#include "postgres.h"
#include "access/heapam.h"
#include "access/htup_details.h"
#include "access/tableam.h"
#include "access/xact.h"
#include "catalog/catalog.h"
#include "commands/trigger.h"
#include "executor/execPartition.h"
#include "executor/executor.h"
#include "executor/nodeModifyTable.h"
#include "foreign/fdwapi.h"
#include "miscadmin.h"
#include "nodes/nodeFuncs.h"
#include "rewrite/rewriteHandler.h"
#include "storage/bufmgr.h"
#include "storage/lmgr.h"
#include "utils/builtins.h"
#include "utils/datum.h"
#include "utils/memutils.h"
#include "utils/rel.h"

Include dependency graph for nodeModifyTable.c:

Data Structures
struct	MTTargetRelLookup

Macros
#define	MT_NRELS_HASH 64

Typedefs
typedef struct MTTargetRelLookup	MTTargetRelLookup

Functions
static void	ExecBatchInsert (ModifyTableState mtstate, ResultRelInfo resultRelInfo, TupleTableSlot slots, TupleTableSlot planSlots, int numSlots, EState *estate, bool canSetTag)

static bool	ExecOnConflictUpdate (ModifyTableState mtstate, ResultRelInfo resultRelInfo, ItemPointer conflictTid, TupleTableSlot planSlot, TupleTableSlot excludedSlot, EState estate, bool canSetTag, TupleTableSlot *returning)

static TupleTableSlot *	ExecPrepareTupleRouting (ModifyTableState mtstate, EState estate, PartitionTupleRouting proute, ResultRelInfo targetRelInfo, TupleTableSlot slot, ResultRelInfo *partRelInfo)

static void	ExecCheckPlanOutput (Relation resultRel, List *targetList)

static TupleTableSlot *	ExecProcessReturning (ResultRelInfo resultRelInfo, TupleTableSlot tupleSlot, TupleTableSlot *planSlot)

static void	ExecCheckTupleVisible (EState estate, Relation rel, TupleTableSlot slot)

static void	ExecCheckTIDVisible (EState estate, ResultRelInfo relinfo, ItemPointer tid, TupleTableSlot *tempSlot)

void	ExecComputeStoredGenerated (ResultRelInfo resultRelInfo, EState estate, TupleTableSlot *slot, CmdType cmdtype)

static void	ExecInitInsertProjection (ModifyTableState mtstate, ResultRelInfo resultRelInfo)

static void	ExecInitUpdateProjection (ModifyTableState mtstate, ResultRelInfo resultRelInfo)

static TupleTableSlot *	ExecGetInsertNewTuple (ResultRelInfo relinfo, TupleTableSlot planSlot)

TupleTableSlot *	ExecGetUpdateNewTuple (ResultRelInfo relinfo, TupleTableSlot planSlot, TupleTableSlot *oldSlot)

static TupleTableSlot *	ExecInsert (ModifyTableState mtstate, ResultRelInfo resultRelInfo, TupleTableSlot slot, TupleTableSlot planSlot, EState *estate, bool canSetTag)

static TupleTableSlot *	ExecDelete (ModifyTableState mtstate, ResultRelInfo resultRelInfo, ItemPointer tupleid, HeapTuple oldtuple, TupleTableSlot planSlot, EPQState epqstate, EState estate, bool processReturning, bool canSetTag, bool changingPart, bool tupleDeleted, TupleTableSlot **epqreturnslot)

static bool	ExecCrossPartitionUpdate (ModifyTableState mtstate, ResultRelInfo resultRelInfo, ItemPointer tupleid, HeapTuple oldtuple, TupleTableSlot slot, TupleTableSlot planSlot, EPQState epqstate, bool canSetTag, TupleTableSlot retry_slot, TupleTableSlot *inserted_tuple)

static TupleTableSlot *	ExecUpdate (ModifyTableState mtstate, ResultRelInfo resultRelInfo, ItemPointer tupleid, HeapTuple oldtuple, TupleTableSlot slot, TupleTableSlot planSlot, EPQState epqstate, EState estate, bool canSetTag)

static void	fireBSTriggers (ModifyTableState *node)

static void	fireASTriggers (ModifyTableState *node)

static void	ExecSetupTransitionCaptureState (ModifyTableState mtstate, EState estate)

static TupleTableSlot *	ExecModifyTable (PlanState *pstate)

ResultRelInfo *	ExecLookupResultRelByOid (ModifyTableState *node, Oid resultoid, bool missing_ok, bool update_cache)

ModifyTableState *	ExecInitModifyTable (ModifyTable node, EState estate, int eflags)

void	ExecEndModifyTable (ModifyTableState *node)

void	ExecReScanModifyTable (ModifyTableState *node)

Macro Definition Documentation

◆ MT_NRELS_HASH

#define MT_NRELS_HASH 64

Referenced by ExecInitModifyTable().

Typedef Documentation

◆ MTTargetRelLookup

typedef struct MTTargetRelLookup MTTargetRelLookup

Function Documentation

◆ ExecBatchInsert()

static void ExecBatchInsert	(	ModifyTableState *	mtstate,
		ResultRelInfo *	resultRelInfo,
		TupleTableSlot **	slots,
		TupleTableSlot **	planSlots,
		int	numSlots,
		EState *	estate,
		bool	canSetTag
	)

static

Definition at line 1009 of file nodeModifyTable.c.

References EState::es_processed, ExecARInsertTriggers(), FdwRoutine::ExecForeignBatchInsert, ExecWithCheckOptions(), i, ModifyTableState::mt_transition_capture, NIL, RelationGetRelid, ResultRelInfo::ri_FdwRoutine, ResultRelInfo::ri_RelationDesc, ResultRelInfo::ri_WithCheckOptions, TupleTableSlot::tts_tableOid, and WCO_VIEW_CHECK.

Referenced by ExecInsert(), and ExecModifyTable().

 {
     int         i;
     int         numInserted = numSlots;
     TupleTableSlot *slot = NULL;
     TupleTableSlot **rslots;
 
     /*
      * insert into foreign table: let the FDW do it
      */
     rslots = resultRelInfo->ri_FdwRoutine->ExecForeignBatchInsert(estate,
                                                                   resultRelInfo,
                                                                   slots,
                                                                   planSlots,
                                                                   &numInserted);
 
     for (i = 0; i < numInserted; i++)
     {
         slot = rslots[i];
 
         /*
          * AFTER ROW Triggers or RETURNING expressions might reference the
          * tableoid column, so (re-)initialize tts_tableOid before evaluating
          * them.
          */
         slot->tts_tableOid = RelationGetRelid(resultRelInfo->ri_RelationDesc);
 
         /* AFTER ROW INSERT Triggers */
         ExecARInsertTriggers(estate, resultRelInfo, slot, NIL,
                              mtstate->mt_transition_capture);
 
         /*
          * Check any WITH CHECK OPTION constraints from parent views.  See the
          * comment in ExecInsert.
          */
         if (resultRelInfo->ri_WithCheckOptions != NIL)
             ExecWithCheckOptions(WCO_VIEW_CHECK, resultRelInfo, slot, estate);
     }
 
     if (canSetTag && numInserted > 0)
         estate->es_processed += numInserted;
 }

◆ ExecCheckPlanOutput()

static void ExecCheckPlanOutput	(	Relation	resultRel,
		List *	targetList
	)

static

Definition at line 101 of file nodeModifyTable.c.

