database/postgres
1
0
Fork 0
mirror of https://github.com/postgres/postgres.git synced 2025-07-03 20:02:46 +03:00
Code Activity

Separate reinitialization of shared parallel-scan state from ExecReScan.

Browse Source 
Previously, the parallel executor logic did reinitialization of shared
state within the ExecReScan code for parallel-aware scan nodes.  This is
problematic, because it means that the ExecReScan call has to occur
synchronously (ie, during the parent Gather node's ReScan call).  That is
swimming very much against the tide so far as the ExecReScan machinery is
concerned; the fact that it works at all today depends on a lot of fragile
assumptions, such as that no plan node between Gather and a parallel-aware
scan node is parameterized.  Another objection is that because ExecReScan
might be called in workers as well as the leader, hacky extra tests are
needed in some places to prevent unwanted shared-state resets.

Hence, let's separate this code into two functions, a ReInitializeDSM
call and the ReScan call proper.  ReInitializeDSM is called only in
the leader and is guaranteed to run before we start new workers.
ReScan is returned to its traditional function of resetting only local
state, which means that ExecReScan's usual habits of delaying or
eliminating child rescan calls are safe again.

As with the preceding commit 7df2c1f8d, it doesn't seem to be necessary
to make these changes in 9.6, which is a good thing because the FDW and
CustomScan APIs are impacted.

Discussion: https://postgr.es/m/CAA4eK1JkByysFJNh9M349u_nNjqETuEnY_y1VUc_kJiU0bxtaQ@mail.gmail.com
This commit is contained in:
Tom Lane
2017-08-30 13:18:16 -04:00
parent 6c2c5bea3c
commit 41b0dd987d
24 changed files with 328 additions and 133 deletions
Download Patch File Download Diff File Expand all files Collapse all files

101
src/backend/executor/execParallel.c
View File

@ -119,6 +119,8 @@ static bool ExecParallelInitializeDSM(PlanState *node,
ExecParallelInitializeDSMContext *d);
static shm_mq_handle **ExecParallelSetupTupleQueues(ParallelContext *pcxt,
bool reinitialize);
static bool ExecParallelReInitializeDSM(PlanState *planstate,
ParallelContext *pcxt);
static bool ExecParallelRetrieveInstrumentation(PlanState *planstate,
SharedExecutorInstrumentation *instrumentation);
@ -255,6 +257,8 @@ ExecParallelEstimate(PlanState *planstate, ExecParallelEstimateContext *e)
case T_SortState:
/* even when not parallel-aware */
ExecSortEstimate((SortState *) planstate, e->pcxt);
break;
default:
break;
}
@ -325,6 +329,8 @@ ExecParallelInitializeDSM(PlanState *planstate,
case T_SortState:
/* even when not parallel-aware */
ExecSortInitializeDSM((SortState *) planstate, d->pcxt);
break;
default:
break;
}
@ -384,18 +390,6 @@ ExecParallelSetupTupleQueues(ParallelContext *pcxt, bool reinitialize)
return responseq;
}
/*
* Re-initialize the parallel executor info such that it can be reused by
* workers.
*/
void
ExecParallelReinitialize(ParallelExecutorInfo *pei)
{
ReinitializeParallelDSM(pei->pcxt);
pei->tqueue = ExecParallelSetupTupleQueues(pei->pcxt, true);
pei->finished = false;
}
/*
* Sets up the required infrastructure for backend workers to perform
* execution and return results to the main backend.
@ -599,7 +593,7 @@ ExecInitParallelPlan(PlanState *planstate, EState *estate, int nworkers,
ExecParallelInitializeDSM(planstate, &d);
/*
* Make sure that the world hasn't shifted under our feat. This could
* Make sure that the world hasn't shifted under our feet. This could
* probably just be an Assert(), but let's be conservative for now.
*/
if (e.nnodes != d.nnodes)
@ -609,6 +603,82 @@ ExecInitParallelPlan(PlanState *planstate, EState *estate, int nworkers,
return pei;
}
/*
* Re-initialize the parallel executor shared memory state before launching
* a fresh batch of workers.
*/
void
ExecParallelReinitialize(PlanState *planstate,
ParallelExecutorInfo *pei)
{
/* Old workers must already be shut down */
Assert(pei->finished);
ReinitializeParallelDSM(pei->pcxt);
pei->tqueue = ExecParallelSetupTupleQueues(pei->pcxt, true);
pei->finished = false;
/* Traverse plan tree and let each child node reset associated state. */
ExecParallelReInitializeDSM(planstate, pei->pcxt);
}
/*
* Traverse plan tree to reinitialize per-node dynamic shared memory state
*/
static bool
ExecParallelReInitializeDSM(PlanState *planstate,
ParallelContext *pcxt)
{
if (planstate == NULL)
return false;
/*
* Call reinitializers for DSM-using plan nodes.
*/
switch (nodeTag(planstate))
{
case T_SeqScanState:
if (planstate->plan->parallel_aware)
ExecSeqScanReInitializeDSM((SeqScanState *) planstate,
pcxt);
break;
case T_IndexScanState:
if (planstate->plan->parallel_aware)
ExecIndexScanReInitializeDSM((IndexScanState *) planstate,
pcxt);
break;
case T_IndexOnlyScanState:
if (planstate->plan->parallel_aware)
ExecIndexOnlyScanReInitializeDSM((IndexOnlyScanState *) planstate,
pcxt);
break;
case T_ForeignScanState:
if (planstate->plan->parallel_aware)
ExecForeignScanReInitializeDSM((ForeignScanState *) planstate,
pcxt);
break;
case T_CustomScanState:
if (planstate->plan->parallel_aware)
ExecCustomScanReInitializeDSM((CustomScanState *) planstate,
pcxt);
break;
case T_BitmapHeapScanState:
if (planstate->plan->parallel_aware)
ExecBitmapHeapReInitializeDSM((BitmapHeapScanState *) planstate,
pcxt);
break;
case T_SortState:
/* even when not parallel-aware */
ExecSortReInitializeDSM((SortState *) planstate, pcxt);
break;
default:
break;
}
return planstate_tree_walker(planstate, ExecParallelReInitializeDSM, pcxt);
}
/*
* Copy instrumentation information about this node and its descendants from
* dynamic shared memory.
@ -845,12 +915,13 @@ ExecParallelInitializeWorker(PlanState *planstate, shm_toc *toc)
break;
case T_BitmapHeapScanState:
if (planstate->plan->parallel_aware)
ExecBitmapHeapInitializeWorker(
(BitmapHeapScanState *) planstate, toc);
ExecBitmapHeapInitializeWorker((BitmapHeapScanState *) planstate, toc);
break;
case T_SortState:
/* even when not parallel-aware */
ExecSortInitializeWorker((SortState *) planstate, toc);
break;
default:
break;
}

