)
#define PARALLEL_TUPLE_QUEUE_SIZE 65536
@@ -141,6 +144,8 @@ static bool ExecParallelRetrieveInstrumentation(PlanState *planstate,
/* Helper function that runs in the parallel worker. */
static DestReceiver *ExecParallelGetReceiver(dsm_segment *seg, shm_toc *toc);
+static void CheckInitialPruningResultsInWorker(EState *estate);
+
/*
* Create a serialized representation of the plan to be sent to each worker.
*/
@@ -620,12 +625,18 @@ ExecInitParallelPlan(PlanState *planstate, EState *estate,
FixedParallelExecutorState *fpes;
char *pstmt_data;
char *pstmt_space;
+ char *part_prune_results_data;
+ char *part_prune_results_space;
+ char *unpruned_relids_data;
+ char *unpruned_relids_space;
char *paramlistinfo_space;
BufferUsage *bufusage_space;
WalUsage *walusage_space;
SharedExecutorInstrumentation *instrumentation = NULL;
SharedJitInstrumentation *jit_instrumentation = NULL;
int pstmt_len;
+ int part_prune_results_len;
+ int unpruned_relids_len;
int paramlistinfo_len;
int instrumentation_len = 0;
int jit_instrumentation_len = 0;
@@ -654,6 +665,8 @@ ExecInitParallelPlan(PlanState *planstate, EState *estate,
/* Fix up and serialize plan to be sent to workers. */
pstmt_data = ExecSerializePlan(planstate->plan, estate);
+ part_prune_results_data = nodeToString(estate->es_part_prune_results);
+ unpruned_relids_data = nodeToString(estate->es_unpruned_relids);
/* Create a parallel context. */
pcxt = CreateParallelContext("postgres", "ParallelQueryMain", nworkers);
@@ -680,6 +693,16 @@ ExecInitParallelPlan(PlanState *planstate, EState *estate,
shm_toc_estimate_chunk(&pcxt->estimator, pstmt_len);
shm_toc_estimate_keys(&pcxt->estimator, 1);
+ /* Estimate space for serialized part_prune_results. */
+ part_prune_results_len = strlen(part_prune_results_data) + 1;
+ shm_toc_estimate_chunk(&pcxt->estimator, part_prune_results_len);
+ shm_toc_estimate_keys(&pcxt->estimator, 1);
+
+ /* Estimate space for serialized unpruned_relids. */
+ unpruned_relids_len = strlen(unpruned_relids_data) + 1;
+ shm_toc_estimate_chunk(&pcxt->estimator, unpruned_relids_len);
+ shm_toc_estimate_keys(&pcxt->estimator, 1);
+
/* Estimate space for serialized ParamListInfo. */
paramlistinfo_len = EstimateParamListSpace(estate->es_param_list_info);
shm_toc_estimate_chunk(&pcxt->estimator, paramlistinfo_len);
@@ -781,6 +804,16 @@ ExecInitParallelPlan(PlanState *planstate, EState *estate,
memcpy(pstmt_space, pstmt_data, pstmt_len);
shm_toc_insert(pcxt->toc, PARALLEL_KEY_PLANNEDSTMT, pstmt_space);
+ /* Store serialized part_prune_results */
+ part_prune_results_space = shm_toc_allocate(pcxt->toc, part_prune_results_len);
+ memcpy(part_prune_results_space, part_prune_results_data, part_prune_results_len);
+ shm_toc_insert(pcxt->toc, PARALLEL_KEY_PARTITION_PRUNE_RESULTS, part_prune_results_space);
+
+ /* Store serialized unpruned_relids */
+ unpruned_relids_space = shm_toc_allocate(pcxt->toc, unpruned_relids_len);
+ memcpy(unpruned_relids_space, unpruned_relids_data, unpruned_relids_len);
+ shm_toc_insert(pcxt->toc, PARALLEL_KEY_UNPRUNED_RELIDS, unpruned_relids_space);
+
/* Store serialized ParamListInfo. */
paramlistinfo_space = shm_toc_allocate(pcxt->toc, paramlistinfo_len);
shm_toc_insert(pcxt->toc, PARALLEL_KEY_PARAMLISTINFO, paramlistinfo_space);
@@ -1280,10 +1313,15 @@ ExecParallelGetQueryDesc(shm_toc *toc, DestReceiver *receiver,
int instrument_options)
{
char *pstmtspace;
+ char *part_prune_results_space;
+ char *unpruned_relids_space;
char *paramspace;
PlannedStmt *pstmt;
+ List *part_prune_results;
+ Bitmapset *unpruned_relids;
ParamListInfo paramLI;
char *queryString;
+ EState *prep_estate = NULL;
/* Get the query string from shared memory */
queryString = shm_toc_lookup(toc, PARALLEL_KEY_QUERY_TEXT, false);
@@ -1296,12 +1334,80 @@ ExecParallelGetQueryDesc(shm_toc *toc, DestReceiver *receiver,
paramspace = shm_toc_lookup(toc, PARALLEL_KEY_PARAMLISTINFO, false);
paramLI = RestoreParamList(¶mspace);
+ /* Reconstruct leader-supplied part_prune_results and unpruned_relids. */
+ part_prune_results_space =
+ shm_toc_lookup(toc, PARALLEL_KEY_PARTITION_PRUNE_RESULTS, false);
+ part_prune_results = (List *) stringToNode(part_prune_results_space);
+ unpruned_relids_space =
+ shm_toc_lookup(toc, PARALLEL_KEY_UNPRUNED_RELIDS, false);
+ unpruned_relids = (Bitmapset *) stringToNode(unpruned_relids_space);
+
+ /*
+ * If pruning was done in the leader, build a prep estate in the worker
+ * and inject the leader's pruning results into it for reuse.
+ */
+ if (pstmt->partPruneInfos)
+ {
+ prep_estate = ExecutorPrep(pstmt, paramLI, CurrentResourceOwner, 0, false);
+ Assert(prep_estate);
+
+ prep_estate->es_part_prune_results = part_prune_results;
+ prep_estate->es_unpruned_relids =
+ bms_add_members(prep_estate->es_unpruned_relids,
+ unpruned_relids);
+
+ /*
+ * A debug-build-only check that the pruning results passed from the
+ * leader match what the worker would independently compute.
+ */
+ CheckInitialPruningResultsInWorker(prep_estate);
+ }
+
/* Create a QueryDesc for the query. */
return CreateQueryDesc(pstmt,
queryString,
GetActiveSnapshot(), InvalidSnapshot,
receiver, paramLI, NULL, instrument_options,
- NULL);
+ prep_estate);
+}
+
+/*
+ * CheckInitialPruningResultsInWorker
+ * Verify partition pruning results passed from the leader process.
+ *
+ * This is intended to be called during parallel worker query setup.
+ * It recomputes initial pruning results locally and compares them with
+ * those received from the leader. Any mismatch may indicate a divergence
+ * between leader and worker logic or environment.
+ *
+ * Only performed in debug builds.
+ */
+static void
+CheckInitialPruningResultsInWorker(EState *estate)
+{
+#ifdef USE_ASSERT_CHECKING
+ ListCell *lc;
+ int i;
+
+ Assert(estate->es_part_prune_results != NULL);
+ i = 0;
+ foreach(lc, estate->es_part_prune_states)
+ {
+ PartitionPruneState *prunestate = (PartitionPruneState *) lfirst(lc);
+ Bitmapset *reuse_validsubplans =
+ list_nth_node(Bitmapset, estate->es_part_prune_results, i++);
+ Bitmapset *validsubplans = NULL;
+ Bitmapset *validsubplan_rtis = NULL;
+
+ if (prunestate->do_initial_prune)
+ validsubplans = ExecFindMatchingSubPlans(prunestate, true,
+ &validsubplan_rtis);
+ if (!bms_equal(validsubplans, reuse_validsubplans))
+ elog(ERROR, "different validsubplans in parallel worker");
+ if (bms_nonempty_difference(validsubplan_rtis, estate->es_unpruned_relids))
+ elog(ERROR, "different unprunable_relids in parallel worker");
+ }
+#endif
}
/*
diff --git a/src/backend/utils/cache/plancache.c b/src/backend/utils/cache/plancache.c
index 2d4c57d3deb..0dd4f40c964 100644
--- a/src/backend/utils/cache/plancache.c
+++ b/src/backend/utils/cache/plancache.c
@@ -2102,7 +2102,7 @@ AcquireExecutorLocksUnpruned(List *stmt_list, bool acquire,
}
prep_estate = ExecutorPrep(plannedstmt, cprep->params,
- cprep->owner, cprep->eflags);
+ cprep->owner, cprep->eflags, true);
Assert(prep_estate);
cprep->prep_estates = lappend(cprep->prep_estates, prep_estate);
diff --git a/src/include/executor/executor.h b/src/include/executor/executor.h
index 24604120c27..38848ba0651 100644
--- a/src/include/executor/executor.h
+++ b/src/include/executor/executor.h
@@ -240,7 +240,8 @@ extern void standard_ExecutorStart(QueryDesc *queryDesc, int eflags);
extern EState *ExecutorPrep(PlannedStmt *pstmt,
ParamListInfo params,
ResourceOwner owner,
- int eflags);
+ int eflags,
+ bool do_initial_pruning);
/*
* Walk a prep_estates list in step with a parallel stmt_list iteration.
--
2.47.3
[application/octet-stream] v8-0003-Use-pruning-aware-locking-in-cached-plans.patch (41.1K, 3-v8-0003-Use-pruning-aware-locking-in-cached-plans.patch)
download | inline diff:
From 2e637cbc71a14775e161bde21e1036eca2644a2b Mon Sep 17 00:00:00 2001
From: Amit Langote <[email protected]>
Date: Thu, 19 Mar 2026 19:02:04 +0900
Subject: [PATCH v8 3/5] Use pruning-aware locking in cached plans
Extend GetCachedPlan() to perform ExecutorPrep() on each planned
statement, capturing unpruned relids and initial pruning results.
Use this data to acquire execution locks only on surviving partitions,
avoiding unnecessary locking of pruned tables even when using cached
plans.
Introduce CachedPlanPrepData to carry the EStates created by
ExecutorPrep() through the plan caching layer. The prep_estates
list is indexed one-to-one with CachedPlan->stmt_list and is
populated when GetCachedPlan() prepares a reused generic plan.
Adjust call sites in SPI, functions, portals, and EXPLAIN to
propagate this data.
Partition pruning expressions may call PL functions that require
an active snapshot (e.g., via EnsurePortalSnapshotExists()).
AcquireExecutorLocksUnpruned() establishes one before calling
ExecutorPrep() if needed, ensuring these expressions can execute
correctly during plan cache validation.
To maintain correctness when all target partitions are pruned, also
reinstate the firstResultRel locking behavior lost in commit
28317de72. That commit required the first ModifyTable target to
remain initialized for executor assumptions to hold. We now
explicitly track these relids in PlannerGlobal and PlannedStmt so they
are locked even if pruned, preserving that rule across cached plan
reuse.
Regression tests are included to verify:
- Only surviving partitions are locked when pruning is enabled, and
all partitions are locked when it is disabled (pg_locks inspection).
- Multiple ModifyTable nodes (via writable CTEs) handle the case where
all target partitions are pruned, exercising firstResultRels.
- Plan invalidation during pruning-aware lock setup (DDL triggered by
a pruning expression) discards the prep state and replans cleanly.
Note for extension authors: code that accesses partition relations
through EState must check that the RT index is a member of
es_unpruned_relids before opening the relation. Previously this was
an optimization (avoid processing pruned partitions); it is now a
correctness requirement, because pruned partitions may not be locked.
ExecGetRangeTableRelation() already enforces this with an error when
called on a pruned relation.