References Assert, ereport, errcode(), errdetail(), errmsg(), ERROR, TargetEntry::expr, exprType(), format_type_be(), IsA, lfirst, TupleDescData::natts, RelationGetDescr, TargetEntry::resjunk, and TupleDescAttr.

Referenced by ExecInitInsertProjection().

 {
     TupleDesc   resultDesc = RelationGetDescr(resultRel);
     int         attno = 0;
     ListCell   *lc;
 
     foreach(lc, targetList)
     {
         TargetEntry *tle = (TargetEntry *) lfirst(lc);
         Form_pg_attribute attr;
 
         Assert(!tle->resjunk);  /* caller removed junk items already */
 
         if (attno >= resultDesc->natts)
             ereport(ERROR,
                     (errcode(ERRCODE_DATATYPE_MISMATCH),
                      errmsg("table row type and query-specified row type do not match"),
                      errdetail("Query has too many columns.")));
         attr = TupleDescAttr(resultDesc, attno);
         attno++;
 
         if (!attr->attisdropped)
         {
             /* Normal case: demand type match */
             if (exprType((Node *) tle->expr) != attr->atttypid)
                 ereport(ERROR,
                         (errcode(ERRCODE_DATATYPE_MISMATCH),
                          errmsg("table row type and query-specified row type do not match"),
                          errdetail("Table has type %s at ordinal position %d, but query expects %s.",
                                    format_type_be(attr->atttypid),
                                    attno,
                                    format_type_be(exprType((Node *) tle->expr)))));
         }
         else
         {
             /*
              * For a dropped column, we can't check atttypid (it's likely 0).
              * In any case the planner has most likely inserted an INT4 null.
              * What we insist on is just *some* NULL constant.
              */
             if (!IsA(tle->expr, Const) ||
                 !((Const *) tle->expr)->constisnull)
                 ereport(ERROR,
                         (errcode(ERRCODE_DATATYPE_MISMATCH),
                          errmsg("table row type and query-specified row type do not match"),
                          errdetail("Query provides a value for a dropped column at ordinal position %d.",
                                    attno)));
         }
     }
     if (attno != resultDesc->natts)
         ereport(ERROR,
                 (errcode(ERRCODE_DATATYPE_MISMATCH),
                  errmsg("table row type and query-specified row type do not match"),
                  errdetail("Query has too few columns.")));
 }

◆ ExecCheckTIDVisible()

static void ExecCheckTIDVisible	(	EState *	estate,
		ResultRelInfo *	relinfo,
		ItemPointer	tid,
		TupleTableSlot *	tempSlot
	)

static

Definition at line 236 of file nodeModifyTable.c.

References elog, ERROR, ExecCheckTupleVisible(), ExecClearTuple(), IsolationUsesXactSnapshot, ResultRelInfo::ri_RelationDesc, SnapshotAny, and table_tuple_fetch_row_version().

Referenced by ExecInsert().

 {
     Relation    rel = relinfo->ri_RelationDesc;
 
     /* Redundantly check isolation level */
     if (!IsolationUsesXactSnapshot())
         return;
 
     if (!table_tuple_fetch_row_version(rel, tid, SnapshotAny, tempSlot))
         elog(ERROR, "failed to fetch conflicting tuple for ON CONFLICT");
     ExecCheckTupleVisible(estate, rel, tempSlot);
     ExecClearTuple(tempSlot);
 }

◆ ExecCheckTupleVisible()

static void ExecCheckTupleVisible	(	EState *	estate,
		Relation	rel,
		TupleTableSlot *	slot
	)

static

Definition at line 202 of file nodeModifyTable.c.

References Assert, DatumGetTransactionId, ereport, errcode(), errmsg(), ERROR, EState::es_snapshot, IsolationUsesXactSnapshot, MinTransactionIdAttributeNumber, slot_getsysattr(), table_tuple_satisfies_snapshot(), and TransactionIdIsCurrentTransactionId().

Referenced by ExecCheckTIDVisible(), and ExecOnConflictUpdate().

 {
     if (!IsolationUsesXactSnapshot())
         return;
 
     if (!table_tuple_satisfies_snapshot(rel, slot, estate->es_snapshot))
     {
         Datum       xminDatum;
         TransactionId xmin;
         bool        isnull;
 
         xminDatum = slot_getsysattr(slot, MinTransactionIdAttributeNumber, &isnull);
         Assert(!isnull);
         xmin = DatumGetTransactionId(xminDatum);
 
         /*
          * We should not raise a serialization failure if the conflict is
          * against a tuple inserted by our own transaction, even if it's not
          * visible to our snapshot.  (This would happen, for example, if
          * conflicting keys are proposed for insertion in a single command.)
          */
         if (!TransactionIdIsCurrentTransactionId(xmin))
             ereport(ERROR,
                     (errcode(ERRCODE_T_R_SERIALIZATION_FAILURE),
                      errmsg("could not serialize access due to concurrent update")));
     }
 }

◆ ExecComputeStoredGenerated()

void ExecComputeStoredGenerated	(	ResultRelInfo *	resultRelInfo,
		EState *	estate,
		TupleTableSlot *	slot,
		CmdType	cmdtype
	)

Definition at line 257 of file nodeModifyTable.c.

References Assert, bms_is_member(), build_column_default(), CMD_UPDATE, TupleDescData::constr, datumCopy(), ExprContext::ecxt_scantuple, elog, ERROR, EState::es_query_cxt, ExecClearTuple(), ExecEvalExpr(), ExecGetExtraUpdatedCols(), ExecMaterializeSlot(), ExecPrepareExpr(), ExecStoreVirtualTuple(), FirstLowInvalidHeapAttributeNumber, GetPerTupleExprContext, GetPerTupleMemoryContext, TupleConstr::has_generated_stored, i, MemoryContextSwitchTo(), TupleDescData::natts, palloc(), RelationGetDescr, RelationGetRelationName, ResultRelInfo::ri_GeneratedExprs, ResultRelInfo::ri_NumGeneratedNeeded, ResultRelInfo::ri_RelationDesc, slot_getallattrs(), TriggerDesc::trig_update_before_row, RelationData::trigdesc, TupleTableSlot::tts_isnull, TupleTableSlot::tts_values, TupleDescAttr, val, and values.

Referenced by CopyFrom(), ExecInsert(), ExecSimpleRelationInsert(), ExecSimpleRelationUpdate(), and ExecUpdate().