42
src/backend/executor/nodeBitmapHeapscan.c
View File

@ -705,23 +705,6 @@ ExecReScanBitmapHeapScan(BitmapHeapScanState *node)
node->shared_tbmiterator = NULL;
node->shared_prefetch_iterator = NULL;
/* Reset parallel bitmap state, if present */
if (node->pstate)
{
dsa_area *dsa = node->ss.ps.state->es_query_dsa;
node->pstate->state = BM_INITIAL;
if (DsaPointerIsValid(node->pstate->tbmiterator))
tbm_free_shared_area(dsa, node->pstate->tbmiterator);
if (DsaPointerIsValid(node->pstate->prefetch_iterator))
tbm_free_shared_area(dsa, node->pstate->prefetch_iterator);
node->pstate->tbmiterator = InvalidDsaPointer;
node->pstate->prefetch_iterator = InvalidDsaPointer;
}
ExecScanReScan(&node->ss);
/*
@ -999,6 +982,31 @@ ExecBitmapHeapInitializeDSM(BitmapHeapScanState *node,
node->pstate = pstate;
}
/* ----------------------------------------------------------------
* ExecBitmapHeapReInitializeDSM
*
* Reset shared state before beginning a fresh scan.
* ----------------------------------------------------------------
*/
void
ExecBitmapHeapReInitializeDSM(BitmapHeapScanState *node,
ParallelContext *pcxt)
{
ParallelBitmapHeapState *pstate = node->pstate;
dsa_area *dsa = node->ss.ps.state->es_query_dsa;
pstate->state = BM_INITIAL;
if (DsaPointerIsValid(pstate->tbmiterator))
tbm_free_shared_area(dsa, pstate->tbmiterator);
if (DsaPointerIsValid(pstate->prefetch_iterator))
tbm_free_shared_area(dsa, pstate->prefetch_iterator);
pstate->tbmiterator = InvalidDsaPointer;
pstate->prefetch_iterator = InvalidDsaPointer;
}
/* ----------------------------------------------------------------
* ExecBitmapHeapInitializeWorker
*