---
src/backend/commands/prepare.c | 17 +-
src/backend/executor/functions.c | 1 +
src/backend/executor/nodeModifyTable.c | 5 +-
src/backend/executor/spi.c | 22 +-
src/backend/optimizer/plan/planner.c | 1 +
src/backend/optimizer/plan/setrefs.c | 20 ++
src/backend/tcop/postgres.c | 7 +-
src/backend/utils/cache/plancache.c | 257 +++++++++++++++++-
src/include/nodes/pathnodes.h | 3 +
src/include/nodes/plannodes.h | 10 +
src/include/utils/plancache.h | 35 ++-
src/test/regress/expected/partition_prune.out | 145 ++++++++++
src/test/regress/expected/plancache.out | 62 +++++
src/test/regress/sql/partition_prune.sql | 77 ++++++
src/test/regress/sql/plancache.sql | 51 ++++
15 files changed, 689 insertions(+), 24 deletions(-)
diff --git a/src/backend/commands/prepare.c b/src/backend/commands/prepare.c
index c7bab14b633..fec83cc6fd4 100644
--- a/src/backend/commands/prepare.c
+++ b/src/backend/commands/prepare.c
@@ -156,6 +156,7 @@ ExecuteQuery(ParseState *pstate,
{
PreparedStatement *entry;
CachedPlan *cplan;
+ CachedPlanPrepData cprep = {0};
List *plan_list;
ParamListInfo paramLI = NULL;
EState *estate = NULL;
@@ -195,7 +196,9 @@ ExecuteQuery(ParseState *pstate,
entry->plansource->query_string);
/* Replan if needed, and increment plan refcount for portal */
- cplan = GetCachedPlan(entry->plansource, paramLI, NULL, NULL);
+ cprep.context = portal->portalContext;
+ cprep.owner = portal->resowner;
+ cplan = GetCachedPlan(entry->plansource, paramLI, NULL, NULL, &cprep);
plan_list = cplan->stmt_list;
/*
@@ -207,7 +210,7 @@ ExecuteQuery(ParseState *pstate,
query_string,
entry->plansource->commandTag,
plan_list,
- NIL,
+ cprep.prep_estates,
cplan);
/*
@@ -577,6 +580,7 @@ ExplainExecuteQuery(ExecuteStmt *execstmt, IntoClause *into, ExplainState *es,
PreparedStatement *entry;
const char *query_string;
CachedPlan *cplan;
+ CachedPlanPrepData cprep = {0};
List *plan_list;
List *prep_estates;
ListCell *p;
@@ -635,8 +639,13 @@ ExplainExecuteQuery(ExecuteStmt *execstmt, IntoClause *into, ExplainState *es,
}
/* Replan if needed, and acquire a transient refcount */
+ cprep.context = CurrentMemoryContext;
+ cprep.owner = CurrentResourceOwner;
+ if (es->generic)
+ cprep.eflags = EXEC_FLAG_EXPLAIN_GENERIC;
cplan = GetCachedPlan(entry->plansource, paramLI,
- CurrentResourceOwner, pstate->p_queryEnv);
+ CurrentResourceOwner, pstate->p_queryEnv,
+ &cprep);
INSTR_TIME_SET_CURRENT(planduration);
INSTR_TIME_SUBTRACT(planduration, planstart);
@@ -655,7 +664,7 @@ ExplainExecuteQuery(ExecuteStmt *execstmt, IntoClause *into, ExplainState *es,
}
plan_list = cplan->stmt_list;
- prep_estates = NIL;
+ prep_estates = cprep.prep_estates;
/* Explain each query */
prep_lc = list_head(prep_estates);
diff --git a/src/backend/executor/functions.c b/src/backend/executor/functions.c
index 952a784c924..c0ca72b38dd 100644
--- a/src/backend/executor/functions.c
+++ b/src/backend/executor/functions.c
@@ -699,6 +699,7 @@ init_execution_state(SQLFunctionCachePtr fcache)
fcache->cplan = GetCachedPlan(plansource,
fcache->paramLI,
fcache->cowner,
+ NULL,
NULL);
/*
diff --git a/src/backend/executor/nodeModifyTable.c b/src/backend/executor/nodeModifyTable.c
index 4cd5e262e0f..9230f2b554f 100644
--- a/src/backend/executor/nodeModifyTable.c
+++ b/src/backend/executor/nodeModifyTable.c
@@ -4865,8 +4865,8 @@ ExecInitModifyTable(ModifyTable *node, EState *estate, int eflags)
* as a reference for building the ResultRelInfo of the target partition.
* In either case, it doesn't matter which result relation is kept, so we
* just keep the first one, if all others have been pruned. See also,
- * ExecDoInitialPruning(), which ensures that this first result relation
- * has been locked.
+ * AcquireExecutorLocksUnpruned(), which ensures that this first result
+ * relation has been locked.
*/
i = 0;
foreach(l, node->resultRelations)
@@ -4880,6 +4880,7 @@ ExecInitModifyTable(ModifyTable *node, EState *estate, int eflags)
/* all result relations pruned; keep the first one */
keep_rel = true;
rti = linitial_int(node->resultRelations);
+ Assert(list_member_int(estate->es_plannedstmt->firstResultRels, rti));
i = 0;
}
diff --git a/src/backend/executor/spi.c b/src/backend/executor/spi.c
index 380bbc44e97..f1d84f7a350 100644
--- a/src/backend/executor/spi.c
+++ b/src/backend/executor/spi.c
@@ -1580,6 +1580,7 @@ SPI_cursor_open_internal(const char *name, SPIPlanPtr plan,
{
CachedPlanSource *plansource;
CachedPlan *cplan;
+ CachedPlanPrepData cprep = {0};
List *stmt_list;
char *query_string;
Snapshot snapshot;
@@ -1660,7 +1661,10 @@ SPI_cursor_open_internal(const char *name, SPIPlanPtr plan,
*/
/* Replan if needed, and increment plan refcount for portal */
- cplan = GetCachedPlan(plansource, paramLI, NULL, _SPI_current->queryEnv);
+ cprep.context = portal->portalContext;
+ cprep.owner = portal->resowner;
+ cplan = GetCachedPlan(plansource, paramLI, NULL, _SPI_current->queryEnv,
+ &cprep);
stmt_list = cplan->stmt_list;
if (!plan->saved)
@@ -1670,7 +1674,10 @@ SPI_cursor_open_internal(const char *name, SPIPlanPtr plan,
* so must copy the plan into the portal's context. An error here
* will result in leaking our refcount on the plan, but it doesn't
* matter because the plan is unsaved and hence transient anyway.
+ *
+ * Unsaved plans use custom plans, so prep should be a no-op.
*/
+ Assert(cprep.prep_estates == NIL);
oldcontext = MemoryContextSwitchTo(portal->portalContext);
stmt_list = copyObject(stmt_list);
MemoryContextSwitchTo(oldcontext);
@@ -1686,7 +1693,7 @@ SPI_cursor_open_internal(const char *name, SPIPlanPtr plan,
query_string,
plansource->commandTag,
stmt_list,
- NIL,
+ cprep.prep_estates,
cplan);
/*
@@ -2104,7 +2111,8 @@ SPI_plan_get_cached_plan(SPIPlanPtr plan)
/* Get the generic plan for the query */
cplan = GetCachedPlan(plansource, NULL,
plan->saved ? CurrentResourceOwner : NULL,
- _SPI_current->queryEnv);
+ _SPI_current->queryEnv,
+ NULL);
Assert(cplan == plansource->gplan);
/* Pop the error context stack */
@@ -2503,6 +2511,7 @@ _SPI_execute_plan(SPIPlanPtr plan, const SPIExecuteOptions *options,
ListCell *lc2;
List *prep_estates;
ListCell *prep_lc;
+ CachedPlanPrepData cprep = {0};
spicallbackarg.query = plansource->query_string;
@@ -2577,11 +2586,14 @@ _SPI_execute_plan(SPIPlanPtr plan, const SPIExecuteOptions *options,
* Replan if needed, and increment plan refcount. If it's a saved
* plan, the refcount must be backed by the plan_owner.
*/
+ cprep.context = CurrentMemoryContext;
+ cprep.owner = CurrentResourceOwner;
cplan = GetCachedPlan(plansource, options->params,
- plan_owner, _SPI_current->queryEnv);
+ plan_owner, _SPI_current->queryEnv,
+ &cprep);
stmt_list = cplan->stmt_list;
- prep_estates = NIL;
+ prep_estates = cprep.prep_estates;
/*
* If we weren't given a specific snapshot to use, and the statement
diff --git a/src/backend/optimizer/plan/planner.c b/src/backend/optimizer/plan/planner.c
index 42604a0f75c..afa61d357c5 100644
--- a/src/backend/optimizer/plan/planner.c
+++ b/src/backend/optimizer/plan/planner.c
@@ -657,6 +657,7 @@ standard_planner(Query *parse, const char *query_string, int cursorOptions,
result->permInfos = glob->finalrteperminfos;
result->subrtinfos = glob->subrtinfos;
result->resultRelations = glob->resultRelations;
+ result->firstResultRels = glob->firstResultRels;
result->appendRelations = glob->appendRelations;
result->subplans = glob->subplans;
result->rewindPlanIDs = glob->rewindPlanIDs;
diff --git a/src/backend/optimizer/plan/setrefs.c b/src/backend/optimizer/plan/setrefs.c
index 1b5b9b5ed9c..ddb7902bc89 100644
--- a/src/backend/optimizer/plan/setrefs.c
+++ b/src/backend/optimizer/plan/setrefs.c
@@ -384,6 +384,26 @@ set_plan_references(PlannerInfo *root, Plan *plan)
}
}
+ /*
+ * Record the first result relation if it belongs to the set of
+ * initially prunable relations. We use bms_next_member() to get
+ * the lowest-numbered leaf result rel, which matches
+ * linitial_int(ModifyTable.resultRelations) because partition
+ * expansion preserves RT index order. There is one ModifyTable
+ * per query level, so this captures exactly one entry per level.
+ * ExecInitModifyTable() asserts that the recorded index matches
+ * what it actually needs.
+ */
+ if (root->leaf_result_relids)
+ {
+ Index firstResultRel = bms_next_member(root->leaf_result_relids, -1);
+
+ firstResultRel += rtoffset;
+ if (bms_is_member(firstResultRel, root->glob->prunableRelids))
+ root->glob->firstResultRels =
+ lappend_int(root->glob->firstResultRels, firstResultRel);
+ }
+
return result;
}
diff --git a/src/backend/tcop/postgres.c b/src/backend/tcop/postgres.c
index 355a490cde9..de362ff1672 100644
--- a/src/backend/tcop/postgres.c
+++ b/src/backend/tcop/postgres.c
@@ -1637,6 +1637,7 @@ exec_bind_message(StringInfo input_message)
int16 *rformats = NULL;
CachedPlanSource *psrc;
CachedPlan *cplan;
+ CachedPlanPrepData cprep = {0};
Portal portal;
char *query_string;
char *saved_stmt_name;
@@ -2018,7 +2019,9 @@ exec_bind_message(StringInfo input_message)
* will be generated in MessageContext. The plan refcount will be
* assigned to the Portal, so it will be released at portal destruction.
*/
- cplan = GetCachedPlan(psrc, params, NULL, NULL);
+ cprep.context = portal->portalContext;
+ cprep.owner = portal->resowner;
+ cplan = GetCachedPlan(psrc, params, NULL, NULL, &cprep);
/*
* Now we can define the portal.
@@ -2031,7 +2034,7 @@ exec_bind_message(StringInfo input_message)
query_string,
psrc->commandTag,
cplan->stmt_list,
- NIL,
+ cprep.prep_estates,
cplan);
/* Portal is defined, set the plan ID based on its contents. */
diff --git a/src/backend/utils/cache/plancache.c b/src/backend/utils/cache/plancache.c
index 182c16e9b9a..2d4c57d3deb 100644
--- a/src/backend/utils/cache/plancache.c
+++ b/src/backend/utils/cache/plancache.c
@@ -93,14 +93,17 @@ static bool StmtPlanRequiresRevalidation(CachedPlanSource *plansource);
static bool BuildingPlanRequiresSnapshot(CachedPlanSource *plansource);
static List *RevalidateCachedQuery(CachedPlanSource *plansource,
QueryEnvironment *queryEnv);
-static bool CheckCachedPlan(CachedPlanSource *plansource);
+static bool CheckCachedPlan(CachedPlanSource *plansource, CachedPlanPrepData *cprep);
static CachedPlan *BuildCachedPlan(CachedPlanSource *plansource, List *qlist,
ParamListInfo boundParams, QueryEnvironment *queryEnv);
static bool choose_custom_plan(CachedPlanSource *plansource,
ParamListInfo boundParams);
static double cached_plan_cost(CachedPlan *plan, bool include_planner);
static Query *QueryListGetPrimaryStmt(List *stmts);
-static void AcquireExecutorLocks(List *stmt_list, bool acquire);
+static void AcquireExecutorLocksAll(List *stmt_list, bool acquire);
+static void AcquireExecutorLocksUnpruned(List *stmt_list, bool acquire,
+ CachedPlanPrepData *cprep);
+static void CachedPlanPrepCleanup(CachedPlanPrepData *cprep);
static void AcquirePlannerLocks(List *stmt_list, bool acquire);
static void ScanQueryForLocks(Query *parsetree, bool acquire);
static bool ScanQueryWalker(Node *node, bool *acquire);
@@ -942,6 +945,11 @@ RevalidateCachedQuery(CachedPlanSource *plansource,
/*
* CheckCachedPlan: see if the CachedPlanSource's generic plan is valid.
*
+ * If 'cprep' is not NULL, ExecutorPrep() is applied to each PlannedStmt to
+ * compute the set of partitions that survive initial runtime pruning in order
+ * to only lock them. The EStates created to do so are saved in cprep for
+ * later reuse by ExecutorStart().
+ *
* Caller must have already called RevalidateCachedQuery to verify that the
* querytree is up to date.
*
@@ -949,7 +957,7 @@ RevalidateCachedQuery(CachedPlanSource *plansource,
* (We must do this for the "true" result to be race-condition-free.)
*/
static bool
-CheckCachedPlan(CachedPlanSource *plansource)
+CheckCachedPlan(CachedPlanSource *plansource, CachedPlanPrepData *cprep)
{
CachedPlan *plan = plansource->gplan;
@@ -983,7 +991,10 @@ CheckCachedPlan(CachedPlanSource *plansource)
*/
Assert(plan->refcount > 0);
- AcquireExecutorLocks(plan->stmt_list, true);
+ if (cprep)
+ AcquireExecutorLocksUnpruned(plan->stmt_list, true, cprep);
+ else
+ AcquireExecutorLocksAll(plan->stmt_list, true);
/*
* If plan was transient, check to see if TransactionXmin has
@@ -1005,7 +1016,13 @@ CheckCachedPlan(CachedPlanSource *plansource)
}
/* Oops, the race case happened. Release useless locks. */
- AcquireExecutorLocks(plan->stmt_list, false);
+ if (cprep)
+ AcquireExecutorLocksUnpruned(plan->stmt_list, false, cprep);
+ else
+ AcquireExecutorLocksAll(plan->stmt_list, false);
+
+ /* Also clean up ExecutorPrep() state, if necessary. */
+ CachedPlanPrepCleanup(cprep);
}
/*
@@ -1285,6 +1302,15 @@ cached_plan_cost(CachedPlan *plan, bool include_planner)
* On return, the plan is valid and we have sufficient locks to begin
* execution.