 {
     Relation    rel = resultRelInfo->ri_RelationDesc;
     TupleDesc   tupdesc = RelationGetDescr(rel);
     int         natts = tupdesc->natts;
     MemoryContext oldContext;
     Datum      *values;
     bool       *nulls;
 
     Assert(tupdesc->constr && tupdesc->constr->has_generated_stored);
 
     /*
      * If first time through for this result relation, build expression
      * nodetrees for rel's stored generation expressions.  Keep them in the
      * per-query memory context so they'll survive throughout the query.
      */
     if (resultRelInfo->ri_GeneratedExprs == NULL)
     {
         oldContext = MemoryContextSwitchTo(estate->es_query_cxt);
 
         resultRelInfo->ri_GeneratedExprs =
             (ExprState **) palloc(natts * sizeof(ExprState *));
         resultRelInfo->ri_NumGeneratedNeeded = 0;
 
         for (int i = 0; i < natts; i++)
         {
             if (TupleDescAttr(tupdesc, i)->attgenerated == ATTRIBUTE_GENERATED_STORED)
             {
                 Expr       *expr;
 
                 /*
                  * If it's an update and the current column was not marked as
                  * being updated, then we can skip the computation.  But if
                  * there is a BEFORE ROW UPDATE trigger, we cannot skip
                  * because the trigger might affect additional columns.
                  */
                 if (cmdtype == CMD_UPDATE &&
                     !(rel->trigdesc && rel->trigdesc->trig_update_before_row) &&
                     !bms_is_member(i + 1 - FirstLowInvalidHeapAttributeNumber,
                                    ExecGetExtraUpdatedCols(resultRelInfo, estate)))
                 {
                     resultRelInfo->ri_GeneratedExprs[i] = NULL;
                     continue;
                 }
 
                 expr = (Expr *) build_column_default(rel, i + 1);
                 if (expr == NULL)
                     elog(ERROR, "no generation expression found for column number %d of table \"%s\"",
                          i + 1, RelationGetRelationName(rel));
 
                 resultRelInfo->ri_GeneratedExprs[i] = ExecPrepareExpr(expr, estate);
                 resultRelInfo->ri_NumGeneratedNeeded++;
             }
         }
 
         MemoryContextSwitchTo(oldContext);
     }
 
     /*
      * If no generated columns have been affected by this change, then skip
      * the rest.
      */
     if (resultRelInfo->ri_NumGeneratedNeeded == 0)
         return;
 
     oldContext = MemoryContextSwitchTo(GetPerTupleMemoryContext(estate));
 
     values = palloc(sizeof(*values) * natts);
     nulls = palloc(sizeof(*nulls) * natts);
 
     slot_getallattrs(slot);
     memcpy(nulls, slot->tts_isnull, sizeof(*nulls) * natts);
 
     for (int i = 0; i < natts; i++)
     {
         Form_pg_attribute attr = TupleDescAttr(tupdesc, i);
 
         if (attr->attgenerated == ATTRIBUTE_GENERATED_STORED &&
             resultRelInfo->ri_GeneratedExprs[i])
         {
             ExprContext *econtext;
             Datum       val;
             bool        isnull;
 
             econtext = GetPerTupleExprContext(estate);
             econtext->ecxt_scantuple = slot;
 
             val = ExecEvalExpr(resultRelInfo->ri_GeneratedExprs[i], econtext, &isnull);
 
             /*
              * We must make a copy of val as we have no guarantees about where
              * memory for a pass-by-reference Datum is located.
              */
             if (!isnull)
                 val = datumCopy(val, attr->attbyval, attr->attlen);
 
             values[i] = val;
             nulls[i] = isnull;
         }
         else
         {
             if (!nulls[i])
                 values[i] = datumCopy(slot->tts_values[i], attr->attbyval, attr->attlen);
         }
     }
 
     ExecClearTuple(slot);
     memcpy(slot->tts_values, values, sizeof(*values) * natts);
     memcpy(slot->tts_isnull, nulls, sizeof(*nulls) * natts);
     ExecStoreVirtualTuple(slot);
     ExecMaterializeSlot(slot);
 
     MemoryContextSwitchTo(oldContext);
 }

◆ ExecCrossPartitionUpdate()

static bool ExecCrossPartitionUpdate	(	ModifyTableState *	mtstate,
		ResultRelInfo *	resultRelInfo,
		ItemPointer	tupleid,
		HeapTuple	oldtuple,
		TupleTableSlot *	slot,
		TupleTableSlot *	planSlot,
		EPQState *	epqstate,
		bool	canSetTag,
		TupleTableSlot **	retry_slot,
		TupleTableSlot **	inserted_tuple
	)

static

Definition at line 1430 of file nodeModifyTable.c.

References Assert, TupleConversionMap::attrMap, elog, ereport, errcode(), errdetail(), errmsg(), ERROR, EState::es_query_cxt, ExecDelete(), ExecGetChildToRootMap(), ExecGetUpdateNewTuple(), ExecInitUpdateProjection(), ExecInsert(), ExecPartitionCheckEmitError(), ExecSetupPartitionTupleRouting(), execute_attr_map_slot(), MemoryContextSwitchTo(), ModifyTableState::mt_partition_tuple_routing, ModifyTableState::mt_root_tuple_slot, ModifyTableState::mt_transition_capture, ONCONFLICT_UPDATE, PlanState::plan, ModifyTableState::ps, ResultRelInfo::ri_oldTupleSlot, ResultRelInfo::ri_projectNewInfoValid, ResultRelInfo::ri_RelationDesc, ModifyTableState::rootResultRelInfo, SnapshotAny, PlanState::state, table_slot_create(), table_tuple_fetch_row_version(), TransitionCaptureState::tcs_original_insert_tuple, TupIsNull, and unlikely.

Referenced by ExecUpdate().

 {
     EState     *estate = mtstate->ps.state;
     TupleConversionMap *tupconv_map;
     bool        tuple_deleted;
     TupleTableSlot *epqslot = NULL;
 
     *inserted_tuple = NULL;
     *retry_slot = NULL;
 
     /*
      * Disallow an INSERT ON CONFLICT DO UPDATE that causes the original row
      * to migrate to a different partition.  Maybe this can be implemented
      * some day, but it seems a fringe feature with little redeeming value.
      */
     if (((ModifyTable *) mtstate->ps.plan)->onConflictAction == ONCONFLICT_UPDATE)
         ereport(ERROR,
                 (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
                  errmsg("invalid ON UPDATE specification"),
                  errdetail("The result tuple would appear in a different partition than the original tuple.")));
 
     /*
      * When an UPDATE is run directly on a leaf partition, simply fail with a
      * partition constraint violation error.
      */
     if (resultRelInfo == mtstate->rootResultRelInfo)
         ExecPartitionCheckEmitError(resultRelInfo, slot, estate);
 
     /* Initialize tuple routing info if not already done. */
     if (mtstate->mt_partition_tuple_routing == NULL)
     {
         Relation    rootRel = mtstate->rootResultRelInfo->ri_RelationDesc;
         MemoryContext oldcxt;
 
