diff options
author | Tom Lane <tgl@sss.pgh.pa.us> | 2009-10-12 18:10:51 +0000 |
---|---|---|
committer | Tom Lane <tgl@sss.pgh.pa.us> | 2009-10-12 18:10:51 +0000 |
commit | 0adaf4cb312fe3eff83e786d6a0b53ae2cdc9302 (patch) | |
tree | 177cf2ea668d6fcabbafc37db4741813ea3f685d /src/backend/executor/execMain.c | |
parent | 05d249717d652f0b16960d8a58611e222f1f907b (diff) | |
download | postgresql-0adaf4cb312fe3eff83e786d6a0b53ae2cdc9302.tar.gz postgresql-0adaf4cb312fe3eff83e786d6a0b53ae2cdc9302.zip |
Move the handling of SELECT FOR UPDATE locking and rechecking out of
execMain.c and into a new plan node type LockRows. Like the recent change
to put table updating into a ModifyTable plan node, this increases planning
flexibility by allowing the operations to occur below the top level of the
plan tree. It's necessary in any case to restore the previous behavior of
having FOR UPDATE locking occur before ModifyTable does.
This partially refactors EvalPlanQual to allow multiple rows-under-test
to be inserted into the EPQ machinery before starting an EPQ test query.
That isn't sufficient to fix EPQ's general bogosity in the face of plans
that return multiple rows per test row, though. Since this patch is
mostly about getting some plan node infrastructure in place and not about
fixing ten-year-old bugs, I will leave EPQ improvements for another day.
Another behavioral change that we could now think about is doing FOR UPDATE
before LIMIT, but that too seems like it should be treated as a followon
patch.
Diffstat (limited to 'src/backend/executor/execMain.c')
-rw-r--r-- | src/backend/executor/execMain.c | 402 |
1 files changed, 174 insertions, 228 deletions
diff --git a/src/backend/executor/execMain.c b/src/backend/executor/execMain.c index 7c788ea6df4..d03ad094184 100644 --- a/src/backend/executor/execMain.c +++ b/src/backend/executor/execMain.c @@ -26,7 +26,7 @@ * * * IDENTIFICATION - * $PostgreSQL: pgsql/src/backend/executor/execMain.c,v 1.332 2009/10/10 01:43:47 tgl Exp $ + * $PostgreSQL: pgsql/src/backend/executor/execMain.c,v 1.333 2009/10/12 18:10:41 tgl Exp $ * *------------------------------------------------------------------------- */ @@ -66,6 +66,8 @@ typedef struct evalPlanQual Index rti; EState *estate; PlanState *planstate; + PlanState *origplanstate; + TupleTableSlot *resultslot; struct evalPlanQual *next; /* stack of active PlanQual plans */ struct evalPlanQual *free; /* list of free PlanQual plans */ } evalPlanQual; @@ -79,7 +81,6 @@ static void ExecutePlan(EState *estate, PlanState *planstate, long numberTuples, ScanDirection direction, DestReceiver *dest); -static TupleTableSlot *EvalPlanQualNext(EState *estate); static void EndEvalPlanQual(EState *estate); static void ExecCheckRTPerms(List *rangeTable); static void ExecCheckRTEPerms(RangeTblEntry *rte); @@ -695,21 +696,9 @@ InitPlan(QueryDesc *queryDesc, int eflags) } /* - * Detect whether we're doing SELECT INTO. If so, set the es_into_oids - * flag appropriately so that the plan tree will be initialized with the - * correct tuple descriptors. (Other SELECT INTO stuff comes later.) - */ - estate->es_select_into = false; - if (operation == CMD_SELECT && plannedstmt->intoClause != NULL) - { - estate->es_select_into = true; - estate->es_into_oids = interpretOidsOption(plannedstmt->intoClause->options); - } - - /* - * Have to lock relations selected FOR UPDATE/FOR SHARE before we - * initialize the plan tree, else we'd be doing a lock upgrade. While we - * are at it, build the ExecRowMark list. + * Similarly, we have to lock relations selected FOR UPDATE/FOR SHARE + * before we initialize the plan tree, else