*
+ * If 'cprep' is not NULL and a generic plan is reused, the function
+ * performs initial pruning via ExecutorPrep() and locks only the
+ * surviving partitions. The resulting EStates are stored in
+ * cprep->prep_estates and must be delivered to ExecutorStart() via
+ * QueryDesc->estate (or the equivalent portal/SPI path). Failure
+ * to do so means the executor will operate on relations for which
+ * locks were never acquired. Passing NULL for cprep is always safe;
+ * all partitions are locked as before.
+ *
* On return, the refcount of the plan has been incremented; a later
* ReleaseCachedPlan() call is expected. If "owner" is not NULL then
* the refcount has been reported to that ResourceOwner (note that this
@@ -1295,7 +1321,8 @@ cached_plan_cost(CachedPlan *plan, bool include_planner)
*/
CachedPlan *
GetCachedPlan(CachedPlanSource *plansource, ParamListInfo boundParams,
- ResourceOwner owner, QueryEnvironment *queryEnv)
+ ResourceOwner owner, QueryEnvironment *queryEnv,
+ CachedPlanPrepData *cprep)
{
CachedPlan *plan = NULL;
List *qlist;
@@ -1317,7 +1344,9 @@ GetCachedPlan(CachedPlanSource *plansource, ParamListInfo boundParams,
if (!customplan)
{
- if (CheckCachedPlan(plansource))
+ if (cprep)
+ cprep->params = boundParams;
+ if (CheckCachedPlan(plansource, cprep))
{
/* We want a generic plan, and we already have a valid one */
plan = plansource->gplan;
@@ -1904,11 +1933,13 @@ QueryListGetPrimaryStmt(List *stmts)
}
/*
- * AcquireExecutorLocks: acquire locks needed for execution of a cached plan;
- * or release them if acquire is false.
+ * AcquireExecutorLocksAll: acquire locks needed for execution of a cached
+ * plan; or release them if acquire is false.
+ *
+ * This locks all relations in a given PlannedStmt's range table.
*/
static void
-AcquireExecutorLocks(List *stmt_list, bool acquire)
+AcquireExecutorLocksAll(List *stmt_list, bool acquire)
{
ListCell *lc1;
@@ -1955,6 +1986,212 @@ AcquireExecutorLocks(List *stmt_list, bool acquire)
}
}
+/*
+ * LockRelids
+ * Acquire or release locks on the specified relids, which reference
+ * entries in the provided range table.
+ *
+ * Helper for AcquireExecutorLocksUnpruned().
+ */
+static void
+LockRelids(List *rtable, Bitmapset *relids, bool acquire)
+{
+ int rtindex = -1;
+
+ while ((rtindex = bms_next_member(relids, rtindex)) >= 0)
+ {
+ RangeTblEntry *rte = list_nth_node(RangeTblEntry, rtable, rtindex - 1);
+
+ Assert(rte->rtekind == RTE_RELATION ||
+ (rte->rtekind == RTE_SUBQUERY && OidIsValid(rte->relid)));
+
+ /*
+ * Acquire the appropriate type of lock on each relation OID. Note
+ * that we don't actually try to open the rel, and hence will not
+ * fail if it's been dropped entirely --- we'll just transiently
+ * acquire a non-conflicting lock.
+ */
+ if (acquire)
+ LockRelationOid(rte->relid, rte->rellockmode);
+ else
+ UnlockRelationOid(rte->relid, rte->rellockmode);
+ }
+}
+
+/*
+ * AcquireExecutorLocksUnpruned
+ * Acquire or release execution locks for only unpruned relations
+ * referenced by the given PlannedStmts.
+ *
+ * On acquire, this:
+ * - locks unprunable rels listed in PlannedStmt.unprunableRelids
+ * - runs ExecutorPrep() to perform initial runtime pruning
+ * - locks the surviving partitions reported in the prep estate
+ * - appends the EState pointer for each PlannedStmt to cprep->prep_estates
+ *
+ * On release, it:
+ * - looks up the EState for each PlannedStmt from cprep->prep_estates
+ * (which must already be populated)
+ * - unlocks the same relations identified during acquire
+ *
+ * prep_estates is extended during acquire and must match stmt_list one-to-one
+ * when releasing locks. Memory allocation for EState happens in
+ * cprep->context. Locks are acquired using cprep->owner.
+ */
+static void
+AcquireExecutorLocksUnpruned(List *stmt_list, bool acquire,
+ CachedPlanPrepData *cprep)
+{
+ MemoryContext oldcontext = MemoryContextSwitchTo(cprep->context);
+ ListCell *lc1;
+ List *prep_estates;
+ ListCell *prep_lc;
+
+ Assert(cprep);
+
+ /*
+ * When releasing locks, use the EState list (if any) created during
+ * acquisition to determine which relids to unlock. The list must match
+ * the PlannedStmt list one-to-one.
+ */
+ prep_estates = cprep->prep_estates;
+ Assert(acquire || list_length(prep_estates) == list_length(stmt_list));
+
+ prep_lc = list_head(prep_estates);
+ foreach(lc1, stmt_list)
+ {
+ PlannedStmt *plannedstmt = lfirst_node(PlannedStmt, lc1);
+ EState *prep_estate;
+
+ if (plannedstmt->commandType == CMD_UTILITY)
+ {
+ /* Same as AcquireExecutorLocks(). */
+ Query *query = UtilityContainsQuery(plannedstmt->utilityStmt);
+
+ if (query)
+ ScanQueryForLocks(query, acquire);
+
+ /* Keep the list one-to-one with stmt_list. */
+ if (acquire)
+ cprep->prep_estates = lappend(cprep->prep_estates, NULL);
+ else
+ (void) next_prep_estate(prep_estates, &prep_lc);
+ continue;
+ }
+
+ /*
+ * Lock tables mentioned in the original query and other unprunable
+ * relations that were added to the plan via inheritance expansion.
+ */
+ LockRelids(plannedstmt->rtable, plannedstmt->unprunableRelids, acquire);
+
+ /* Lock partitions surviving runtime initial pruning. */
+ if (acquire)
+ {
+ /*
+ * Pruning expressions may call PL functions that require an active
+ * snapshot (e.g., via EnsurePortalSnapshotExists()). Establish one
+ * if needed.
+ */
+ bool snap_pushed = false;
+
+ if (!ActiveSnapshotSet())
+ {
+ PushActiveSnapshot(GetTransactionSnapshot());
+ snap_pushed = true;
+ }
+
+ prep_estate = ExecutorPrep(plannedstmt, cprep->params,
+ cprep->owner, cprep->eflags);
+ Assert(prep_estate);
+ cprep->prep_estates = lappend(cprep->prep_estates, prep_estate);
+
+ if (snap_pushed)
+ PopActiveSnapshot();
+ }
+ else
+ prep_estate = next_prep_estate(prep_estates, &prep_lc);
+
+ if (prep_estate)
+ {
+ /*
+ * es_unpruned_relids includes plannedstmt->unprunableRelids,
+ * which we've already locked. Filter them out to avoid double-locking.
+ */
+ Bitmapset *lock_relids = bms_difference(prep_estate->es_unpruned_relids,
+ plannedstmt->unprunableRelids);
+
+ /*
+ * We must always include the first result relation of each
+ * ModifyTable node in the plan, that is, the one mentioned in
+ * plannedstmt->firstResultRels in the set of relations to be
+ * locked to satisfy executor assumptions described
+ * in ExecInitModifyTable(). This can be wasteful, because we
+ * may not need to use the first result relation at all if other
+ * result relations are unpruned and thus sufficient for the
+ * ModifyTable node's needs. Unfortunately, we don't have per-node
+ * unpruned_relids set to determine that other result relations
+ * are included.
+ */
+ if (plannedstmt->resultRelations)
+ {
+ ListCell *lc2;
+
+ foreach(lc2, plannedstmt->firstResultRels)
+ {
+ Index firstResultRel = lfirst_int(lc2);
+
+ if (!bms_is_member(firstResultRel, lock_relids))
+ lock_relids = bms_add_member(lock_relids, firstResultRel);
+ }
+ }
+
+ LockRelids(plannedstmt->rtable, lock_relids, acquire);
+ bms_free(lock_relids);
+ }
+ }
+
+ MemoryContextSwitchTo(oldcontext);
+}
+
+/*
+ * CachedPlanPrepCleanup
+ * Clean up EState built for a generic plan.
+ *
+ * This is used in the corner case where CheckCachedPlan() discovers
+ * that a CachedPlan has become invalid after AcquireExecutorLocksUnpruned()
+ * has already run. In that case we must both release the execution locks
+ * and dispose of the ExecPrep list stored in CachedPlanPrepData, since the
+ * executor will never see or clean it up.
+ */
+static void
+CachedPlanPrepCleanup(CachedPlanPrepData *cprep)
+{
+ ListCell *lc;
+ ResourceOwner oldowner;
+
+ if (cprep == NULL)
+ return;
+
+ /* Switch to owner that ExecutorPrep() would have used. */
+ oldowner = CurrentResourceOwner;
+ CurrentResourceOwner = cprep->owner;
+ foreach(lc, cprep->prep_estates)
+ {
+ EState *prep_estate = (EState *) lfirst(lc);
+
+ if (prep_estate == NULL)
+ continue;
+
+ ExecCloseRangeTableRelations(prep_estate);
+ FreeExecutorState(prep_estate);
+ }
+ CurrentResourceOwner = oldowner;
+
+ list_free(cprep->prep_estates);
+ cprep->prep_estates = NIL;
+}
+
/*
* AcquirePlannerLocks: acquire locks needed for planning of a querytree list;
* or release them if acquire is false.
diff --git a/src/include/nodes/pathnodes.h b/src/include/nodes/pathnodes.h
index 27758ec16fe..4fd9d9bcc56 100644
--- a/src/include/nodes/pathnodes.h
+++ b/src/include/nodes/pathnodes.h
@@ -217,6 +217,9 @@ typedef struct PlannerGlobal
/* "flat" list of integer RT indexes */
List *resultRelations;
+ /* "flat" list of integer RT indexes (one per ModifyTable node) */
+ List *firstResultRels;
+
/* "flat" list of AppendRelInfos */
List *appendRelations;
diff --git a/src/include/nodes/plannodes.h b/src/include/nodes/plannodes.h
index b6185825fcb..55279cbbda8 100644
--- a/src/include/nodes/plannodes.h
+++ b/src/include/nodes/plannodes.h
@@ -121,6 +121,16 @@ typedef struct PlannedStmt
/* integer list of RT indexes, or NIL */
List *resultRelations;
+ /*
+ * rtable indexes of first target relation in each ModifyTable node in the
+ * plan for INSERT/UPDATE/DELETE/MERGE. NIL if resultRelations is NIL.
+ *
+ * These are used by AcquireExecutorLocksUnpruned() to ensure that the
+ * first result rel for each ModifyTable remains locked even if pruned;
+ * see ExecInitModifyTable() for the executor side assumptions.
+ */
+ List *firstResultRels;
+
/* list of AppendRelInfo nodes */
List *appendRelations;
diff --git a/src/include/utils/plancache.h b/src/include/utils/plancache.h
index 984c51515c6..c22f832d0b1 100644
--- a/src/include/utils/plancache.h
+++ b/src/include/utils/plancache.h
@@ -197,6 +197,38 @@ typedef struct CachedExpression
} CachedExpression;
+/*
+ * CachedPlanPrepData
+ * Carries ExecutorPrep results for each PlannedStmt in a CachedPlan,
+ * along with context and owner information needed to allocate them.
+ *
+ * prep_estates is indexed one-to-one with CachedPlan->stmt_list, and is
+ * populated when GetCachedPlan() prepares a reused generic plan. If the
+ * plan is found invalid after locking, the same list is used to determine
+ * which relations to unlock before retrying.
+ *
+ * ExecutorPrep state is allocated in 'context' and owned by 'owner'.
+ *
+ * eflags controls ExecutorPrep() behavior during initial pruning.
+ * Normally zero; set EXEC_FLAG_EXPLAIN_GENERIC to suppress pruning
+ * in EXPLAIN (GENERIC_PLAN). Need not match the eflags later passed
+ * to ExecutorStart().
+ *
+ * prep_estates must reach ExecutorStart() to be adopted for execution.
+ * If the plan is invalidated before that happens, CachedPlanPrepCleanup()
+ * frees them instead. The EStates are allocated in 'context' and their
+ * resources tracked under 'owner', which the caller sets to match the
+ * execution environment (e.g., portal context and resowner).