         /* Things built here have to last for the query duration. */
         oldcxt = MemoryContextSwitchTo(estate->es_query_cxt);
 
         mtstate->mt_partition_tuple_routing =
             ExecSetupPartitionTupleRouting(estate, rootRel);
 
         /*
          * Before a partition's tuple can be re-routed, it must first be
          * converted to the root's format, so we'll need a slot for storing
          * such tuples.
          */
         Assert(mtstate->mt_root_tuple_slot == NULL);
         mtstate->mt_root_tuple_slot = table_slot_create(rootRel, NULL);
 
         MemoryContextSwitchTo(oldcxt);
     }
 
     /*
      * Row movement, part 1.  Delete the tuple, but skip RETURNING processing.
      * We want to return rows from INSERT.
      */
     ExecDelete(mtstate, resultRelInfo, tupleid, oldtuple, planSlot,
                epqstate, estate,
                false,           /* processReturning */
                false,           /* canSetTag */
                true,            /* changingPart */
                &tuple_deleted, &epqslot);
 
     /*
      * For some reason if DELETE didn't happen (e.g. trigger prevented it, or
      * it was already deleted by self, or it was concurrently deleted by
      * another transaction), then we should skip the insert as well;
      * otherwise, an UPDATE could cause an increase in the total number of
      * rows across all partitions, which is clearly wrong.
      *
      * For a normal UPDATE, the case where the tuple has been the subject of a
      * concurrent UPDATE or DELETE would be handled by the EvalPlanQual
      * machinery, but for an UPDATE that we've translated into a DELETE from
      * this partition and an INSERT into some other partition, that's not
      * available, because CTID chains can't span relation boundaries.  We
      * mimic the semantics to a limited extent by skipping the INSERT if the
      * DELETE fails to find a tuple.  This ensures that two concurrent
      * attempts to UPDATE the same tuple at the same time can't turn one tuple
      * into two, and that an UPDATE of a just-deleted tuple can't resurrect
      * it.
      */
     if (!tuple_deleted)
     {
         /*
          * epqslot will be typically NULL.  But when ExecDelete() finds that
          * another transaction has concurrently updated the same row, it
          * re-fetches the row, skips the delete, and epqslot is set to the
          * re-fetched tuple slot.  In that case, we need to do all the checks
          * again.
          */
         if (TupIsNull(epqslot))
             return true;
         else
         {
             /* Fetch the most recent version of old tuple. */
             TupleTableSlot *oldSlot;
 
             /* ... but first, make sure ri_oldTupleSlot is initialized. */
             if (unlikely(!resultRelInfo->ri_projectNewInfoValid))
                 ExecInitUpdateProjection(mtstate, resultRelInfo);
             oldSlot = resultRelInfo->ri_oldTupleSlot;
             if (!table_tuple_fetch_row_version(resultRelInfo->ri_RelationDesc,
                                                tupleid,
                                                SnapshotAny,
                                                oldSlot))
                 elog(ERROR, "failed to fetch tuple being updated");
             *retry_slot = ExecGetUpdateNewTuple(resultRelInfo, epqslot,
                                                 oldSlot);
             return false;
         }
     }
 
     /*
      * resultRelInfo is one of the per-relation resultRelInfos.  So we should
      * convert the tuple into root's tuple descriptor if needed, since
      * ExecInsert() starts the search from root.
      */
     tupconv_map = ExecGetChildToRootMap(resultRelInfo);
     if (tupconv_map != NULL)
         slot = execute_attr_map_slot(tupconv_map->attrMap,
                                      slot,
                                      mtstate->mt_root_tuple_slot);
 
     /* Tuple routing starts from the root table. */
     *inserted_tuple = ExecInsert(mtstate, mtstate->rootResultRelInfo, slot,
                                  planSlot, estate, canSetTag);
 
     /*
      * Reset the transition state that may possibly have been written by
      * INSERT.
      */
     if (mtstate->mt_transition_capture)
         mtstate->mt_transition_capture->tcs_original_insert_tuple = NULL;
 
     /* We're done moving. */
     return true;
 }

◆ ExecDelete()

static TupleTableSlot* ExecDelete	(	ModifyTableState *	mtstate,
		ResultRelInfo *	resultRelInfo,
		ItemPointer	tupleid,
		HeapTuple	oldtuple,
		TupleTableSlot *	planSlot,
		EPQState *	epqstate,
		EState *	estate,
		bool	processReturning,
		bool	canSetTag,
		bool	changingPart,
		bool *	tupleDeleted,
		TupleTableSlot **	epqreturnslot
	)

static

Definition at line 1081 of file nodeModifyTable.c.

References Assert, TM_FailureData::cmax, CMD_UPDATE, elog, ereport, errcode(), errhint(), errmsg(), ERROR, EState::es_crosscheck_snapshot, EState::es_output_cid, EState::es_processed, EState::es_snapshot, EvalPlanQual(), EvalPlanQualBegin(), EvalPlanQualSlot(), ExecARDeleteTriggers(), ExecARUpdateTriggers(), ExecBRDeleteTriggers(), ExecClearTuple(), ExecForceStoreHeapTuple(), FdwRoutine::ExecForeignDelete, ExecGetReturningSlot(), ExecIRDeleteTriggers(), ExecMaterializeSlot(), ExecProcessReturning(), ExecStoreAllNullTuple(), IsolationUsesXactSnapshot, LockTupleExclusive, LockWaitBlock, ModifyTableState::mt_transition_capture, ModifyTableState::operation, RelationGetRelid, ResultRelInfo::ri_FdwRoutine, ResultRelInfo::ri_projectReturning, ResultRelInfo::ri_RangeTableIndex, ResultRelInfo::ri_RelationDesc, ResultRelInfo::ri_TrigDesc, SnapshotAny, table_tuple_delete(), table_tuple_fetch_row_version(), table_tuple_lock(), TransitionCaptureState::tcs_update_old_table, TM_Deleted, TM_Ok, TM_SelfModified, TM_Updated, TM_FailureData::traversed, TriggerDesc::trig_delete_before_row, TriggerDesc::trig_delete_instead_row, TTS_EMPTY, TupleTableSlot::tts_tableOid, TupIsNull, and TUPLE_LOCK_FLAG_FIND_LAST_VERSION.

Referenced by ExecCrossPartitionUpdate(), and ExecModifyTable().