we'd be risking lock + * upgrades. While we are at it, build the ExecRowMark list. */ estate->es_rowMarks = NIL; foreach(l, plannedstmt->rowMarks) @@ -729,9 +718,10 @@ InitPlan(QueryDesc *queryDesc, int eflags) erm->relation = relation; erm->rti = rc->rti; erm->prti = rc->prti; + erm->rowmarkId = rc->rowmarkId; erm->forUpdate = rc->forUpdate; erm->noWait = rc->noWait; - /* We'll locate the junk attrs below */ + /* remaining fields are filled during LockRows plan node init */ erm->ctidAttNo = InvalidAttrNumber; erm->toidAttNo = InvalidAttrNumber; ItemPointerSetInvalid(&(erm->curCtid)); @@ -739,6 +729,18 @@ InitPlan(QueryDesc *queryDesc, int eflags) } /* + * Detect whether we're doing SELECT INTO. If so, set the es_into_oids + * flag appropriately so that the plan tree will be initialized with the + * correct tuple descriptors. (Other SELECT INTO stuff comes later.) + */ + estate->es_select_into = false; + if (operation == CMD_SELECT && plannedstmt->intoClause != NULL) + { + estate->es_select_into = true; + estate->es_into_oids = interpretOidsOption(plannedstmt->intoClause->options); + } + + /* * Initialize the executor's tuple table to empty. */ estate->es_tupleTable = NIL; @@ -749,7 +751,6 @@ InitPlan(QueryDesc *queryDesc, int eflags) estate->es_evalPlanQual = NULL; estate->es_evTupleNull = NULL; estate->es_evTuple = NULL; - estate->es_useEvalPlan = false; /* * Initialize private state information for each SubPlan. We must do this @@ -826,37 +827,6 @@ InitPlan(QueryDesc *queryDesc, int eflags) /* Want to return the cleaned tuple type */ tupType = j->jf_cleanTupType; - - /* For SELECT FOR UPDATE/SHARE, find the junk attrs now */ - foreach(l, estate->es_rowMarks) - { - ExecRowMark *erm = (ExecRowMark *) lfirst(l); - char resname[32]; - - /* always need the ctid */ - snprintf(resname, sizeof(resname), "ctid%u", - erm->prti); - erm->ctidAttNo = ExecFindJunkAttribute(j, resname); - if (!AttributeNumberIsValid(erm->ctidAttNo)) - elog(ERROR, "could not find junk \"%s\" column", - resname); - /* if child relation, need tableoid too */ - if (erm->rti != erm->prti) - { - snprintf(resname, sizeof(resname), "tableoid%u", - erm->prti); - erm->toidAttNo = ExecFindJunkAttribute(j, resname); - if (!AttributeNumberIsValid(erm->toidAttNo)) - elog(ERROR, "could not find junk \"%s\" column", - resname); - } - } - } - else - { - estate->es_junkFilter = NULL; - if (estate->es_rowMarks) - elog(ERROR, "SELECT FOR UPDATE/SHARE, but no junk columns"); } } @@ -1190,8 +1160,6 @@ ExecutePlan(EState *estate, ScanDirection direction, DestReceiver *dest) { - JunkFilter *junkfilter; - TupleTableSlot *planSlot; TupleTableSlot *slot; long current_tuple_count; @@ -1216,23 +1184,14 @@ ExecutePlan(EState *estate, /* * Execute the plan and obtain a tuple */ -lnext: ; - if (estate->es_useEvalPlan) - { - planSlot = EvalPlanQualNext(estate); - if (TupIsNull(planSlot)) - planSlot = ExecProcNode(planstate); - } - else - planSlot = ExecProcNode(planstate); + slot = ExecProcNode(planstate); /* * if the tuple is null, then we assume there is nothing more to * process so we just end the loop... */ - if (TupIsNull(planSlot)) + if (TupIsNull(slot)) break; - slot = planSlot; /* * If we have a junk filter, then project a new tuple with the junk @@ -1241,128 +1200,13 @@ lnext: ; * Store this new "clean" tuple in the junkfilter's resultSlot. * (Formerly, we stored it back over the "dirty" tuple, which is WRONG * because that tuple slot has the wrong descriptor.) - * - * But first, extract all the junk information we need. */ - if ((junkfilter = estate->es_junkFilter) != NULL) - { - /* - * Process any FOR UPDATE