+ */
+typedef struct CachedPlanPrepData
+{
+ List *prep_estates; /* one EState per PlannedStmt, or NULL */
+ ParamListInfo params; /* params visible to ExecutorPrep */
+ MemoryContext context; /* where to allocate EState and its fields */
+ ResourceOwner owner; /* ResourceOwner for ExecutorPrep state */
+ int eflags; /* executor flags to control ExecutorPrep */
+} CachedPlanPrepData;
+
extern void InitPlanCache(void);
extern void ResetPlanCache(void);
@@ -240,7 +272,8 @@ extern List *CachedPlanGetTargetList(CachedPlanSource *plansource,
extern CachedPlan *GetCachedPlan(CachedPlanSource *plansource,
ParamListInfo boundParams,
ResourceOwner owner,
- QueryEnvironment *queryEnv);
+ QueryEnvironment *queryEnv,
+ CachedPlanPrepData *cprep);
extern void ReleaseCachedPlan(CachedPlan *plan, ResourceOwner owner);
extern bool CachedPlanAllowsSimpleValidityCheck(CachedPlanSource *plansource,
diff --git a/src/test/regress/expected/partition_prune.out b/src/test/regress/expected/partition_prune.out
index deacdd75807..8e0cc98baca 100644
--- a/src/test/regress/expected/partition_prune.out
+++ b/src/test/regress/expected/partition_prune.out
@@ -4824,3 +4824,148 @@ select min(a) over (partition by a order by a) from part_abc where a >= stable_o
drop view part_abc_view;
drop table part_abc;
+--
+-- Verify that pruning-aware locking skips pruned partitions
+-- when reusing a generic cached plan.
+--
+set plan_cache_mode to force_generic_plan;
+create table prunelock_p (a int) partition by list (a);
+create table prunelock_p1 partition of prunelock_p for values in (1);
+create table prunelock_p2 partition of prunelock_p for values in (2);
+create table prunelock_p3 partition of prunelock_p for values in (3);
+prepare prunelock_q (int) as select * from prunelock_p where a = $1;
+-- Force generic plan creation
+explain (costs off) execute prunelock_q(1);
+ QUERY PLAN
+----------------------------------------------
+ Append
+ Subplans Removed: 2
+ -> Seq Scan on prunelock_p1 prunelock_p_1
+ Filter: (a = $1)
+(4 rows)
+
+-- Execute and check which child partitions are locked
+begin;
+execute prunelock_q(1);
+ a
+---
+(0 rows)
+
+select c.relname
+ from pg_locks l
+ join pg_class c on c.oid = l.relation
+ where l.pid = pg_backend_pid()
+ and c.relname like 'prunelock_p_'
+ order by c.relname;
+ relname
+--------------
+ prunelock_p1
+(1 row)
+
+commit;
+deallocate prunelock_q;
+-- Turn pruning off
+set enable_partition_pruning to off;
+prepare prunelock_q (int) as select * from prunelock_p where a = $1;
+-- Force generic plan creation
+explain (costs off) execute prunelock_q(1);
+ QUERY PLAN
+----------------------------------------------
+ Append
+ -> Seq Scan on prunelock_p1 prunelock_p_1
+ Filter: (a = $1)
+ -> Seq Scan on prunelock_p2 prunelock_p_2
+ Filter: (a = $1)
+ -> Seq Scan on prunelock_p3 prunelock_p_3
+ Filter: (a = $1)
+(7 rows)
+
+-- Execute and check which child partitions are locked
+begin;
+execute prunelock_q(1);
+ a
+---
+(0 rows)
+
+select c.relname
+ from pg_locks l
+ join pg_class c on c.oid = l.relation
+ where l.pid = pg_backend_pid()
+ and c.relname like 'prunelock_p_'
+ order by c.relname;
+ relname
+--------------
+ prunelock_p1
+ prunelock_p2
+ prunelock_p3
+(3 rows)
+
+commit;
+deallocate prunelock_q;
+reset enable_partition_pruning;
+--
+-- Verify firstResultRels handling with multiple ModifyTable nodes
+-- (writable CTEs) targeting a partitioned table. When a pruning
+-- parameter matches no partition, all result relations are pruned
+-- and the executor must still find a usable first result relation
+-- for each ModifyTable node.
+--
+prepare prunelock_mt_q (int, int) as
+ with upd1 as (update prunelock_p set a = a),
+ upd2 as (update prunelock_p set a = a where a = $2)
+ update prunelock_p set a = a where a = $1;
+-- Force generic plan creation
+explain (costs off) execute prunelock_mt_q(1, 2);
+ QUERY PLAN
+------------------------------------------------------------
+ Update on prunelock_p
+ Update on prunelock_p1 prunelock_p_1
+ CTE upd1
+ -> Update on prunelock_p prunelock_p_3
+ Update on prunelock_p1 prunelock_p_4
+ Update on prunelock_p2 prunelock_p_5
+ Update on prunelock_p3 prunelock_p_6
+ -> Append
+ -> Seq Scan on prunelock_p1 prunelock_p_4
+ -> Seq Scan on prunelock_p2 prunelock_p_5
+ -> Seq Scan on prunelock_p3 prunelock_p_6
+ CTE upd2
+ -> Update on prunelock_p prunelock_p_7
+ Update on prunelock_p2 prunelock_p_8
+ -> Append
+ Subplans Removed: 2
+ -> Seq Scan on prunelock_p2 prunelock_p_8
+ Filter: (a = $2)
+ -> Append
+ Subplans Removed: 2
+ -> Seq Scan on prunelock_p1 prunelock_p_1
+ Filter: (a = $1)
+(22 rows)
+
+-- All partitions pruned: value 4 matches no partition, so each
+-- ModifyTable must still initialize correctly with no matching
+-- result relations.
+explain (costs off) execute prunelock_mt_q(4, 5);
+ QUERY PLAN
+------------------------------------------------------------
+ Update on prunelock_p
+ CTE upd1
+ -> Update on prunelock_p prunelock_p_2
+ Update on prunelock_p1 prunelock_p_3
+ Update on prunelock_p2 prunelock_p_4
+ Update on prunelock_p3 prunelock_p_5
+ -> Append
+ -> Seq Scan on prunelock_p1 prunelock_p_3
+ -> Seq Scan on prunelock_p2 prunelock_p_4
+ -> Seq Scan on prunelock_p3 prunelock_p_5
+ CTE upd2
+ -> Update on prunelock_p prunelock_p_6
+ -> Append
+ Subplans Removed: 3
+ -> Append
+ Subplans Removed: 3
+(16 rows)
+
+deallocate prunelock_mt_q;
+drop table prunelock_p;
+reset plan_cache_mode;
diff --git a/src/test/regress/expected/plancache.out b/src/test/regress/expected/plancache.out
index 4e59188196c..1d69ab0a1c2 100644
--- a/src/test/regress/expected/plancache.out
+++ b/src/test/regress/expected/plancache.out
@@ -398,3 +398,65 @@ select name, generic_plans, custom_plans from pg_prepared_statements
(1 row)
drop table test_mode;
+-- Test invalidation of a generic plan during pruning-aware lock setup.
+-- The pruning expression uses a stable SQL function that calls a volatile
+-- plpgsql function. That function performs DDL on a partition when a
+-- separate "signal" table says to do so. The second EXECUTE should
+-- replan cleanly after the DDL.
+set plan_cache_mode to force_generic_plan;
+create table inval_during_pruning_p (a int) partition by list (a);
+create table inval_during_pruning_p1 partition of inval_during_pruning_p for values in (1);
+create table inval_during_pruning_p2 partition of inval_during_pruning_p for values in (2);
+insert into inval_during_pruning_p values (1), (2);
+create table inval_during_pruning_signal (create_idx bool not null);
+insert into inval_during_pruning_signal values (false);
+create or replace function invalidate_plancache_func() returns int
+as $$
+declare
+ create_index bool;
+begin
+ -- Perform DDL on a partition if asked to
+ select create_idx into create_index from inval_during_pruning_signal for update;
+ if create_index = true then
+ raise notice 'creating index on partition inval_during_pruning_p1';
+ create index on inval_during_pruning_p1 (a);
+ update inval_during_pruning_signal set create_idx = false;
+ end if;
+ -- pruning parameter
+ return 1;
+end;
+$$ language plpgsql volatile;
+create or replace function stable_pruning_val() returns int as $$
+ select invalidate_plancache_func();
+$$ language sql stable;
+prepare inval_during_pruning_q as select * from inval_during_pruning_p where a = stable_pruning_val();
+-- Build a generic plan and run pruning once, but don't set the signal
+-- for invalidate_plancache_func() to perform the DDL.
+explain (verbose, costs off) execute inval_during_pruning_q;
+ QUERY PLAN
+---------------------------------------------------------------------------
+ Append
+ Subplans Removed: 1
+ -> Seq Scan on public.inval_during_pruning_p1 inval_during_pruning_p_1
+ Output: inval_during_pruning_p_1.a
+ Filter: (inval_during_pruning_p_1.a = stable_pruning_val())
+(5 rows)
+
+-- Reuse the generic plan. Make invalidate_plancache_func() perform DDL
+-- during this execution, which should force replanning without errors.
+update inval_during_pruning_signal set create_idx = true;
+explain (verbose, costs off) execute inval_during_pruning_q;
+NOTICE: creating index on partition inval_during_pruning_p1
+ QUERY PLAN
+---------------------------------------------------------------------------
+ Append
+ Subplans Removed: 1
+ -> Seq Scan on public.inval_during_pruning_p1 inval_during_pruning_p_1
+ Output: inval_during_pruning_p_1.a
+ Filter: (inval_during_pruning_p_1.a = stable_pruning_val())
+(5 rows)
+
+drop table inval_during_pruning_p, inval_during_pruning_signal;
+drop function invalidate_plancache_func, stable_pruning_val;
+deallocate inval_during_pruning_q;
+reset plan_cache_mode;
diff --git a/src/test/regress/sql/partition_prune.sql b/src/test/regress/sql/partition_prune.sql
index d93c0c03bab..804dd3c8f4e 100644
--- a/src/test/regress/sql/partition_prune.sql
+++ b/src/test/regress/sql/partition_prune.sql
@@ -1447,3 +1447,80 @@ select min(a) over (partition by a order by a) from part_abc where a >= stable_o
drop view part_abc_view;
drop table part_abc;
+
+--
+-- Verify that pruning-aware locking skips pruned partitions
+-- when reusing a generic cached plan.
+--
+set plan_cache_mode to force_generic_plan;
+
+create table prunelock_p (a int) partition by list (a);
+create table prunelock_p1 partition of prunelock_p for values in (1);
+create table prunelock_p2 partition of prunelock_p for values in (2);
+create table prunelock_p3 partition of prunelock_p for values in (3);
+
+prepare prunelock_q (int) as select * from prunelock_p where a = $1;
+
+-- Force generic plan creation
+explain (costs off) execute prunelock_q(1);
+
+-- Execute and check which child partitions are locked
+begin;
+execute prunelock_q(1);
+
+select c.relname
+ from pg_locks l
+ join pg_class c on c.oid = l.relation
+ where l.pid = pg_backend_pid()
+ and c.relname like 'prunelock_p_'
+ order by c.relname;
+commit;
+
+deallocate prunelock_q;
+
+-- Turn pruning off
+set enable_partition_pruning to off;
+
+prepare prunelock_q (int) as select * from prunelock_p where a = $1;
+
+-- Force generic plan creation
+explain (costs off) execute prunelock_q(1);
+
+-- Execute and check which child partitions are locked
+begin;
+execute prunelock_q(1);
+
+select c.relname
+ from pg_locks l
+ join pg_class c on c.oid = l.relation
+ where l.pid = pg_backend_pid()
+ and c.relname like 'prunelock_p_'
+ order by c.relname;
+commit;
+
+deallocate prunelock_q;
+reset enable_partition_pruning;
+
+--
+-- Verify firstResultRels handling with multiple ModifyTable nodes
+-- (writable CTEs) targeting a partitioned table. When a pruning
+-- parameter matches no partition, all result relations are pruned
+-- and the executor must still find a usable first result relation
+-- for each ModifyTable node.
+--
+prepare prunelock_mt_q (int, int) as
+ with upd1 as (update prunelock_p set a = a),
+ upd2 as (update prunelock_p set a = a where a = $2)
+ update prunelock_p set a = a where a = $1;
+
+-- Force generic plan creation
+explain (costs off) execute prunelock_mt_q(1, 2);
+
+-- All partitions pruned: value 4 matches no partition, so each
+-- ModifyTable must still initialize correctly with no matching
+-- result relations.
+explain (costs off) execute prunelock_mt_q(4, 5);
+
+deallocate prunelock_mt_q;
+drop table prunelock_p;
+reset plan_cache_mode;
diff --git a/src/test/regress/sql/plancache.sql b/src/test/regress/sql/plancache.sql
index 4b2f11dcc64..139b4688fd6 100644
--- a/src/test/regress/sql/plancache.sql
+++ b/src/test/regress/sql/plancache.sql
@@ -223,3 +223,54 @@ select name, generic_plans, custom_plans from pg_prepared_statements
where name = 'test_mode_pp';
drop table test_mode;
+
+-- Test invalidation of a generic plan during pruning-aware lock setup.
+-- The pruning expression uses a stable SQL function that calls a volatile
+-- plpgsql function. That function performs DDL on a partition when a
+-- separate "signal" table says to do so. The second EXECUTE should
+-- replan cleanly after the DDL.