 {
     Relation    resultRelationDesc = resultRelInfo->ri_RelationDesc;
     TM_Result   result;
     TM_FailureData tmfd;
     TupleTableSlot *slot = NULL;
     TransitionCaptureState *ar_delete_trig_tcs;
 
     if (tupleDeleted)
         *tupleDeleted = false;
 
     /* BEFORE ROW DELETE Triggers */
     if (resultRelInfo->ri_TrigDesc &&
         resultRelInfo->ri_TrigDesc->trig_delete_before_row)
     {
         bool        dodelete;
 
         dodelete = ExecBRDeleteTriggers(estate, epqstate, resultRelInfo,
                                         tupleid, oldtuple, epqreturnslot);
 
         if (!dodelete)          /* "do nothing" */
             return NULL;
     }
 
     /* INSTEAD OF ROW DELETE Triggers */
     if (resultRelInfo->ri_TrigDesc &&
         resultRelInfo->ri_TrigDesc->trig_delete_instead_row)
     {
         bool        dodelete;
 
         Assert(oldtuple != NULL);
         dodelete = ExecIRDeleteTriggers(estate, resultRelInfo, oldtuple);
 
         if (!dodelete)          /* "do nothing" */
             return NULL;
     }
     else if (resultRelInfo->ri_FdwRoutine)
     {
         /*
          * delete from foreign table: let the FDW do it
          *
          * We offer the returning slot as a place to store RETURNING data,
          * although the FDW can return some other slot if it wants.
          */
         slot = ExecGetReturningSlot(estate, resultRelInfo);
         slot = resultRelInfo->ri_FdwRoutine->ExecForeignDelete(estate,
                                                                resultRelInfo,
                                                                slot,
                                                                planSlot);
 
         if (slot == NULL)       /* "do nothing" */
             return NULL;
 
         /*
          * RETURNING expressions might reference the tableoid column, so
          * (re)initialize tts_tableOid before evaluating them.
          */
         if (TTS_EMPTY(slot))
             ExecStoreAllNullTuple(slot);
 
         slot->tts_tableOid = RelationGetRelid(resultRelationDesc);
     }
     else
     {
         /*
          * delete the tuple
          *
          * Note: if es_crosscheck_snapshot isn't InvalidSnapshot, we check
          * that the row to be deleted is visible to that snapshot, and throw a
          * can't-serialize error if not. This is a special-case behavior
          * needed for referential integrity updates in transaction-snapshot
          * mode transactions.
          */
 ldelete:;
         result = table_tuple_delete(resultRelationDesc, tupleid,
                                     estate->es_output_cid,
                                     estate->es_snapshot,
                                     estate->es_crosscheck_snapshot,
                                     true /* wait for commit */ ,
                                     &tmfd,
                                     changingPart);
 
         switch (result)
         {
             case TM_SelfModified:
 
                 /*
                  * The target tuple was already updated or deleted by the
                  * current command, or by a later command in the current
                  * transaction.  The former case is possible in a join DELETE
                  * where multiple tuples join to the same target tuple. This
                  * is somewhat questionable, but Postgres has always allowed
                  * it: we just ignore additional deletion attempts.
                  *
                  * The latter case arises if the tuple is modified by a
                  * command in a BEFORE trigger, or perhaps by a command in a
                  * volatile function used in the query.  In such situations we
                  * should not ignore the deletion, but it is equally unsafe to
                  * proceed.  We don't want to discard the original DELETE
                  * while keeping the triggered actions based on its deletion;
                  * and it would be no better to allow the original DELETE
                  * while discarding updates that it triggered.  The row update
                  * carries some information that might be important according
                  * to business rules; so throwing an error is the only safe
                  * course.
                  *
                  * If a trigger actually intends this type of interaction, it
                  * can re-execute the DELETE and then return NULL to cancel
                  * the outer delete.
                  */
                 if (tmfd.cmax != estate->es_output_cid)
                     ereport(ERROR,
                             (errcode(ERRCODE_TRIGGERED_DATA_CHANGE_VIOLATION),
                              errmsg("tuple to be deleted was already modified by an operation triggered by the current command"),
                              errhint("Consider using an AFTER trigger instead of a BEFORE trigger to propagate changes to other rows.")));
 
                 /* Else, already deleted by self; nothing to do */
                 return NULL;
 
             case TM_Ok:
                 break;
 
             case TM_Updated:
                 {
                     TupleTableSlot *inputslot;
                     TupleTableSlot *epqslot;
 
                     if (IsolationUsesXactSnapshot())
                         ereport(ERROR,
                                 (errcode(ERRCODE_T_R_SERIALIZATION_FAILURE),
                                  errmsg("could not serialize access due to concurrent update")));
 
                     /*
                      * Already know that we're going to need to do EPQ, so
                      * fetch tuple directly into the right slot.
                      */
                     EvalPlanQualBegin(epqstate);
                     inputslot = EvalPlanQualSlot(epqstate, resultRelationDesc,
                                                  resultRelInfo->ri_RangeTableIndex);
 
                     result = table_tuple_lock(resultRelationDesc, tupleid,
                                               estate->es_snapshot,
                                               inputslot, estate->es_output_cid,
                                               LockTupleExclusive, LockWaitBlock,
                                               TUPLE_LOCK_FLAG_FIND_LAST_VERSION,
                                               &tmfd);
 
                     switch (result)
                     {
                         case TM_Ok:
                             Assert(tmfd.traversed);
                             epqslot = EvalPlanQual(epqstate,
                                                    resultRelationDesc,
                                                    resultRelInfo->ri_RangeTableIndex,
                                                    inputslot);
                             if (TupIsNull(epqslot))
                                 /* Tuple not passing quals anymore, exiting... */
                                 return NULL;
 
                             /*
                              * If requested, skip delete and pass back the
                              * updated row.
                              */
                             if (epqreturnslot)
                             {
                                 *epqreturnslot = epqslot;
                                 return NULL;
                             }
                             else
                                 goto ldelete;
 
                         case TM_SelfModified:
 
                             /*
                              * This can be reached when following an update
                              * chain from a tuple updated by another session,
                              * reaching a tuple that was already updated in
                              * this transaction. If previously updated by this
                              * command, ignore the delete, otherwise error
                              * out.
                              *
                              * See also TM_SelfModified response to
                              * table_tuple_delete() above.
                              */
                             if (tmfd.cmax != estate->es_output_cid)
                                 ereport(ERROR,
                                         (errcode(ERRCODE_TRIGGERED_DATA_CHANGE_VIOLATION),
                                          errmsg("tuple to be deleted was already modified by an operation triggered by the current command"),
                                          errhint("Consider using an AFTER trigger instead of a BEFORE trigger to propagate changes to other rows.")));
                             return NULL;
 
                         case TM_Deleted:
                             /* tuple already deleted; nothing to do */
                             return NULL;
 
                         default:
 
                             /*
                              * TM_Invisible should be impossible because we're
                              * waiting for updated row versions, and would
                              * already have errored out if the first version
                              * is invisible.
                              *
                              * TM_Updated should be impossible, because we're
                              * locking the latest version via
                              * TUPLE_LOCK_FLAG_FIND_LAST_VERSION.
                              */
                             elog(ERROR, "unexpected table_tuple_lock status: %u",
                                  result);
                             return NULL;
                     }
 