15
src/backend/executor/nodeCustom.c
View File

@ -194,6 +194,21 @@ ExecCustomScanInitializeDSM(CustomScanState *node, ParallelContext *pcxt)
}
}
void
ExecCustomScanReInitializeDSM(CustomScanState *node, ParallelContext *pcxt)
{
const CustomExecMethods *methods = node->methods;
if (methods->ReInitializeDSMCustomScan)
{
int plan_node_id = node->ss.ps.plan->plan_node_id;
void *coordinate;
coordinate = shm_toc_lookup(pcxt->toc, plan_node_id, false);
methods->ReInitializeDSMCustomScan(node, pcxt, coordinate);
}
}
void
ExecCustomScanInitializeWorker(CustomScanState *node, shm_toc *toc)
{

23
src/backend/executor/nodeForeignscan.c
View File

@ -332,7 +332,28 @@ ExecForeignScanInitializeDSM(ForeignScanState *node, ParallelContext *pcxt)
}
/* ----------------------------------------------------------------
* ExecForeignScanInitializeDSM
* ExecForeignScanReInitializeDSM
*
* Reset shared state before beginning a fresh scan.
* ----------------------------------------------------------------
*/
void
ExecForeignScanReInitializeDSM(ForeignScanState *node, ParallelContext *pcxt)
{
FdwRoutine *fdwroutine = node->fdwroutine;
if (fdwroutine->ReInitializeDSMForeignScan)
{
int plan_node_id = node->ss.ps.plan->plan_node_id;
void *coordinate;
coordinate = shm_toc_lookup(pcxt->toc, plan_node_id, false);
fdwroutine->ReInitializeDSMForeignScan(node, pcxt, coordinate);
}
}
/* ----------------------------------------------------------------
* ExecForeignScanInitializeWorker
*
* Initialization according to the parallel coordination information
* ----------------------------------------------------------------

29
src/backend/executor/nodeGather.c
View File

@ -153,12 +153,15 @@ ExecGather(PlanState *pstate)
{
ParallelContext *pcxt;
/* Initialize the workers required to execute Gather node. */
/* Initialize, or re-initialize, shared state needed by workers. */
if (!node->pei)
node->pei = ExecInitParallelPlan(node->ps.lefttree,
estate,
gather->num_workers,
node->tuples_needed);
else
ExecParallelReinitialize(node->ps.lefttree,
node->pei);
/*
* Register backend workers. We might not get as many as we
@ -426,7 +429,7 @@ ExecShutdownGather(GatherState *node)
/* ----------------------------------------------------------------
* ExecReScanGather
*
* Re-initialize the workers and rescans a relation via them.
* Prepare to re-scan the result of a Gather.
* ----------------------------------------------------------------
*/
void
@ -435,19 +438,12 @@ ExecReScanGather(GatherState *node)
Gather *gather = (Gather *) node->ps.plan;
PlanState *outerPlan = outerPlanState(node);
/*
* Re-initialize the parallel workers to perform rescan of relation. We
* want to gracefully shutdown all the workers so that they should be able
* to propagate any error or other information to master backend before
* dying. Parallel context will be reused for rescan.
*/
/* Make sure any existing workers are gracefully shut down */
ExecShutdownGatherWorkers(node);
/* Mark node so that shared state will be rebuilt at next call */
node->initialized = false;
if (node->pei)
ExecParallelReinitialize(node->pei);
/*
* Set child node's chgParam to tell it that the next scan might deliver a
* different set of rows within the leader process. (The overall rowset
@ -459,10 +455,15 @@ ExecReScanGather(GatherState *node)
outerPlan->chgParam = bms_add_member(outerPlan->chgParam,
gather->rescan_param);
/*
* if chgParam of subnode is not null then plan will be re-scanned by
* first ExecProcNode.
* If chgParam of subnode is not null then plan will be re-scanned by
* first ExecProcNode. Note: because this does nothing if we have a
* rescan_param, it's currently guaranteed that parallel-aware child nodes
* will not see a ReScan call until after they get a ReInitializeDSM call.
* That ordering might not be something to rely on, though. A good rule
* of thumb is that ReInitializeDSM should reset only shared state, ReScan
* should reset only local state, and anything that depends on both of
* those steps being finished must wait until the first ExecProcNode call.
*/
if (outerPlan->chgParam == NULL)
ExecReScan(outerPlan);