+set plan_cache_mode to force_generic_plan;
+create table inval_during_pruning_p (a int) partition by list (a);
+create table inval_during_pruning_p1 partition of inval_during_pruning_p for values in (1);
+create table inval_during_pruning_p2 partition of inval_during_pruning_p for values in (2);
+insert into inval_during_pruning_p values (1), (2);
+
+create table inval_during_pruning_signal (create_idx bool not null);
+insert into inval_during_pruning_signal values (false);
+create or replace function invalidate_plancache_func() returns int
+as $$
+declare
+ create_index bool;
+begin
+ -- Perform DDL on a partition if asked to
+ select create_idx into create_index from inval_during_pruning_signal for update;
+ if create_index = true then
+ raise notice 'creating index on partition inval_during_pruning_p1';
+ create index on inval_during_pruning_p1 (a);
+ update inval_during_pruning_signal set create_idx = false;
+ end if;
+ -- pruning parameter
+ return 1;
+end;
+$$ language plpgsql volatile;
+
+create or replace function stable_pruning_val() returns int as $$
+ select invalidate_plancache_func();
+$$ language sql stable;
+
+prepare inval_during_pruning_q as select * from inval_during_pruning_p where a = stable_pruning_val();
+
+-- Build a generic plan and run pruning once, but don't set the signal
+-- for invalidate_plancache_func() to perform the DDL.
+explain (verbose, costs off) execute inval_during_pruning_q;
+
+-- Reuse the generic plan. Make invalidate_plancache_func() perform DDL
+-- during this execution, which should force replanning without errors.
+update inval_during_pruning_signal set create_idx = true;
+explain (verbose, costs off) execute inval_during_pruning_q;
+
+drop table inval_during_pruning_p, inval_during_pruning_signal;
+drop function invalidate_plancache_func, stable_pruning_val;
+deallocate inval_during_pruning_q;
+
+reset plan_cache_mode;
--
2.47.3
[application/octet-stream] v8-0004-Make-SQL-function-executor-track-ExecutorPrep-sta.patch (7.8K, 4-v8-0004-Make-SQL-function-executor-track-ExecutorPrep-sta.patch)
download | inline diff:
From 2ab5fefb9644118a1f1528a53b9a6af90e063edb Mon Sep 17 00:00:00 2001
From: Amit Langote <[email protected]>
Date: Tue, 10 Feb 2026 22:09:23 +0900
Subject: [PATCH v8 4/5] Make SQL function executor track ExecutorPrep state
Extend the SQL function executor to use the ExecutorPrep results
returned by GetCachedPlan(). init_execution_state() now passes a
CachedPlanPrepData to GetCachedPlan() and stores the per statement
ExecPrep pointers in the execution_state nodes.
At execution time, postquel_start() reparents the prep estate's
es_query_cxt under the function's subcontext so that prep state
follows the usual per call context hierarchy.
This allows SQL language functions to participate in the same
ExecutorPrep machinery as other plan cache users.
Add a regression test where rule rewrite expands a single UPDATE
into multiple PlannedStmts, exercising the SQL function plan cache
and the generic plan reuse path that now invokes ExecutorPrep.
---
src/backend/executor/functions.c | 29 +++++++++++++--
src/test/regress/expected/plancache.out | 48 +++++++++++++++++++++++++
src/test/regress/sql/plancache.sql | 34 ++++++++++++++++++
3 files changed, 109 insertions(+), 2 deletions(-)
diff --git a/src/backend/executor/functions.c b/src/backend/executor/functions.c
index c0ca72b38dd..f246f051c25 100644
--- a/src/backend/executor/functions.c
+++ b/src/backend/executor/functions.c
@@ -73,6 +73,7 @@ typedef struct execution_state
bool setsResult; /* true if this query produces func's result */
bool lazyEval; /* true if should fetch one row at a time */
PlannedStmt *stmt; /* plan for this query */
+ EState *prep_estate; /* EState created in ExecutorPrep() for this plan */
QueryDesc *qd; /* null unless status == RUN */
} execution_state;
@@ -658,6 +659,8 @@ init_execution_state(SQLFunctionCachePtr fcache)
execution_state *lasttages = NULL;
int nstmts;
ListCell *lc;
+ CachedPlanPrepData cprep = {0};
+ ListCell *prep_lc;
/*
* Clean up after previous query, if there was one.
@@ -696,11 +699,20 @@ init_execution_state(SQLFunctionCachePtr fcache)
* CurrentResourceOwner will be the same when ShutdownSQLFunction runs.)
*/
fcache->cowner = CurrentResourceOwner;
+
+ /*
+ * Have ExecutorPrep() allocate under fcache->fcontext. The prep
+ * EStates it creates will initially live there; postquel_start()
+ * will later reparent their es_query_cxt into fcache->subcontext
+ * when using them for execution.
+ */
+ cprep.context = fcache->fcontext;
+ cprep.owner = fcache->cowner;
fcache->cplan = GetCachedPlan(plansource,
fcache->paramLI,
fcache->cowner,
NULL,
- NULL);
+ &cprep);
/*
* If necessary, make esarray[] bigger to hold the needed state.
@@ -721,9 +733,11 @@ init_execution_state(SQLFunctionCachePtr fcache)
/*
* Build execution_state list to match the number of contained plans.
*/
+ prep_lc = list_head(cprep.prep_estates);
foreach(lc, fcache->cplan->stmt_list)
{
PlannedStmt *stmt = lfirst_node(PlannedStmt, lc);
+ EState *prep_estate = next_prep_estate(cprep.prep_estates, &prep_lc);
execution_state *newes;
/*
@@ -765,6 +779,7 @@ init_execution_state(SQLFunctionCachePtr fcache)
newes->setsResult = false; /* might change below */
newes->lazyEval = false; /* might change below */
newes->stmt = stmt;
+ newes->prep_estate = prep_estate;
newes->qd = NULL;
if (stmt->canSetTag)
@@ -1363,6 +1378,15 @@ postquel_start(execution_state *es, SQLFunctionCachePtr fcache)
else
dest = None_Receiver;
+ /*
+ * Prep EStates were built under fcache->fcontext. For execution,
+ * make their es_query_cxt a child of fcache->subcontext so they
+ * follow the usual per call lifetime.
+ */
+ if (es->prep_estate)
+ MemoryContextSetParent(es->prep_estate->es_query_cxt,
+ fcache->subcontext);
+
es->qd = CreateQueryDesc(es->stmt,
fcache->func->src,
GetActiveSnapshot(),
@@ -1371,7 +1395,7 @@ postquel_start(execution_state *es, SQLFunctionCachePtr fcache)
fcache->paramLI,
es->qd ? es->qd->queryEnv : NULL,
0,
- NULL);
+ es->prep_estate);
/* Utility commands don't need Executor. */
if (es->qd->operation != CMD_UTILITY)
@@ -1462,6 +1486,7 @@ postquel_end(execution_state *es, SQLFunctionCachePtr fcache)
FreeQueryDesc(es->qd);
es->qd = NULL;
+ es->prep_estate = NULL;
MemoryContextSwitchTo(oldcontext);
diff --git a/src/test/regress/expected/plancache.out b/src/test/regress/expected/plancache.out
index 1d69ab0a1c2..371673a6e96 100644
--- a/src/test/regress/expected/plancache.out
+++ b/src/test/regress/expected/plancache.out
@@ -459,4 +459,52 @@ NOTICE: creating index on partition inval_during_pruning_p1
drop table inval_during_pruning_p, inval_during_pruning_signal;
drop function invalidate_plancache_func, stable_pruning_val;
deallocate inval_during_pruning_q;
+-- exercise sql-function plan cache when rewrite expands a single statement
+-- into multiple planned statements. this forces cachedplan->stmt_list to
+-- contain more than one entry and checks that executor state for the first
+-- rewritten statement does not destroy state needed by the second one.
+set plan_cache_mode = force_generic_plan;
+create table sqlf_base(id int, val int) partition by list (id);
+create table sqlf_base_1 partition of sqlf_base for values in (1);
+create table sqlf_base_2 partition of sqlf_base for values in (2);
+create table sqlf_log(id int, note text);
+insert into sqlf_base values (1, 10);
+create rule sqlf_base_upd_log as
+on update to sqlf_base do also
+ insert into sqlf_log(id, note)
+ values (new.id, 'logged by rule');
+create or replace function sqlf_execprep_test(a int, v int)
+returns void
+language sql
+as $$
+ update sqlf_base set val = v where id = a;
+$$;
+select sqlf_execprep_test(1, 20);
+ sqlf_execprep_test
+--------------------
+
+(1 row)
+
+select sqlf_execprep_test(1, 30);
+ sqlf_execprep_test
+--------------------
+
+(1 row)
+
+select * from sqlf_base order by 1;
+ id | val
+----+-----
+ 1 | 30
+(1 row)
+
+select * from sqlf_log order by 1;
+ id | note
+----+----------------
+ 1 | logged by rule
+ 1 | logged by rule
+(2 rows)
+
+drop rule sqlf_base_upd_log on sqlf_base;
+drop table sqlf_base, sqlf_log;
+drop function sqlf_execprep_test;
reset plan_cache_mode;
diff --git a/src/test/regress/sql/plancache.sql b/src/test/regress/sql/plancache.sql
index 139b4688fd6..b89c9ad69a4 100644
--- a/src/test/regress/sql/plancache.sql
+++ b/src/test/regress/sql/plancache.sql
@@ -273,4 +273,38 @@ drop table inval_during_pruning_p, inval_during_pruning_signal;
drop function invalidate_plancache_func, stable_pruning_val;
deallocate inval_during_pruning_q;
+-- exercise sql-function plan cache when rewrite expands a single statement
+-- into multiple planned statements. this forces cachedplan->stmt_list to
+-- contain more than one entry and checks that executor state for the first
+-- rewritten statement does not destroy state needed by the second one.
+
+set plan_cache_mode = force_generic_plan;
+
+create table sqlf_base(id int, val int) partition by list (id);
+create table sqlf_base_1 partition of sqlf_base for values in (1);
+create table sqlf_base_2 partition of sqlf_base for values in (2);
+create table sqlf_log(id int, note text);
+
+insert into sqlf_base values (1, 10);
+
+create rule sqlf_base_upd_log as
+on update to sqlf_base do also
+ insert into sqlf_log(id, note)
+ values (new.id, 'logged by rule');
+
+create or replace function sqlf_execprep_test(a int, v int)
+returns void
+language sql
+as $$
+ update sqlf_base set val = v where id = a;
+$$;
+
+select sqlf_execprep_test(1, 20);
+select sqlf_execprep_test(1, 30);
+select * from sqlf_base order by 1;
+select * from sqlf_log order by 1;
+
+drop rule sqlf_base_upd_log on sqlf_base;
+drop table sqlf_base, sqlf_log;
+drop function sqlf_execprep_test;
reset plan_cache_mode;
--
2.47.3
[application/octet-stream] v8-0002-Introduce-ExecutorPrep-and-refactor-executor-star.patch (27.1K, 5-v8-0002-Introduce-ExecutorPrep-and-refactor-executor-star.patch)
download | inline diff:
From a2a0befc44d25df8b549644a7e179923270a0fc6 Mon Sep 17 00:00:00 2001
From: Amit Langote <[email protected]>
Date: Tue, 11 Nov 2025 21:47:46 +0900
Subject: [PATCH v8 2/5] Introduce ExecutorPrep and refactor executor startup
Factor permission checks, range table initialization, and initial
partition pruning out of InitPlan() into a new ExecutorPrep()
helper. ExecutorPrep() builds an EState containing the executor
metadata needed before plan execution, including partition
pruning state where partPruneInfos are present, and returns it
directly to the caller.
ExecutorStart() now checks if QueryDesc->estate is already set
(indicating ExecutorPrep() was called earlier). If so, it reuses
the EState to avoid redoing range table setup and pruning.
Otherwise, it invokes ExecutorPrep() itself and adopts the
resulting EState for the duration of the query. This keeps the
executor startup behavior unchanged while making the setup work
callable separately when needed.
CreateQueryDesc() grows a prep_estate argument to accept an
optionally pre-created EState and stores it in the QueryDesc.
Portals, SPI, SQL functions, and EXPLAIN are wired to carry
optional EState pointers alongside the PlannedStmt list, but most
callers still pass NULL and let ExecutorStart() perform the setup
lazily.
ExecutorPrep() requires the caller to have established an active
snapshot, as partition pruning expressions may call PL functions
that internally require one (e.g., via EnsurePortalSnapshotExists()).
Update executor/README and related comments to document the new
control flow and the separation between preparation and execution.
Note that as of this commit, ExecutorStart() is the only caller of
ExecutorPrep(), so there is no semantic change in behavior. Later
commits will add specialized callers that invoke ExecutorPrep()
earlier to enable pruning-aware locking in cached plans.