                     Assert(false);
                     break;
                 }
 
             case TM_Deleted:
                 if (IsolationUsesXactSnapshot())
                     ereport(ERROR,
                             (errcode(ERRCODE_T_R_SERIALIZATION_FAILURE),
                              errmsg("could not serialize access due to concurrent delete")));
                 /* tuple already deleted; nothing to do */
                 return NULL;
 
             default:
                 elog(ERROR, "unrecognized table_tuple_delete status: %u",
                      result);
                 return NULL;
         }
 
         /*
          * Note: Normally one would think that we have to delete index tuples
          * associated with the heap tuple now...
          *
          * ... but in POSTGRES, we have no need to do this because VACUUM will
          * take care of it later.  We can't delete index tuples immediately
          * anyway, since the tuple is still visible to other transactions.
          */
     }
 
     if (canSetTag)
         (estate->es_processed)++;
 
     /* Tell caller that the delete actually happened. */
     if (tupleDeleted)
         *tupleDeleted = true;
 
     /*
      * If this delete is the result of a partition key update that moved the
      * tuple to a new partition, put this row into the transition OLD TABLE,
      * if there is one. We need to do this separately for DELETE and INSERT
      * because they happen on different tables.
      */
     ar_delete_trig_tcs = mtstate->mt_transition_capture;
     if (mtstate->operation == CMD_UPDATE && mtstate->mt_transition_capture
         && mtstate->mt_transition_capture->tcs_update_old_table)
     {
         ExecARUpdateTriggers(estate, resultRelInfo,
                              tupleid,
                              oldtuple,
                              NULL,
                              NULL,
                              mtstate->mt_transition_capture);
 
         /*
          * We've already captured the NEW TABLE row, so make sure any AR
          * DELETE trigger fired below doesn't capture it again.
          */
         ar_delete_trig_tcs = NULL;
     }
 
     /* AFTER ROW DELETE Triggers */
     ExecARDeleteTriggers(estate, resultRelInfo, tupleid, oldtuple,
                          ar_delete_trig_tcs);
 
     /* Process RETURNING if present and if requested */
     if (processReturning && resultRelInfo->ri_projectReturning)
     {
         /*
          * We have to put the target tuple into a slot, which means first we
          * gotta fetch it.  We can use the trigger tuple slot.
          */
         TupleTableSlot *rslot;
 
         if (resultRelInfo->ri_FdwRoutine)
         {
             /* FDW must have provided a slot containing the deleted row */
             Assert(!TupIsNull(slot));
         }
         else
         {
             slot = ExecGetReturningSlot(estate, resultRelInfo);
             if (oldtuple != NULL)
             {
                 ExecForceStoreHeapTuple(oldtuple, slot, false);
             }
             else
             {
                 if (!table_tuple_fetch_row_version(resultRelationDesc, tupleid,
                                                    SnapshotAny, slot))
                     elog(ERROR, "failed to fetch deleted tuple for DELETE RETURNING");
             }
         }
 
         rslot = ExecProcessReturning(resultRelInfo, slot, planSlot);
 
         /*
          * Before releasing the target tuple again, make sure rslot has a
          * local copy of any pass-by-reference values.
          */
         ExecMaterializeSlot(rslot);
 
         ExecClearTuple(slot);
 
         return rslot;
     }
 
     return NULL;
 }

◆ ExecEndModifyTable()

void ExecEndModifyTable ( ModifyTableState * node )

Definition at line 3169 of file nodeModifyTable.c.

References FdwRoutine::EndForeignModify, EvalPlanQualEnd(), ExecCleanupTupleRouting(), ExecClearTuple(), ExecDropSingleTupleTableSlot(), ExecEndNode(), ExecFreeExprContext(), i, ModifyTableState::mt_epqstate, ModifyTableState::mt_nrels, ModifyTableState::mt_partition_tuple_routing, ModifyTableState::mt_root_tuple_slot, outerPlanState, ModifyTableState::ps, PlanState::ps_ResultTupleSlot, ModifyTableState::resultRelInfo, ResultRelInfo::ri_FdwRoutine, ResultRelInfo::ri_NumSlotsInitialized, ResultRelInfo::ri_PlanSlots, ResultRelInfo::ri_Slots, ResultRelInfo::ri_usesFdwDirectModify, and PlanState::state.

Referenced by ExecEndNode().

 {
     int         i;
 
     /*
      * Allow any FDWs to shut down
      */
     for (i = 0; i < node->mt_nrels; i++)
     {
         int         j;
         ResultRelInfo *resultRelInfo = node->resultRelInfo + i;
 
         if (!resultRelInfo->ri_usesFdwDirectModify &&
             resultRelInfo->ri_FdwRoutine != NULL &&
             resultRelInfo->ri_FdwRoutine->EndForeignModify != NULL)
             resultRelInfo->ri_FdwRoutine->EndForeignModify(node->ps.state,
                                                            resultRelInfo);
 
         /*
          * Cleanup the initialized batch slots. This only matters for FDWs
          * with batching, but the other cases will have ri_NumSlotsInitialized
          * == 0.
          */
         for (j = 0; j < resultRelInfo->ri_NumSlotsInitialized; j++)
         {
             ExecDropSingleTupleTableSlot(resultRelInfo->ri_Slots[j]);
             ExecDropSingleTupleTableSlot(resultRelInfo->ri_PlanSlots[j]);
         }
     }
 
     /*
      * Close all the partitioned tables, leaf partitions, and their indices
      * and release the slot used for tuple routing, if set.
      */
     if (node->mt_partition_tuple_routing)
     {
         ExecCleanupTupleRouting(node, node->mt_partition_tuple_routing);
 
         if (node->mt_root_tuple_slot)
             ExecDropSingleTupleTableSlot(node->mt_root_tuple_slot);
     }
 
     /*
      * Free the exprcontext
      */
     ExecFreeExprContext(&node->ps);
 
     /*
      * clean out the tuple table
      */
     if (node->ps.ps_ResultTupleSlot)
         ExecClearTuple(node->ps.ps_ResultTupleSlot);
 
     /*
      * Terminate EPQ execution if active
      */
     EvalPlanQualEnd(&node->mt_epqstate);
 
     /*
      * shut down subplan
      */
     ExecEndNode(outerPlanState(node));
 }

◆ ExecGetInsertNewTuple()

static TupleTableSlot* ExecGetInsertNewTuple	(	ResultRelInfo *	relinfo,
		TupleTableSlot *	planSlot
	)

static

Definition at line 512 of file nodeModifyTable.c.

References ExprContext::ecxt_outertuple, ExecCopySlot(), ExecProject(), ProjectionInfo::pi_exprContext, ResultRelInfo::ri_newTupleSlot, ResultRelInfo::ri_projectNew, and TupleTableSlot::tts_ops.

Referenced by ExecModifyTable().