or FOR SHARE locking requested. - */ - if (estate->es_rowMarks != NIL) - { - ListCell *l; - - lmark: ; - foreach(l, estate->es_rowMarks) - { - ExecRowMark *erm = lfirst(l); - Datum datum; - bool isNull; - HeapTupleData tuple; - Buffer buffer; - ItemPointerData update_ctid; - TransactionId update_xmax; - TupleTableSlot *newSlot; - LockTupleMode lockmode; - HTSU_Result test; - - /* if child rel, must check whether it produced this row */ - if (erm->rti != erm->prti) - { - Oid tableoid; - - datum = ExecGetJunkAttribute(slot, - erm->toidAttNo, - &isNull); - /* shouldn't ever get a null result... */ - if (isNull) - elog(ERROR, "tableoid is NULL"); - tableoid = DatumGetObjectId(datum); - - if (tableoid != RelationGetRelid(erm->relation)) - { - /* this child is inactive right now */ - ItemPointerSetInvalid(&(erm->curCtid)); - continue; - } - } - - /* okay, fetch the tuple by ctid */ - datum = ExecGetJunkAttribute(slot, - erm->ctidAttNo, - &isNull); - /* shouldn't ever get a null result... */ - if (isNull) - elog(ERROR, "ctid is NULL"); - tuple.t_self = *((ItemPointer) DatumGetPointer(datum)); - - if (erm->forUpdate) - lockmode = LockTupleExclusive; - else - lockmode = LockTupleShared; - - test = heap_lock_tuple(erm->relation, &tuple, &buffer, - &update_ctid, &update_xmax, - estate->es_output_cid, - lockmode, erm->noWait); - ReleaseBuffer(buffer); - switch (test) - { - case HeapTupleSelfUpdated: - /* treat it as deleted; do not process */ - goto lnext; - - case HeapTupleMayBeUpdated: - break; - - case HeapTupleUpdated: - if (IsXactIsoLevelSerializable) - ereport(ERROR, - (errcode(ERRCODE_T_R_SERIALIZATION_FAILURE), - errmsg("could not serialize access due to concurrent update"))); - if (!ItemPointerEquals(&update_ctid, - &tuple.t_self)) - { - /* updated, so look at updated version */ - newSlot = EvalPlanQual(estate, - erm->rti, - planstate, - &update_ctid, - update_xmax); - if (!TupIsNull(newSlot)) - { - slot = planSlot = newSlot; - estate->es_useEvalPlan = true; - goto lmark; - } - } - - /* - * if tuple was deleted or PlanQual failed for - * updated tuple - we must not return this tuple! - */ - goto lnext; - - default: - elog(ERROR, "unrecognized heap_lock_tuple status: %u", - test); - } - - /* Remember tuple TID for WHERE CURRENT OF */ - erm->curCtid = tuple.t_self; - } - } - - /* - * Create a new "clean" tuple with all junk attributes removed. - */ - slot = ExecFilterJunk(junkfilter, slot); - } + if (estate->es_junkFilter != NULL) + slot = ExecFilterJunk(estate->es_junkFilter, slot); /* * If we are supposed to send the tuple somewhere, do so. - * (In practice this is probably always the case at this point.) + * (In practice, this is probably always the case at this point.) */ if (sendTuples) (*dest->receiveSlot) (slot, dest); @@ -1509,18 +1353,104 @@ EvalPlanQual(EState *estate, Index rti, PlanState *subplanstate, ItemPointer tid, TransactionId priorXmax) { - evalPlanQual *epq; - EState *epqstate; + TupleTableSlot *slot; + HeapTuple copyTuple; + + Assert(rti != 0); + + /* + * Get the updated version of the row; if fail, return NULL. + */ + copyTuple = EvalPlanQualFetch(estate, rti, tid, priorXmax); + + if (copyTuple == NULL) + return NULL; + + /* + * For UPDATE/DELETE we have to return tid of actual row we're executing + * PQ for. + */ + *tid = copyTuple->t_self; + + /* + * Need to run a recheck subquery. Find or create a PQ stack entry. + */ + EvalPlanQualPush(estate, rti, subplanstate); + + /* + * free old RTE' tuple, if any, and store target tuple where relation's + * scan node will see it + */ + EvalPlanQualSetTuple(estate, rti, copyTuple); + + /* + * Run