---
src/backend/commands/copyto.c | 2 +-
src/backend/commands/createas.c | 2 +-
src/backend/commands/explain.c | 8 +-
src/backend/commands/extension.c | 2 +-
src/backend/commands/matview.c | 2 +-
src/backend/commands/portalcmds.c | 1 +
src/backend/commands/prepare.c | 9 +-
src/backend/executor/README | 11 +-
src/backend/executor/execMain.c | 164 +++++++++++++++++++++++-----
src/backend/executor/execParallel.c | 3 +-
src/backend/executor/functions.c | 3 +-
src/backend/executor/spi.c | 9 +-
src/backend/tcop/postgres.c | 2 +
src/backend/tcop/pquery.c | 24 +++-
src/backend/utils/mmgr/portalmem.c | 2 +
src/include/commands/explain.h | 3 +-
src/include/executor/execdesc.h | 5 +-
src/include/executor/executor.h | 26 +++++
src/include/nodes/execnodes.h | 1 -
src/include/utils/portal.h | 2 +
20 files changed, 229 insertions(+), 52 deletions(-)
diff --git a/src/backend/commands/copyto.c b/src/backend/commands/copyto.c
index 499ce9ad3db..e09303491d2 100644
--- a/src/backend/commands/copyto.c
+++ b/src/backend/commands/copyto.c
@@ -877,7 +877,7 @@ BeginCopyTo(ParseState *pstate,
cstate->queryDesc = CreateQueryDesc(plan, pstate->p_sourcetext,
GetActiveSnapshot(),
InvalidSnapshot,
- dest, NULL, NULL, 0);
+ dest, NULL, NULL, 0, NULL);
/*
* Call ExecutorStart to prepare the plan for execution.
diff --git a/src/backend/commands/createas.c b/src/backend/commands/createas.c
index 270e9bf3110..b4a9808955a 100644
--- a/src/backend/commands/createas.c
+++ b/src/backend/commands/createas.c
@@ -336,7 +336,7 @@ ExecCreateTableAs(ParseState *pstate, CreateTableAsStmt *stmt,
/* Create a QueryDesc, redirecting output to our tuple receiver */
queryDesc = CreateQueryDesc(plan, pstate->p_sourcetext,
GetActiveSnapshot(), InvalidSnapshot,
- dest, params, queryEnv, 0);
+ dest, params, queryEnv, 0, NULL);
/* call ExecutorStart to prepare the plan for execution */
ExecutorStart(queryDesc, GetIntoRelEFlags(into));
diff --git a/src/backend/commands/explain.c b/src/backend/commands/explain.c
index 296ea8a1ed2..02027c429e1 100644
--- a/src/backend/commands/explain.c
+++ b/src/backend/commands/explain.c
@@ -372,7 +372,7 @@ standard_ExplainOneQuery(Query *query, int cursorOptions,
}
/* run it (if needed) and produce output */
- ExplainOnePlan(plan, into, es, queryString, params, queryEnv,
+ ExplainOnePlan(plan, NULL, into, es, queryString, params, queryEnv,
&planduration, (es->buffers ? &bufusage : NULL),
es->memory ? &mem_counters : NULL);
}
@@ -494,7 +494,8 @@ ExplainOneUtility(Node *utilityStmt, IntoClause *into, ExplainState *es,
* to call it.
*/
void
-ExplainOnePlan(PlannedStmt *plannedstmt, IntoClause *into, ExplainState *es,
+ExplainOnePlan(PlannedStmt *plannedstmt, EState *prep_estate,
+ IntoClause *into, ExplainState *es,
const char *queryString, ParamListInfo params,
QueryEnvironment *queryEnv, const instr_time *planduration,
const BufferUsage *bufusage,
@@ -552,7 +553,8 @@ ExplainOnePlan(PlannedStmt *plannedstmt, IntoClause *into, ExplainState *es,
/* Create a QueryDesc for the query */
queryDesc = CreateQueryDesc(plannedstmt, queryString,
GetActiveSnapshot(), InvalidSnapshot,
- dest, params, queryEnv, instrument_option);
+ dest, params, queryEnv, instrument_option,
+ prep_estate);
/* Select execution options */
if (es->analyze)
diff --git a/src/backend/commands/extension.c b/src/backend/commands/extension.c
index b98801d08f2..939e7a632f0 100644
--- a/src/backend/commands/extension.c
+++ b/src/backend/commands/extension.c
@@ -1174,7 +1174,7 @@ execute_sql_string(const char *sql, const char *filename)
qdesc = CreateQueryDesc(stmt,
sql,
GetActiveSnapshot(), NULL,
- dest, NULL, NULL, 0);
+ dest, NULL, NULL, 0, NULL);
ExecutorStart(qdesc, 0);
ExecutorRun(qdesc, ForwardScanDirection, 0);
diff --git a/src/backend/commands/matview.c b/src/backend/commands/matview.c
index 81a55a33ef2..2cdfdcf984b 100644
--- a/src/backend/commands/matview.c
+++ b/src/backend/commands/matview.c
@@ -439,7 +439,7 @@ refresh_matview_datafill(DestReceiver *dest, Query *query,
/* Create a QueryDesc, redirecting output to our tuple receiver */
queryDesc = CreateQueryDesc(plan, queryString,
GetActiveSnapshot(), InvalidSnapshot,
- dest, NULL, NULL, 0);
+ dest, NULL, NULL, 0, NULL);
/* call ExecutorStart to prepare the plan for execution */
ExecutorStart(queryDesc, 0);
diff --git a/src/backend/commands/portalcmds.c b/src/backend/commands/portalcmds.c
index 01efac3319e..1e880a6d7c9 100644
--- a/src/backend/commands/portalcmds.c
+++ b/src/backend/commands/portalcmds.c
@@ -118,6 +118,7 @@ PerformCursorOpen(ParseState *pstate, DeclareCursorStmt *cstmt, ParamListInfo pa
queryString,
CMDTAG_SELECT, /* cursor's query is always a SELECT */
list_make1(plan),
+ NIL,
NULL);
/*----------
diff --git a/src/backend/commands/prepare.c b/src/backend/commands/prepare.c
index 876aad2100a..c7bab14b633 100644
--- a/src/backend/commands/prepare.c
+++ b/src/backend/commands/prepare.c
@@ -207,6 +207,7 @@ ExecuteQuery(ParseState *pstate,
query_string,
entry->plansource->commandTag,
plan_list,
+ NIL,
cplan);
/*
@@ -577,7 +578,9 @@ ExplainExecuteQuery(ExecuteStmt *execstmt, IntoClause *into, ExplainState *es,
const char *query_string;
CachedPlan *cplan;
List *plan_list;
+ List *prep_estates;
ListCell *p;
+ ListCell *prep_lc;
ParamListInfo paramLI = NULL;
EState *estate = NULL;
instr_time planstart;
@@ -652,14 +655,18 @@ ExplainExecuteQuery(ExecuteStmt *execstmt, IntoClause *into, ExplainState *es,
}
plan_list = cplan->stmt_list;
+ prep_estates = NIL;
/* Explain each query */
+ prep_lc = list_head(prep_estates);
foreach(p, plan_list)
{
PlannedStmt *pstmt = lfirst_node(PlannedStmt, p);
+ EState *prep_estate = next_prep_estate(prep_estates, &prep_lc);
if (pstmt->commandType != CMD_UTILITY)
- ExplainOnePlan(pstmt, into, es, query_string, paramLI, pstate->p_queryEnv,
+ ExplainOnePlan(pstmt, prep_estate,
+ into, es, query_string, paramLI, pstate->p_queryEnv,
&planduration, (es->buffers ? &bufusage : NULL),
es->memory ? &mem_counters : NULL);
else
diff --git a/src/backend/executor/README b/src/backend/executor/README
index 54f4782f31b..d749ceb6687 100644
--- a/src/backend/executor/README
+++ b/src/backend/executor/README
@@ -291,11 +291,18 @@ Query Processing Control Flow
This is a sketch of control flow for full query processing:
+ ExecutorPrep
+ May be run before ExecutorStart (e.g., for plan validation), or
+ implicitly from ExecutorStart if not done earlier. Creates EState,
+ performs range table initialization, permission checks, and initial
+ partition pruning. Returns the EState that ExecutorStart() should
+ reuse.
+
CreateQueryDesc
ExecutorStart
- CreateExecutorState
- creates per-query context
+ ExecutorPrep (if not already done, indicated by NULL QueryDesc.estate)
+ creates EState and per-query context
switch to per-query context to run ExecInitNode
AfterTriggerBeginQuery
ExecInitNode --- recursively scans plan tree
diff --git a/src/backend/executor/execMain.c b/src/backend/executor/execMain.c
index c58a2abe9a7..0f95ad88497 100644
--- a/src/backend/executor/execMain.c
+++ b/src/backend/executor/execMain.c
@@ -57,6 +57,7 @@
#include "parser/parse_relation.h"
#include "pgstat.h"
#include "rewrite/rewriteHandler.h"
+#include "storage/lmgr.h"
#include "tcop/utility.h"
#include "utils/acl.h"
#include "utils/backend_status.h"
@@ -147,7 +148,6 @@ standard_ExecutorStart(QueryDesc *queryDesc, int eflags)
/* sanity checks: queryDesc must not be started already */
Assert(queryDesc != NULL);
- Assert(queryDesc->estate == NULL);
/* caller must ensure the query's snapshot is active */
Assert(GetActiveSnapshot() == queryDesc->snapshot);
@@ -173,9 +173,70 @@ standard_ExecutorStart(QueryDesc *queryDesc, int eflags)
/*
* Build EState, switch into per-query memory context for startup.
- */
- estate = CreateExecutorState();
- queryDesc->estate = estate;
+ *
+ * If ExecutorPrep() ran earlier (e.g., to do initial pruning during plan
+ * validity checking), reuse its EState to avoid redoing range table setup
+ * and pruning. Otherwise, create a fresh EState as usual.
+ *
+ * In assert builds, verify that the expected locks are held. When
+ * no prep EState was provided, AcquireExecutorLocks() should have
+ * locked every relation in the plan. When one was provided,
+ * pruning-aware locking should have locked at least the unpruned
+ * relations. Both checks are skipped in parallel workers, which
+ * acquire relation locks lazily in ExecGetRangeTableRelation().
+ */
+ if (queryDesc->estate == NULL)
+ {
+#ifdef USE_ASSERT_CHECKING
+ if (!IsParallelWorker())
+ {
+ ListCell *lc;
+
+ foreach(lc, queryDesc->plannedstmt->rtable)
+ {
+ RangeTblEntry *rte = lfirst_node(RangeTblEntry, lc);
+
+ if (rte->rtekind == RTE_RELATION ||
+ (rte->rtekind == RTE_SUBQUERY && rte->relid != InvalidOid))
+ Assert(CheckRelationOidLockedByMe(rte->relid,
+ rte->rellockmode,
+ true));
+ }
+ }
+#endif
+ queryDesc->estate = ExecutorPrep(queryDesc->plannedstmt,
+ queryDesc->params,
+ CurrentResourceOwner,
+ eflags);
+ }
+#ifdef USE_ASSERT_CHECKING
+ else
+ {
+ /*
+ * A prep EState was provided, meaning pruning-aware locking
+ * should have locked at least the unpruned relations.
+ */
+ if (!IsParallelWorker())
+ {
+ int rtindex = -1;
+
+ while ((rtindex = bms_next_member(queryDesc->estate->es_unpruned_relids,
+ rtindex)) >= 0)
+ {
+ RangeTblEntry *rte = exec_rt_fetch(rtindex, queryDesc->estate);
+
+ Assert(rte->rtekind == RTE_RELATION ||
+ (rte->rtekind == RTE_SUBQUERY &&
+ rte->relid != InvalidOid));
+ Assert(CheckRelationOidLockedByMe(rte->relid,
+ rte->rellockmode, true));
+ }
+ }
+ }
+#endif
+
+ estate = queryDesc->estate;
+ Assert(estate);
oldcontext = MemoryContextSwitchTo(estate->es_query_cxt);
@@ -265,6 +326,73 @@ standard_ExecutorStart(QueryDesc *queryDesc, int eflags)
MemoryContextSwitchTo(oldcontext);
}
+/*
+ * ExecutorPrep: prepare executor state for a PlannedStmt outside ExecutorStart.
+ *
+ * Performs range table initialization, permission checks, and initial
+ * partition pruning if partPruneInfos are present.
+ *
+ * Returns an EState that the caller must either pass to ExecutorStart()
+ * for reuse or free via FreeExecutorState() if execution will not proceed.
+ * GetCachedPlan() uses this to determine which partitions to lock after
+ * pruning; if the resulting EState is not delivered to ExecutorStart(),
+ * the executor would operate on unlocked relations.
+ */
+EState *
+ExecutorPrep(PlannedStmt *pstmt, ParamListInfo params, ResourceOwner owner,
+ int eflags)
+{
+ ResourceOwner oldowner;
+ EState *estate;
+
+ if (pstmt->commandType == CMD_UTILITY)
+ return NULL;
+
+ /* Caller must have established an active snapshot. */
+ Assert(ActiveSnapshotSet());
+
+ estate = CreateExecutorState();
+ estate->es_plannedstmt = pstmt;
+ estate->es_part_prune_infos = pstmt->partPruneInfos;
+ estate->es_param_list_info = params;
+ estate->es_top_eflags = eflags;
+
+ /*
+ * Do permissions checks.
+ */
+ ExecCheckPermissions(pstmt->rtable, pstmt->permInfos, true);
+
+ /*
+ * Initialize range table.
+ */
+ ExecInitRangeTable(estate, pstmt->rtable, pstmt->permInfos,
+ bms_copy(pstmt->unprunableRelids));
+
+ /*
+ * Ensure locks taken during initial pruning are tracked under the given
+ * ResourceOwner (e.g., one associated with CachedPlan validation).
+ */
+ oldowner = CurrentResourceOwner;
+ CurrentResourceOwner = owner;
+
+ /*
+ * Set up PartitionPruneState structures needed for both initial and
+ * runtime partition pruning. These structures are built from the
+ * PartitionPruneInfo entries in the plan tree.
+ *
+ * Also perform initial pruning to compute the subset of child subplans
+ * that will be executed. The results, which are bitmapsets of selected
+ * child indexes, are saved in es_part_prune_results. This list is parallel
+ * to es_part_prune_infos.