 {
     ProjectionInfo *newProj = relinfo->ri_projectNew;
     ExprContext *econtext;
 
     /*
      * If there's no projection to be done, just make sure the slot is of the
      * right type for the target rel.  If the planSlot is the right type we
      * can use it as-is, else copy the data into ri_newTupleSlot.
      */
     if (newProj == NULL)
     {
         if (relinfo->ri_newTupleSlot->tts_ops != planSlot->tts_ops)
         {
             ExecCopySlot(relinfo->ri_newTupleSlot, planSlot);
             return relinfo->ri_newTupleSlot;
         }
         else
             return planSlot;
     }
 
     /*
      * Else project; since the projection output slot is ri_newTupleSlot, this
      * will also fix any slot-type problem.
      *
      * Note: currently, this is dead code, because INSERT cases don't receive
      * any junk columns so there's never a projection to be done.
      */
     econtext = newProj->pi_exprContext;
     econtext->ecxt_outertuple = planSlot;
     return ExecProject(newProj);
 }

◆ ExecGetUpdateNewTuple()

TupleTableSlot* ExecGetUpdateNewTuple	(	ResultRelInfo *	relinfo,
		TupleTableSlot *	planSlot,
		TupleTableSlot *	oldSlot
	)

Definition at line 556 of file nodeModifyTable.c.

References Assert, ExprContext::ecxt_outertuple, ExprContext::ecxt_scantuple, ExecProject(), ProjectionInfo::pi_exprContext, ResultRelInfo::ri_projectNew, ResultRelInfo::ri_projectNewInfoValid, and TTS_EMPTY.

Referenced by exec_rt_fetch(), ExecBRUpdateTriggers(), ExecCrossPartitionUpdate(), ExecModifyTable(), and ExecUpdate().

 {
     ProjectionInfo *newProj = relinfo->ri_projectNew;
     ExprContext *econtext;
 
     /* Use a few extra Asserts to protect against outside callers */
     Assert(relinfo->ri_projectNewInfoValid);
     Assert(planSlot != NULL && !TTS_EMPTY(planSlot));
     Assert(oldSlot != NULL && !TTS_EMPTY(oldSlot));
 
     econtext = newProj->pi_exprContext;
     econtext->ecxt_outertuple = planSlot;
     econtext->ecxt_scantuple = oldSlot;
     return ExecProject(newProj);
 }

◆ ExecInitInsertProjection()

static void ExecInitInsertProjection	(	ModifyTableState *	mtstate,
		ResultRelInfo *	resultRelInfo
	)

static

Definition at line 384 of file nodeModifyTable.c.

References EState::es_tupleTable, ExecAssignExprContext(), ExecBuildProjectionInfo(), ExecCheckPlanOutput(), lappend(), lfirst, NIL, outerPlan, PlanState::plan, ModifyTableState::ps, PlanState::ps_ExprContext, RelationGetDescr, TargetEntry::resjunk, ResultRelInfo::ri_newTupleSlot, ResultRelInfo::ri_projectNew, ResultRelInfo::ri_projectNewInfoValid, ResultRelInfo::ri_RelationDesc, PlanState::state, and table_slot_create().

Referenced by ExecModifyTable().

 {
     ModifyTable *node = (ModifyTable *) mtstate->ps.plan;
     Plan       *subplan = outerPlan(node);
     EState     *estate = mtstate->ps.state;
     List       *insertTargetList = NIL;
     bool        need_projection = false;
     ListCell   *l;
 
     /* Extract non-junk columns of the subplan's result tlist. */
     foreach(l, subplan->targetlist)
     {
         TargetEntry *tle = (TargetEntry *) lfirst(l);
 
         if (!tle->resjunk)
             insertTargetList = lappend(insertTargetList, tle);
         else
             need_projection = true;
     }
 
     /*
      * The junk-free list must produce a tuple suitable for the result
      * relation.
      */
     ExecCheckPlanOutput(resultRelInfo->ri_RelationDesc, insertTargetList);
 
     /* We'll need a slot matching the table's format. */
     resultRelInfo->ri_newTupleSlot =
         table_slot_create(resultRelInfo->ri_RelationDesc,
                           &estate->es_tupleTable);
 
     /* Build ProjectionInfo if needed (it probably isn't). */
     if (need_projection)
     {
         TupleDesc   relDesc = RelationGetDescr(resultRelInfo->ri_RelationDesc);
 
         /* need an expression context to do the projection */
         if (mtstate->ps.ps_ExprContext == NULL)
             ExecAssignExprContext(estate, &mtstate->ps);
 
         resultRelInfo->ri_projectNew =
             ExecBuildProjectionInfo(insertTargetList,
                                     mtstate->ps.ps_ExprContext,
                                     resultRelInfo->ri_newTupleSlot,
                                     &mtstate->ps,
                                     relDesc);
     }
 
     resultRelInfo->ri_projectNewInfoValid = true;
 }

◆ ExecInitModifyTable()

ModifyTableState* ExecInitModifyTable	(	ModifyTable *	node,
		EState *	estate,
		int	eflags
	)

Definition at line 2708 of file nodeModifyTable.c.

Referenced by ExecInitNode().

 {
     ModifyTableState *mtstate;
     Plan       *subplan = outerPlan(node);
     CmdType     operation = node->operation;
     int         nrels = list_length(node->resultRelations);
     ResultRelInfo *resultRelInfo;
     List       *arowmarks;
     ListCell   *l;
     int         i;
     Relation    rel;
 
     /* check for unsupported flags */
     Assert(!(eflags & (EXEC_FLAG_BACKWARD | EXEC_FLAG_MARK)));
 
     /*
      * create state structure
      */
     mtstate = makeNode(ModifyTableState);
     mtstate->ps.plan = (Plan *) node;
     mtstate->ps.state = estate;
     mtstate->ps.ExecProcNode = ExecModifyTable;
 
     mtstate->operation = operation;
     mtstate->canSetTag = node->canSetTag;
     mtstate->mt_done = false;
 
     mtstate->mt_nrels = nrels;
     mtstate->resultRelInfo = (ResultRelInfo *)
         palloc(nrels * sizeof(ResultRelInfo));
 
     /*----------
      * Resolve the target relation. This is the same as:
      *
      * - the relation for which we will fire FOR STATEMENT triggers,
      * - the relation into whose tuple format all captured transition tuples
      *   must be converted, and
      * - the root partitioned table used for tuple routing.
      *
      * If it's a partitioned table, the root partition doesn't appear
      * elsewhere in the plan and its RT index is given explicitly in
      * node->rootRelation.  Otherwise (i.e. table inheritance) the target
      * relation is the first relation in the node->resultRelations list.
      *----------
      */
     if (node->rootRelation > 0)
     {
         mtstate->rootResultRelInfo = makeNode(ResultRelInfo);
         ExecInitResultRelation(estate, mtstate->rootResultRelInfo,
                                node->rootRelation);
     }
     else
     {
         mtstate->rootResultRelInfo = mtstate->resultRelInfo;
         ExecInitResultRelation(estate, mtstate->resultRelInfo,
                                linitial_int(node->resultRelations));
     }
 
     /* set up epqstate with dummy subplan data for the moment */
     EvalPlanQualInit(&mtstate->mt_epqstate, estate, NULL, NIL, node->epqParam);
     mtstate->fireBSTriggers = true;
 