29
src/backend/executor/nodeGatherMerge.c
View File

@ -187,12 +187,15 @@ ExecGatherMerge(PlanState *pstate)
{
ParallelContext *pcxt;
/* Initialize data structures for workers. */
/* Initialize, or re-initialize, shared state needed by workers. */
if (!node->pei)
node->pei = ExecInitParallelPlan(node->ps.lefttree,
estate,
gm->num_workers,
node->tuples_needed);
else
ExecParallelReinitialize(node->ps.lefttree,
node->pei);
/* Try to launch workers. */
pcxt = node->pei->pcxt;
@ -321,7 +324,7 @@ ExecShutdownGatherMergeWorkers(GatherMergeState *node)
/* ----------------------------------------------------------------
* ExecReScanGatherMerge
*
* Re-initialize the workers and rescans a relation via them.
* Prepare to re-scan the result of a GatherMerge.
* ----------------------------------------------------------------
*/
void
@ -330,20 +333,13 @@ ExecReScanGatherMerge(GatherMergeState *node)
GatherMerge *gm = (GatherMerge *) node->ps.plan;
PlanState *outerPlan = outerPlanState(node);
/*
* Re-initialize the parallel workers to perform rescan of relation. We
* want to gracefully shutdown all the workers so that they should be able
* to propagate any error or other information to master backend before
* dying. Parallel context will be reused for rescan.
*/
/* Make sure any existing workers are gracefully shut down */
ExecShutdownGatherMergeWorkers(node);
/* Mark node so that shared state will be rebuilt at next call */
node->initialized = false;
node->gm_initialized = false;
if (node->pei)
ExecParallelReinitialize(node->pei);
/*
* Set child node's chgParam to tell it that the next scan might deliver a
* different set of rows within the leader process. (The overall rowset
@ -355,10 +351,15 @@ ExecReScanGatherMerge(GatherMergeState *node)
outerPlan->chgParam = bms_add_member(outerPlan->chgParam,
gm->rescan_param);
/*
* if chgParam of subnode is not null then plan will be re-scanned by
* first ExecProcNode.
* If chgParam of subnode is not null then plan will be re-scanned by
* first ExecProcNode. Note: because this does nothing if we have a
* rescan_param, it's currently guaranteed that parallel-aware child nodes
* will not see a ReScan call until after they get a ReInitializeDSM call.
* That ordering might not be something to rely on, though. A good rule
* of thumb is that ReInitializeDSM should reset only shared state, ReScan
* should reset only local state, and anything that depends on both of
* those steps being finished must wait until the first ExecProcNode call.
*/
if (outerPlan->chgParam == NULL)
ExecReScan(outerPlan);

29
src/backend/executor/nodeIndexonlyscan.c
View File

@ -25,6 +25,7 @@
* parallel index-only scan
* ExecIndexOnlyScanInitializeDSM initialize DSM for parallel
* index-only scan
* ExecIndexOnlyScanReInitializeDSM reinitialize DSM for fresh scan
* ExecIndexOnlyScanInitializeWorker attach to DSM info in parallel worker
*/
#include "postgres.h"
@ -336,16 +337,6 @@ ExecIndexOnlyScan(PlanState *pstate)
void
ExecReScanIndexOnlyScan(IndexOnlyScanState *node)
{
bool reset_parallel_scan = true;
/*
* If we are here to just update the scan keys, then don't reset parallel
* scan. For detailed reason behind this look in the comments for
* ExecReScanIndexScan.
*/
if (node->ioss_NumRuntimeKeys != 0 && !node->ioss_RuntimeKeysReady)
reset_parallel_scan = false;
/*
* If we are doing runtime key calculations (ie, any of the index key
* values weren't simple Consts), compute the new key values. But first,
@ -366,15 +357,10 @@ ExecReScanIndexOnlyScan(IndexOnlyScanState *node)
/* reset index scan */
if (node->ioss_ScanDesc)
{
index_rescan(node->ioss_ScanDesc,
node->ioss_ScanKeys, node->ioss_NumScanKeys,
node->ioss_OrderByKeys, node->ioss_NumOrderByKeys);
if (reset_parallel_scan && node->ioss_ScanDesc->parallel_scan)
index_parallelrescan(node->ioss_ScanDesc);
}
ExecScanReScan(&node->ss);
}
@ -671,6 +657,19 @@ ExecIndexOnlyScanInitializeDSM(IndexOnlyScanState *node,
node->ioss_OrderByKeys, node->ioss_NumOrderByKeys);
}
/* ----------------------------------------------------------------
* ExecIndexOnlyScanReInitializeDSM
*
* Reset shared state before beginning a fresh scan.
* ----------------------------------------------------------------
*/
void
ExecIndexOnlyScanReInitializeDSM(IndexOnlyScanState *node,
ParallelContext *pcxt)
{
index_parallelrescan(node->ioss_ScanDesc);
}
/* ----------------------------------------------------------------
* ExecIndexOnlyScanInitializeWorker
*