+ */
+ ExecCreatePartitionPruneStates(estate);
+ ExecDoInitialPruning(estate);
+
+ CurrentResourceOwner = oldowner;
+
+ return estate;
+}
+
/* ----------------------------------------------------------------
* ExecutorRun
*
@@ -840,38 +968,14 @@ InitPlan(QueryDesc *queryDesc, int eflags)
CmdType operation = queryDesc->operation;
PlannedStmt *plannedstmt = queryDesc->plannedstmt;
Plan *plan = plannedstmt->planTree;
- List *rangeTable = plannedstmt->rtable;
EState *estate = queryDesc->estate;
PlanState *planstate;
TupleDesc tupType;
ListCell *l;
int i;
- /*
- * Do permissions checks
- */
- ExecCheckPermissions(rangeTable, plannedstmt->permInfos, true);
-
- /*
- * initialize the node's execution state
- */
- ExecInitRangeTable(estate, rangeTable, plannedstmt->permInfos,
- bms_copy(plannedstmt->unprunableRelids));
-
- estate->es_plannedstmt = plannedstmt;
- estate->es_part_prune_infos = plannedstmt->partPruneInfos;
-
- /*
- * Perform runtime "initial" pruning to identify which child subplans,
- * corresponding to the children of plan nodes that contain
- * PartitionPruneInfo such as Append, will not be executed. The results,
- * which are bitmapsets of indexes of the child subplans that will be
- * executed, are saved in es_part_prune_results. These results correspond
- * to each PartitionPruneInfo entry, and the es_part_prune_results list is
- * parallel to es_part_prune_infos.
- */
- ExecCreatePartitionPruneStates(estate);
- ExecDoInitialPruning(estate);
+ /* ExecutorPrep() must have been done. */
+ Assert(queryDesc->estate);
/*
* Next, build the ExecRowMark array from the PlanRowMark(s), if any.
diff --git a/src/backend/executor/execParallel.c b/src/backend/executor/execParallel.c
index ac84af294c9..024780d3516 100644
--- a/src/backend/executor/execParallel.c
+++ b/src/backend/executor/execParallel.c
@@ -1300,7 +1300,8 @@ ExecParallelGetQueryDesc(shm_toc *toc, DestReceiver *receiver,
return CreateQueryDesc(pstmt,
queryString,
GetActiveSnapshot(), InvalidSnapshot,
- receiver, paramLI, NULL, instrument_options);
+ receiver, paramLI, NULL, instrument_options,
+ NULL);
}
/*
diff --git a/src/backend/executor/functions.c b/src/backend/executor/functions.c
index 88109348817..952a784c924 100644
--- a/src/backend/executor/functions.c
+++ b/src/backend/executor/functions.c
@@ -1369,7 +1369,8 @@ postquel_start(execution_state *es, SQLFunctionCachePtr fcache)
dest,
fcache->paramLI,
es->qd ? es->qd->queryEnv : NULL,
- 0);
+ 0,
+ NULL);
/* Utility commands don't need Executor. */
if (es->qd->operation != CMD_UTILITY)
diff --git a/src/backend/executor/spi.c b/src/backend/executor/spi.c
index 52f3b11301c..380bbc44e97 100644
--- a/src/backend/executor/spi.c
+++ b/src/backend/executor/spi.c
@@ -1686,6 +1686,7 @@ SPI_cursor_open_internal(const char *name, SPIPlanPtr plan,
query_string,
plansource->commandTag,
stmt_list,
+ NIL,
cplan);
/*
@@ -2500,6 +2501,8 @@ _SPI_execute_plan(SPIPlanPtr plan, const SPIExecuteOptions *options,
CachedPlanSource *plansource = (CachedPlanSource *) lfirst(lc1);
List *stmt_list;
ListCell *lc2;
+ List *prep_estates;
+ ListCell *prep_lc;
spicallbackarg.query = plansource->query_string;
@@ -2578,6 +2581,7 @@ _SPI_execute_plan(SPIPlanPtr plan, const SPIExecuteOptions *options,
plan_owner, _SPI_current->queryEnv);
stmt_list = cplan->stmt_list;
+ prep_estates = NIL;
/*
* If we weren't given a specific snapshot to use, and the statement
@@ -2615,9 +2619,11 @@ _SPI_execute_plan(SPIPlanPtr plan, const SPIExecuteOptions *options,
}
}
+ prep_lc = list_head(prep_estates);
foreach(lc2, stmt_list)
{
PlannedStmt *stmt = lfirst_node(PlannedStmt, lc2);
+ EState *prep_estate = next_prep_estate(prep_estates, &prep_lc);
bool canSetTag = stmt->canSetTag;
DestReceiver *dest;
@@ -2695,7 +2701,8 @@ _SPI_execute_plan(SPIPlanPtr plan, const SPIExecuteOptions *options,
dest,
options->params,
_SPI_current->queryEnv,
- 0);
+ 0,
+ prep_estate);
res = _SPI_pquery(qdesc, fire_triggers,
canSetTag ? options->tcount : 0);
FreeQueryDesc(qdesc);
diff --git a/src/backend/tcop/postgres.c b/src/backend/tcop/postgres.c
index b3563113219..355a490cde9 100644
--- a/src/backend/tcop/postgres.c
+++ b/src/backend/tcop/postgres.c
@@ -1231,6 +1231,7 @@ exec_simple_query(const char *query_string)
query_string,
commandTag,
plantree_list,
+ NIL,
NULL);
/*
@@ -2030,6 +2031,7 @@ exec_bind_message(StringInfo input_message)
query_string,
psrc->commandTag,
cplan->stmt_list,
+ NIL,
cplan);
/* Portal is defined, set the plan ID based on its contents. */
diff --git a/src/backend/tcop/pquery.c b/src/backend/tcop/pquery.c
index d8fc75d0bb9..b18266487bb 100644
--- a/src/backend/tcop/pquery.c
+++ b/src/backend/tcop/pquery.c
@@ -37,6 +37,7 @@ Portal ActivePortal = NULL;
static void ProcessQuery(PlannedStmt *plan,
+ EState *prep_estate,
const char *sourceText,
ParamListInfo params,
QueryEnvironment *queryEnv,
@@ -72,7 +73,8 @@ CreateQueryDesc(PlannedStmt *plannedstmt,
DestReceiver *dest,
ParamListInfo params,
QueryEnvironment *queryEnv,
- int instrument_options)
+ int instrument_options,
+ EState *prep_estate)
{
QueryDesc *qd = palloc_object(QueryDesc);
@@ -93,6 +95,9 @@ CreateQueryDesc(PlannedStmt *plannedstmt,
qd->planstate = NULL;
qd->totaltime = NULL;
+ /* Use the EState created by ExecutorPrep() if already done. */
+ qd->estate = prep_estate;
+
/* not yet executed */
qd->already_executed = false;
@@ -123,6 +128,7 @@ FreeQueryDesc(QueryDesc *qdesc)
* PORTAL_ONE_RETURNING, or PORTAL_ONE_MOD_WITH portal
*
* plan: the plan tree for the query
+ * prep_estate: EState created in ExecutorPrep() for the query, if any
* sourceText: the source text of the query
* params: any parameters needed
* dest: where to send results
@@ -135,6 +141,7 @@ FreeQueryDesc(QueryDesc *qdesc)
*/
static void
ProcessQuery(PlannedStmt *plan,
+ EState *prep_estate,
const char *sourceText,
ParamListInfo params,
QueryEnvironment *queryEnv,
@@ -148,7 +155,8 @@ ProcessQuery(PlannedStmt *plan,
*/
queryDesc = CreateQueryDesc(plan, sourceText,
GetActiveSnapshot(), InvalidSnapshot,
- dest, params, queryEnv, 0);
+ dest, params, queryEnv, 0,
+ prep_estate);
/*
* Call ExecutorStart to prepare the plan for execution
@@ -495,7 +503,10 @@ PortalStart(Portal portal, ParamListInfo params,
None_Receiver,
params,
portal->queryEnv,
- 0);
+ 0,
+ portal->prep_estates ?
+ (EState *) linitial(portal->prep_estates) :
+ NULL);
/*
* If it's a scrollable cursor, executor needs to support
@@ -1185,6 +1196,7 @@ PortalRunMulti(Portal portal,
{
bool active_snapshot_set = false;
ListCell *stmtlist_item;
+ ListCell *prep_lc;
/*
* If the destination is DestRemoteExecute, change to DestNone. The
@@ -1205,9 +1217,11 @@ PortalRunMulti(Portal portal,
* Loop to handle the individual queries generated from a single parsetree
* by analysis and rewrite.
*/
+ prep_lc = list_head(portal->prep_estates);
foreach(stmtlist_item, portal->stmts)
{
PlannedStmt *pstmt = lfirst_node(PlannedStmt, stmtlist_item);
+ EState *prep_estate = next_prep_estate(portal->prep_estates, &prep_lc);
/*
* If we got a cancel signal in prior command, quit
@@ -1265,7 +1279,7 @@ PortalRunMulti(Portal portal,
if (pstmt->canSetTag)
{
/* statement can set tag string */
- ProcessQuery(pstmt,
+ ProcessQuery(pstmt, prep_estate,
portal->sourceText,
portal->portalParams,
portal->queryEnv,
@@ -1274,7 +1288,7 @@ PortalRunMulti(Portal portal,
else
{
/* stmt added by rewrite cannot set tag */
- ProcessQuery(pstmt,
+ ProcessQuery(pstmt, prep_estate,
portal->sourceText,
portal->portalParams,
portal->queryEnv,
diff --git a/src/backend/utils/mmgr/portalmem.c b/src/backend/utils/mmgr/portalmem.c
index 493f9b0ee19..443b583637c 100644
--- a/src/backend/utils/mmgr/portalmem.c
+++ b/src/backend/utils/mmgr/portalmem.c
@@ -286,6 +286,7 @@ PortalDefineQuery(Portal portal,
const char *sourceText,
CommandTag commandTag,
List *stmts,
+ List *prep_estates,
CachedPlan *cplan)
{
Assert(PortalIsValid(portal));
@@ -299,6 +300,7 @@ PortalDefineQuery(Portal portal,
portal->commandTag = commandTag;
SetQueryCompletion(&portal->qc, commandTag, 0);
portal->stmts = stmts;
+ portal->prep_estates = prep_estates;
portal->cplan = cplan;
portal->status = PORTAL_DEFINED;
}
diff --git a/src/include/commands/explain.h b/src/include/commands/explain.h
index 472e141bba3..71ebe38bc86 100644
--- a/src/include/commands/explain.h
+++ b/src/include/commands/explain.h
@@ -64,7 +64,8 @@ extern void ExplainOneUtility(Node *utilityStmt, IntoClause *into,
ExplainState *es, ParseState *pstate,
ParamListInfo params);
-extern void ExplainOnePlan(PlannedStmt *plannedstmt, IntoClause *into,
+extern void ExplainOnePlan(PlannedStmt *plannedstmt, EState *prep_estate,
+ IntoClause *into,
ExplainState *es, const char *queryString,
ParamListInfo params, QueryEnvironment *queryEnv,
const instr_time *planduration,
diff --git a/src/include/executor/execdesc.h b/src/include/executor/execdesc.h
index d3a57242844..3a2169c9613 100644
--- a/src/include/executor/execdesc.h
+++ b/src/include/executor/execdesc.h
@@ -43,7 +43,7 @@ typedef struct QueryDesc
QueryEnvironment *queryEnv; /* query environment passed in */
int instrument_options; /* OR of InstrumentOption flags */
- /* These fields are set by ExecutorStart */
+ /* These fields are set by ExecutorStart or ExecutorPrep */
TupleDesc tupDesc; /* descriptor for result tuples */
EState *estate; /* executor's query-wide state */
PlanState *planstate; /* tree of per-plan-node state */
@@ -63,7 +63,8 @@ extern QueryDesc *CreateQueryDesc(PlannedStmt *plannedstmt,
DestReceiver *dest,
ParamListInfo params,
QueryEnvironment *queryEnv,
- int instrument_options);
+ int instrument_options,
+ EState *prep_estate);
extern void FreeQueryDesc(QueryDesc *qdesc);
diff --git a/src/include/executor/executor.h b/src/include/executor/executor.h
index 064df01811e..24604120c27 100644
--- a/src/include/executor/executor.h
+++ b/src/include/executor/executor.h
@@ -21,6 +21,7 @@
#include "nodes/lockoptions.h"
#include "nodes/parsenodes.h"
#include "utils/memutils.h"
+#include "utils/resowner.h"
/*
@@ -235,6 +236,31 @@ ExecGetJunkAttribute(TupleTableSlot *slot, AttrNumber attno, bool *isNull)
*/
extern void ExecutorStart(QueryDesc *queryDesc, int eflags);
extern void standard_ExecutorStart(QueryDesc *queryDesc, int eflags);
+
+extern EState *ExecutorPrep(PlannedStmt *pstmt,
+ ParamListInfo params,
+ ResourceOwner owner,
+ int eflags);
+
+/*
+ * Walk a prep_estates list in step with a parallel stmt_list iteration.
+ * Returns the next EState (or NULL) and advances *lc.
+ *
+ * Safe when prep_estates is NIL; just returns NULL for every call.