the EPQ query, but just for one tuple. + */ + slot = EvalPlanQualNext(estate); + + /* + * If we got a result, we must copy it out of the EPQ query's local + * context before we shut down the EPQ query. + */ + if (TupIsNull(slot)) + slot = NULL; /* in case we got back an empty slot */ + else + { + TupleDesc tupdesc = CreateTupleDescCopy(slot->tts_tupleDescriptor); + evalPlanQual *epq = estate->es_evalPlanQual; + + if (epq->resultslot == NULL) + { + epq->resultslot = ExecInitExtraTupleSlot(estate); + ExecSetSlotDescriptor(epq->resultslot, tupdesc); + } + else + { + TupleDesc oldtupdesc = epq->resultslot->tts_tupleDescriptor; + + ExecSetSlotDescriptor(epq->resultslot, tupdesc); + FreeTupleDesc(oldtupdesc); + } + + slot = ExecCopySlot(epq->resultslot, slot); + } + + /* + * Shut it down ... + */ + EvalPlanQualPop(estate, subplanstate); + + return slot; +} + +/* + * Fetch a copy of the newest version of an outdated tuple + * + * estate - executor state data + * rti - rangetable index of table containing tuple + * *tid - t_ctid from the outdated tuple (ie, next updated version) + * priorXmax - t_xmax from the outdated tuple + * + * Returns a palloc'd copy of the newest tuple version, or NULL if we find + * that there is no newest version (ie, the row was deleted not updated). + * + * XXX this does not lock the new row version ... wouldn't it be better if + * it did? As-is, caller might have to repeat all its work. + */ +HeapTuple +EvalPlanQualFetch(EState *estate, Index rti, + ItemPointer tid, TransactionId priorXmax) +{ + HeapTuple copyTuple = NULL; Relation relation; HeapTupleData tuple; - HeapTuple copyTuple = NULL; SnapshotData SnapshotDirty; - bool endNode; Assert(rti != 0); /* - * find relation containing target tuple + * Find relation containing target tuple --- must be either a result + * relation of the query, or a SELECT FOR UPDATE target */ if (estate->es_result_relation_info != NULL && estate->es_result_relation_info->ri_RangeTableIndex == rti) @@ -1661,21 +1591,29 @@ EvalPlanQual(EState *estate, Index rti, } /* - * For UPDATE/DELETE we have to return tid of actual row we're executing - * PQ for. + * Return the copied tuple */ - *tid = tuple.t_self; + return copyTuple; +} + +/* + * Push a new level of EPQ state, and prepare to execute the given subplan + */ +void +EvalPlanQualPush(EState *estate, Index rti, PlanState *subplanstate) +{ + evalPlanQual *epq; + bool endNode; + + Assert(rti != 0); - /* - * Need to run a recheck subquery. Find or create a PQ stack entry. - */ epq = estate->es_evalPlanQual; endNode = true; if (epq != NULL && epq->rti == 0) { /* Top PQ stack entry is idle, so re-use it */ - Assert(!(estate->es_useEvalPlan) && epq->next == NULL); + Assert(epq->next == NULL); epq->rti = rti; endNode = false; } @@ -1720,6 +1658,8 @@ EvalPlanQual(EState *estate, Index rti, newepq->free = NULL; newepq->estate = NULL; newepq->planstate = NULL; + newepq->origplanstate = NULL; + newepq->resultslot = NULL; } else { @@ -1736,6 +1676,7 @@ EvalPlanQual(EState *estate, Index rti, } Assert(epq->rti == rti); + Assert(estate->es_evalPlanQual == epq); /* * Ok - we're requested for the same RTE. Unfortunately we still have to @@ -1764,7 +1705,20 @@ EvalPlanQual(EState *estate, Index rti, * es_result_relation_info) and reset locally changeable * state in the epq (including es_param_exec_vals, es_evTupleNull). */ + epq->origplanstate = subplanstate; EvalPlanQualStart(epq, estate, subplanstate->plan, epq->next); +} + +/* + * Install one test tuple into current EPQ level + */ +void +EvalPlanQualSetTuple(EState *estate, Index rti, HeapTuple tuple) +{ + evalPlanQual *epq = estate->es_evalPlanQual; + EState *epqstate; + + Assert(rti != 0); /* * free old RTE' tuple, if any, and store target tuple where relation's @@ -1773,12 +1727,13 @@ EvalPlanQual(EState *estate, Index rti, epqstate = epq->estate; if (epqstate->es_evTuple[rti - 1] != NULL) heap_freetuple(epqstate->es_evTuple[rti - 1]); - epqstate->es_evTuple[rti - 1] = copyTuple; - - return EvalPlanQualNext(estate); + epqstate->es_evTuple[rti - 1] = tuple; } -static TupleTableSlot * +/* + * Fetch the next row (if any) from EvalPlanQual testing + */ +TupleTableSlot * EvalPlanQualNext(EState *estate) { evalPlanQual *epq = estate->es_evalPlanQual; @@ -1787,39 +1742,48 @@ EvalPlanQualNext(EState *estate) Assert(epq->rti != 0); -lpqnext:; oldcontext = MemoryContextSwitchTo(epq->estate->es_query_cxt); slot = ExecProcNode(epq->planstate); MemoryContextSwitchTo(oldcontext); - /* - * No more tuples for this PQ. Continue previous one. - */ - if (TupIsNull(slot)) + return slot; +} + +/* + * Shut down and pop the specified level of EvalPlanQual machinery, + * plus any levels nested within it + */ +void +EvalPlanQualPop(EState *estate, PlanState *subplanstate) +{ + evalPlanQual *epq = estate->es_evalPlanQual; + + for (;;) { + PlanState *epqplanstate = epq->origplanstate; evalPlanQual *oldepq; + Assert(epq->rti != 0); + /* stop execution */ EvalPlanQualStop(epq); + epq->origplanstate = NULL; /* pop old PQ from the stack */ oldepq = epq->next; if (oldepq == NULL) { /* this is the first (oldest) PQ - mark as free */ epq->rti = 0; - estate->es_useEvalPlan = false; - /* and continue Query execution */ - return NULL; + break; } Assert(oldepq->rti != 0); /* push current PQ to freePQ stack */ oldepq->free = epq; epq = oldepq; estate->es_evalPlanQual = epq; - goto lpqnext; + if (epqplanstate == subplanstate) + break; } - - return slot; } static void @@ -1839,13 +1803,13 @@ EndEvalPlanQual(EState *estate) /* stop execution */ EvalPlanQualStop(epq); + epq->origplanstate = NULL; /* pop old PQ from the stack */ oldepq = epq->next; if (oldepq == NULL) { /* this is the first (oldest) PQ - mark as free */ epq->rti = 0; - estate->es_useEvalPlan = false; break; } Assert(oldepq->rti != 0); @@ -1887,11 +1851,11 @@ EvalPlanQualStart(evalPlanQual *epq, EState *estate, Plan *planTree, epqstate->es_snapshot = estate->es_snapshot; epqstate->es_crosscheck_snapshot = estate->es_crosscheck_snapshot; epqstate->es_range_table = estate->es_range_table; + epqstate->es_junkFilter = estate->es_junkFilter; epqstate->es_output_cid = estate->es_output_cid; epqstate->es_result_relations = estate->es_result_relations; epqstate->es_num_result_relations = estate->es_num_result_relations; epqstate->es_result_relation_info = estate->es_result_relation_info; - epqstate->es_junkFilter = estate->es_junkFilter; /* es_trig_target_relations must NOT be copied */ epqstate->es_param_list_info = estate->es_param_list_info; if (estate->es_plannedstmt->nParamExec > 0) @@ -2005,24 +1969,6 @@ EvalPlanQualStop(evalPlanQual *epq) epq->planstate = NULL; } -/* - * ExecGetActivePlanTree --- get the active PlanState tree from a QueryDesc - * - * Ordinarily this is just the one mentioned in the QueryDesc, but if we - * are looking at a row returned by the EvalPlanQual machinery, we need - * to look at the subsidiary state instead. - */ -PlanState * -ExecGetActivePlanTree(QueryDesc *queryDesc) -{ - EState *estate = queryDesc->estate; - - if (estate && estate->es_useEvalPlan && estate->es_evalPlanQual != NULL) - return estate->es_evalPlanQual->planstate; - else - return queryDesc->planstate; -} - /* * Support for SELECT INTO (a/k/a CREATE TABLE AS) |