38
src/backend/executor/nodeIndexscan.c
View File

@ -24,6 +24,7 @@
* ExecIndexRestrPos restores scan position.
* ExecIndexScanEstimate estimates DSM space needed for parallel index scan
* ExecIndexScanInitializeDSM initialize DSM for parallel indexscan
* ExecIndexScanReInitializeDSM reinitialize DSM for fresh scan
* ExecIndexScanInitializeWorker attach to DSM info in parallel worker
*/
#include "postgres.h"
@ -577,18 +578,6 @@ ExecIndexScan(PlanState *pstate)
void
ExecReScanIndexScan(IndexScanState *node)
{
bool reset_parallel_scan = true;
/*
* If we are here to just update the scan keys, then don't reset parallel
* scan. We don't want each of the participating process in the parallel
* scan to update the shared parallel scan state at the start of the scan.
* It is quite possible that one of the participants has already begun
* scanning the index when another has yet to start it.
*/
if (node->iss_NumRuntimeKeys != 0 && !node->iss_RuntimeKeysReady)
reset_parallel_scan = false;
/*
* If we are doing runtime key calculations (ie, any of the index key
* values weren't simple Consts), compute the new key values. But first,
@ -614,21 +603,11 @@ ExecReScanIndexScan(IndexScanState *node)
reorderqueue_pop(node);
}
/*
* Reset (parallel) index scan. For parallel-aware nodes, the scan
* descriptor is initialized during actual execution of node and we can
* reach here before that (ex. during execution of nest loop join). So,
* avoid updating the scan descriptor at that time.
*/
/* reset index scan */
if (node->iss_ScanDesc)
{
index_rescan(node->iss_ScanDesc,
node->iss_ScanKeys, node->iss_NumScanKeys,
node->iss_OrderByKeys, node->iss_NumOrderByKeys);
if (reset_parallel_scan && node->iss_ScanDesc->parallel_scan)
index_parallelrescan(node->iss_ScanDesc);
}
node->iss_ReachedEnd = false;
ExecScanReScan(&node->ss);
@ -1716,6 +1695,19 @@ ExecIndexScanInitializeDSM(IndexScanState *node,
node->iss_OrderByKeys, node->iss_NumOrderByKeys);
}
/* ----------------------------------------------------------------
* ExecIndexScanReInitializeDSM
*
* Reset shared state before beginning a fresh scan.
* ----------------------------------------------------------------
*/
void
ExecIndexScanReInitializeDSM(IndexScanState *node,
ParallelContext *pcxt)
{
index_parallelrescan(node->iss_ScanDesc);
}
/* ----------------------------------------------------------------
* ExecIndexScanInitializeWorker
*

16
src/backend/executor/nodeSeqscan.c
View File

@ -22,6 +22,7 @@
*
* ExecSeqScanEstimate estimates DSM space needed for parallel scan
* ExecSeqScanInitializeDSM initialize DSM for parallel scan
* ExecSeqScanReInitializeDSM reinitialize DSM for fresh parallel scan
* ExecSeqScanInitializeWorker attach to DSM info in parallel worker
*/
#include "postgres.h"
@ -324,6 +325,21 @@ ExecSeqScanInitializeDSM(SeqScanState *node,
heap_beginscan_parallel(node->ss.ss_currentRelation, pscan);
}
/* ----------------------------------------------------------------
* ExecSeqScanReInitializeDSM
*
* Reset shared state before beginning a fresh scan.
* ----------------------------------------------------------------
*/
void
ExecSeqScanReInitializeDSM(SeqScanState *node,
ParallelContext *pcxt)
{
HeapScanDesc scan = node->ss.ss_currentScanDesc;
heap_parallelscan_reinitialize(scan->rs_parallel);
}
/* ----------------------------------------------------------------
* ExecSeqScanInitializeWorker
*

17
src/backend/executor/nodeSort.c
View File

@ -396,6 +396,23 @@ ExecSortInitializeDSM(SortState *node, ParallelContext *pcxt)
node->shared_info);
}
/* ----------------------------------------------------------------
* ExecSortReInitializeDSM
*
* Reset shared state before beginning a fresh scan.
* ----------------------------------------------------------------
*/
void
ExecSortReInitializeDSM(SortState *node, ParallelContext *pcxt)
{
/* If there's any instrumentation space, clear it for next time */
if (node->shared_info != NULL)
{
memset(node->shared_info->sinstrument, 0,
node->shared_info->num_workers * sizeof(TuplesortInstrumentation));
}
}
/* ----------------------------------------------------------------
* ExecSortInitializeWorker
*