+ */
+static inline EState *
+next_prep_estate(List *prep_estates, ListCell **lc)
+{
+ EState *result = NULL;
+
+ if (*lc != NULL)
+ {
+ result = (EState *) lfirst(*lc);
+ *lc = lnext(prep_estates, *lc);
+ }
+ return result;
+}
+
extern void ExecutorRun(QueryDesc *queryDesc,
ScanDirection direction, uint64 count);
extern void standard_ExecutorRun(QueryDesc *queryDesc,
diff --git a/src/include/nodes/execnodes.h b/src/include/nodes/execnodes.h
index 0716c5a9aed..42d75693d43 100644
--- a/src/include/nodes/execnodes.h
+++ b/src/include/nodes/execnodes.h
@@ -784,7 +784,6 @@ typedef struct EState
List *es_insert_pending_modifytables;
} EState;
-
/*
* ExecRowMark -
* runtime representation of FOR [KEY] UPDATE/SHARE clauses
diff --git a/src/include/utils/portal.h b/src/include/utils/portal.h
index a7bedb12c18..f69b4b9b479 100644
--- a/src/include/utils/portal.h
+++ b/src/include/utils/portal.h
@@ -137,6 +137,7 @@ typedef struct PortalData
CommandTag commandTag; /* command tag for original query */
QueryCompletion qc; /* command completion data for executed query */
List *stmts; /* list of PlannedStmts */
+ List *prep_estates; /* list of EStates where needed */
CachedPlan *cplan; /* CachedPlan, if stmts are from one */
ParamListInfo portalParams; /* params to pass to query */
@@ -240,6 +241,7 @@ extern void PortalDefineQuery(Portal portal,
const char *sourceText,
CommandTag commandTag,
List *stmts,
+ List *prep_estates,
CachedPlan *cplan);
extern PlannedStmt *PortalGetPrimaryStmt(Portal portal);
extern void PortalCreateHoldStore(Portal portal);
--
2.47.3
[application/octet-stream] v8-0001-Refactor-partition-pruning-initialization-for-cla.patch (10.2K, 6-v8-0001-Refactor-partition-pruning-initialization-for-cla.patch)
download | inline diff:
From a79af61882f1ff696d46f612a5b3a8ce50ee75d6 Mon Sep 17 00:00:00 2001
From: Amit Langote <[email protected]>
Date: Tue, 10 Feb 2026 15:08:52 +0900
Subject: [PATCH v8 1/5] Refactor partition pruning initialization for clarity
and modularity
Move the creation of PartitionPruneState structures out of
ExecDoInitialPruning() into a new ExecCreatePartitionPruneStates()
function. This separates the setup of pruning state from the execution
of initial pruning logic, making the code clearer and easier to
maintain.
Simplify handling of unpruned relids by moving responsibility
for recording them in EState into CreatePartitionPruneState(),
avoiding the need to pass all_leafpart_rtis as an out parameter.
Also move the setting of ecxt_param_exec_vals from
ExecCreatePartitionPruneState() to InitExecPartitionPruneContexts(),
to allow the former to be called at a time when the PARAM_EXEC
parameters have not yet been set up.
This refactoring allows callers to reuse the pruning setup logic
without always triggering pruning, a capability useful for future use
cases that may only need metadata initialization.
---
src/backend/executor/execMain.c | 1 +
src/backend/executor/execPartition.c | 103 +++++++++++++++++++--------
src/include/executor/execPartition.h | 1 +
3 files changed, 74 insertions(+), 31 deletions(-)
diff --git a/src/backend/executor/execMain.c b/src/backend/executor/execMain.c
index 58b84955c2b..c58a2abe9a7 100644
--- a/src/backend/executor/execMain.c
+++ b/src/backend/executor/execMain.c
@@ -870,6 +870,7 @@ InitPlan(QueryDesc *queryDesc, int eflags)
* to each PartitionPruneInfo entry, and the es_part_prune_results list is
* parallel to es_part_prune_infos.
*/
+ ExecCreatePartitionPruneStates(estate);
ExecDoInitialPruning(estate);
/*
diff --git a/src/backend/executor/execPartition.c b/src/backend/executor/execPartition.c
index d96d4f9947b..feea9fdfde0 100644
--- a/src/backend/executor/execPartition.c
+++ b/src/backend/executor/execPartition.c
@@ -185,8 +185,7 @@ static char *ExecBuildSlotPartitionKeyDescription(Relation rel,
static List *adjust_partition_colnos(List *colnos, ResultRelInfo *leaf_part_rri);
static List *adjust_partition_colnos_using_map(List *colnos, AttrMap *attrMap);
static PartitionPruneState *CreatePartitionPruneState(EState *estate,
- PartitionPruneInfo *pruneinfo,
- Bitmapset **all_leafpart_rtis);
+ PartitionPruneInfo *pruneinfo);
static void InitPartitionPruneContext(PartitionPruneContext *context,
List *pruning_steps,
PartitionDesc partdesc,
@@ -1943,6 +1942,9 @@ adjust_partition_colnos_using_map(List *colnos, AttrMap *attrMap)
*
* Functions:
*
+ * ExecCreatePartitionPruneStates
+ * Create PartitionPruneState for all PartitionPruneInfos in the EState
+ *
* ExecDoInitialPruning:
* Perform runtime "initial" pruning, if necessary, to determine the set
* of child subnodes that need to be initialized during ExecInitNode() for
@@ -1967,6 +1969,29 @@ adjust_partition_colnos_using_map(List *colnos, AttrMap *attrMap)
*-------------------------------------------------------------------------
*/
+/*
+ * ExecCreatePartitionPruneStates
+ *
+ * Create a PartitionPruneState for each PartitionPruneInfo in the estate,
+ * and save them in estate->es_part_prune_states. This setup is required
+ * before any initial or runtime pruning can occur.
+ */
+void
+ExecCreatePartitionPruneStates(EState *estate)
+{
+ ListCell *lc;
+
+ foreach(lc, estate->es_part_prune_infos)
+ {
+ PartitionPruneInfo *pruneinfo = lfirst_node(PartitionPruneInfo, lc);
+ PartitionPruneState *prunestate;
+
+ /* Create and save the PartitionPruneState. */
+ prunestate = CreatePartitionPruneState(estate, pruneinfo);
+ estate->es_part_prune_states = lappend(estate->es_part_prune_states,
+ prunestate);
+ }
+}
/*
* ExecDoInitialPruning
@@ -1974,11 +1999,11 @@ adjust_partition_colnos_using_map(List *colnos, AttrMap *attrMap)
* of child subnodes that need to be initialized during ExecInitNode() for
* plan nodes that support partition pruning.
*
- * This function iterates over each PartitionPruneInfo entry in
- * estate->es_part_prune_infos. For each entry, it creates a PartitionPruneState
- * and adds it to es_part_prune_states. ExecInitPartitionExecPruning() accesses
+ * This function iterates over each PartitionPruneState in
+ * estate->es_part_prune_states, which must have been populated earlier by
+ * ExecCreatePartitionPruneStates(). ExecInitPartitionExecPruning() accesses
* these states through their corresponding indexes in es_part_prune_states and
- * assign each state to the parent node's PlanState, from where it will be used
+ * assigns each state to the parent node's PlanState, from where it will be used
* for "exec" pruning.
*
* If initial pruning steps exist for a PartitionPruneInfo entry, this function
@@ -1996,20 +2021,13 @@ ExecDoInitialPruning(EState *estate)
{
ListCell *lc;
- foreach(lc, estate->es_part_prune_infos)
+ Assert(estate->es_part_prune_results == NULL);
+ foreach(lc, estate->es_part_prune_states)
{
- PartitionPruneInfo *pruneinfo = lfirst_node(PartitionPruneInfo, lc);
- PartitionPruneState *prunestate;
+ PartitionPruneState *prunestate = (PartitionPruneState *) lfirst(lc);
Bitmapset *validsubplans = NULL;
- Bitmapset *all_leafpart_rtis = NULL;
Bitmapset *validsubplan_rtis = NULL;
- /* Create and save the PartitionPruneState. */
- prunestate = CreatePartitionPruneState(estate, pruneinfo,
- &all_leafpart_rtis);
- estate->es_part_prune_states = lappend(estate->es_part_prune_states,
- prunestate);
-
/*
* Perform initial pruning steps, if any, and save the result
* bitmapset or NULL as described in the header comment.
@@ -2017,8 +2035,6 @@ ExecDoInitialPruning(EState *estate)
if (prunestate->do_initial_prune)
validsubplans = ExecFindMatchingSubPlans(prunestate, true,
&validsubplan_rtis);
- else
- validsubplan_rtis = all_leafpart_rtis;
estate->es_unpruned_relids = bms_add_members(estate->es_unpruned_relids,
validsubplan_rtis);
@@ -2136,14 +2152,12 @@ ExecInitPartitionExecPruning(PlanState *planstate,
* parent plan node's PlanState.
*
* If initial pruning steps are to be skipped (e.g., during EXPLAIN
- * (GENERIC_PLAN)), *all_leafpart_rtis will be populated with the RT indexes of
- * all leaf partitions whose scanning subnode is included in the parent plan
- * node's list of child plans. The caller must add these RT indexes to
- * estate->es_unpruned_relids.
+ * (GENERIC_PLAN)), the RT indexes of all leaf partitions whose scanning
+ * subnode is included in the parent plan node's list of child plans are
+ * added to estate->es_unpruned_relids.
*/
static PartitionPruneState *
-CreatePartitionPruneState(EState *estate, PartitionPruneInfo *pruneinfo,
- Bitmapset **all_leafpart_rtis)
+CreatePartitionPruneState(EState *estate, PartitionPruneInfo *pruneinfo)
{
PartitionPruneState *prunestate;
int n_part_hierarchies;
@@ -2377,8 +2391,8 @@ CreatePartitionPruneState(EState *estate, PartitionPruneInfo *pruneinfo,
pinfo->execparamids);
/*
- * Return all leaf partition indexes if we're skipping pruning in
- * the EXPLAIN (GENERIC_PLAN) case.
+ * Add all leaf partition indexes to es_unpruned_relids if we're
+ * skipping pruning in the EXPLAIN (GENERIC_PLAN) case.
*/
if (pinfo->initial_pruning_steps && !prunestate->do_initial_prune)
{
@@ -2390,10 +2404,29 @@ CreatePartitionPruneState(EState *estate, PartitionPruneInfo *pruneinfo,
Index rtindex = pprune->leafpart_rti_map[part_index];
if (rtindex)
- *all_leafpart_rtis = bms_add_member(*all_leafpart_rtis,
- rtindex);
+ estate->es_unpruned_relids =
+ bms_add_member(estate->es_unpruned_relids, rtindex);
}
}
+ else if (pinfo->initial_pruning_steps == NIL)
+ {
+ /*
+ * All partitions better be present in es_unpruned_relids when
+ * none are initially prunable.
+ */
+#ifdef USE_ASSERT_CHECKING
+ int part_index = -1;
+
+ while ((part_index = bms_next_member(pprune->present_parts,
+ part_index)) >= 0)
+ {
+ Index rtindex = pprune->leafpart_rti_map[part_index];
+
+ if (rtindex)
+ Assert(bms_is_member(rtindex, estate->es_unpruned_relids));
+ }
+#endif
+ }
j++;
}
@@ -2490,9 +2523,10 @@ InitPartitionPruneContext(PartitionPruneContext *context,
* Initialize exec pruning contexts deferred by CreatePartitionPruneState()
*
* This function finalizes exec pruning setup for a PartitionPruneState by
- * initializing contexts for pruning steps that require the parent plan's
- * PlanState. It iterates over PartitionPruningData entries and sets up the
- * necessary execution contexts for pruning during query execution.
+ * initializing contexts for pruning steps that require PARAM_EXEC parameters
+ * and the parent plan's PlanState. It iterates over PartitionPruningData
+ * entries and sets up the necessary execution contexts for pruning during
+ * query execution.
*
* Also fix the mapping of partition indexes to subplan indexes contained in
* prunestate by considering the new list of subplans that survived initial
@@ -2520,9 +2554,16 @@ InitExecPartitionPruneContexts(PartitionPruneState *prunestate,
bool fix_subplan_map = false;
Assert(prunestate->do_exec_prune);
+ Assert(prunestate->econtext);
Assert(parent_plan != NULL);
estate = parent_plan->state;
+ /*
+ * These might not be available when ExecCreatePartitionPruneState() is
+ * called.
+ */
+ prunestate->econtext->ecxt_param_exec_vals = estate->es_param_exec_vals;
+
/*
* No need to fix subplans maps if initial pruning didn't eliminate any
* subplans.
diff --git a/src/include/executor/execPartition.h b/src/include/executor/execPartition.h
index 82063ec2a16..4c96808c376 100644
--- a/src/include/executor/execPartition.h
+++ b/src/include/executor/execPartition.h
@@ -130,6 +130,7 @@ typedef struct PartitionPruneState
PartitionPruningData *partprunedata[FLEXIBLE_ARRAY_MEMBER];
} PartitionPruneState;
+extern void ExecCreatePartitionPruneStates(EState *estate);
extern void ExecDoInitialPruning(EState *estate);
extern PartitionPruneState *ExecInitPartitionExecPruning(PlanState *planstate,
int n_total_subplans,
--
2.47.3
view thread (114+ messages) latest in thread
reply
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Reply to all the recipients using the --to and --cc options:
reply via email
To: [email protected]
Cc: [email protected], [email protected], [email protected], [email protected], [email protected], [email protected], [email protected], [email protected], [email protected], [email protected], [email protected], [email protected]
Subject: Re: generic plans and "initial" pruning
In-Reply-To: <CA+HiwqHBxDL=3qQa1f-sBOBZqB88EiVAiagXF3X8Kagpr6Yhpw@mail.gmail.com>
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
This inbox is served by agora; see mirroring instructions
for how to clone and mirror all data and code used for this inbox