     /*
      * Build state for collecting transition tuples.  This requires having a
      * valid trigger query context, so skip it in explain-only mode.
      */
     if (!(eflags & EXEC_FLAG_EXPLAIN_ONLY))
         ExecSetupTransitionCaptureState(mtstate, estate);
 
     /*
      * Open all the result relations and initialize the ResultRelInfo structs.
      * (But root relation was initialized above, if it's part of the array.)
      * We must do this before initializing the subplan, because direct-modify
      * FDWs expect their ResultRelInfos to be available.
      */
     resultRelInfo = mtstate->resultRelInfo;
     i = 0;
     foreach(l, node->resultRelations)
     {
         Index       resultRelation = lfirst_int(l);
 
         if (resultRelInfo != mtstate->rootResultRelInfo)
         {
             ExecInitResultRelation(estate, resultRelInfo, resultRelation);
 
             /*
              * For child result relations, store the root result relation
              * pointer.  We do so for the convenience of places that want to
              * look at the query's original target relation but don't have the
              * mtstate handy.
              */
             resultRelInfo->ri_RootResultRelInfo = mtstate->rootResultRelInfo;
         }
 
         /* Initialize the usesFdwDirectModify flag */
         resultRelInfo->ri_usesFdwDirectModify = bms_is_member(i,
                                                               node->fdwDirectModifyPlans);
 
         /*
          * Verify result relation is a valid target for the current operation
          */
         CheckValidResultRel(resultRelInfo, operation);
 
         resultRelInfo++;
         i++;
     }
 
     /*
      * Now we may initialize the subplan.
      */
     outerPlanState(mtstate) = ExecInitNode(subplan, estate, eflags);
 
     /*
      * Do additional per-result-relation initialization.
      */
     for (i = 0; i < nrels; i++)
     {
         resultRelInfo = &mtstate->resultRelInfo[i];
 
         /* Let FDWs init themselves for foreign-table result rels */
         if (!resultRelInfo->ri_usesFdwDirectModify &&
             resultRelInfo->ri_FdwRoutine != NULL &&
             resultRelInfo->ri_FdwRoutine->BeginForeignModify != NULL)
         {
             List       *fdw_private = (List *) list_nth(node->fdwPrivLists, i);
 
             resultRelInfo->ri_FdwRoutine->BeginForeignModify(mtstate,
                                                              resultRelInfo,
                                                              fdw_private,
                                                              i,
                                                              eflags);
         }
 
         /*
          * For UPDATE/DELETE, find the appropriate junk attr now, either a
          * 'ctid' or 'wholerow' attribute depending on relkind.  For foreign
          * tables, the FDW might have created additional junk attr(s), but
          * those are no concern of ours.
          */
         if (operation == CMD_UPDATE || operation == CMD_DELETE)
         {
             char        relkind;
 
             relkind = resultRelInfo->ri_RelationDesc->rd_rel->relkind;
             if (relkind == RELKIND_RELATION ||
                 relkind == RELKIND_MATVIEW ||
                 relkind == RELKIND_PARTITIONED_TABLE)
             {
                 resultRelInfo->ri_RowIdAttNo =
                     ExecFindJunkAttributeInTlist(subplan->targetlist, "ctid");
                 if (!AttributeNumberIsValid(resultRelInfo->ri_RowIdAttNo))
                     elog(ERROR, "could not find junk ctid column");
             }
             else if (relkind == RELKIND_FOREIGN_TABLE)
             {
                 /*
                  * When there is a row-level trigger, there should be a
                  * wholerow attribute.  We also require it to be present in
                  * UPDATE, so we can get the values of unchanged columns.
                  */
                 resultRelInfo->ri_RowIdAttNo =
                     ExecFindJunkAttributeInTlist(subplan->targetlist,
                                                  "wholerow");
                 if (mtstate->operation == CMD_UPDATE &&
                     !AttributeNumberIsValid(resultRelInfo->ri_RowIdAttNo))
                     elog(ERROR, "could not find junk wholerow column");
             }
             else
             {
                 /* Other valid target relkinds must provide wholerow */
                 resultRelInfo->ri_RowIdAttNo =
                     ExecFindJunkAttributeInTlist(subplan->targetlist,
                                                  "wholerow");
                 if (!AttributeNumberIsValid(resultRelInfo->ri_RowIdAttNo))
                     elog(ERROR, "could not find junk wholerow column");
             }
         }
     }
 
     /*
      * If this is an inherited update/delete, there will be a junk attribute
      * named "tableoid" present in the subplan's targetlist.  It will be used
      * to identify the result relation for a given tuple to be
      * updated/deleted.
      */
     mtstate->mt_resultOidAttno =
         ExecFindJunkAttributeInTlist(subplan->targetlist, "tableoid");
     Assert(AttributeNumberIsValid(mtstate->mt_resultOidAttno) || nrels == 1);
     mtstate->mt_lastResultOid = InvalidOid; /* force lookup at first tuple */
     mtstate->mt_lastResultIndex = 0;    /* must be zero if no such attr */
 
     /* Get the root target relation */
     rel = mtstate->rootResultRelInfo->ri_RelationDesc;
 
     /*
      * Build state for tuple routing if it's a partitioned INSERT.  An UPDATE
      * might need this too, but only if it actually moves tuples between
      * partitions; in that case setup is done by ExecCrossPartitionUpdate.
      */
     if (rel->rd_rel->relkind == RELKIND_PARTITIONED_TABLE &&
         operation == CMD_INSERT)
         mtstate->mt_partition_tuple_routing =
             ExecSetupPartitionTupleRouting(estate, rel);
 
     /*
      * Initialize any WITH CHECK OPTION constraints if needed.
      */
     resultRelInfo = mtstate->resultRelInfo;
     foreach(l, node->withCheckOptionLists)
     {
         List       *wcoList = (List *) lfirst(l);
         List       *wcoExprs = NIL;
         ListCell   *ll;
 
         foreach(ll, wcoList)
         {
             WithCheckOption *wco = (WithCheckOption *) lfirst(ll);
             ExprState  *wcoExpr = ExecInitQual((List *) wco->qual,
                                                &mtstate->ps);
 
             wcoExprs = lappend(wcoExprs, wcoExpr);
         }
 
         resultRelInfo->ri_WithCheckOptions = wcoList;
         resultRelInfo->ri_WithCheckOptionExprs = wcoExprs;
         resultRelInfo++;
     }
 
     /*
      * Initialize RETURNING projections if needed.
      */
     if (node->returningLists)
     {
         TupleTableSlot *slot;
         ExprContext *econtext;
 
         /*
          * Initialize result tuple slot and assign its rowtype using the first
          * RETURNING list.  We assume the rest will look the same.
          */
         mtstate->ps.plan->targetlist = (List *) linitial(node->returningLists);
 
         /* Set up a slot for the output of the RETURNING projection(s) */
         ExecInitResultTupleSlotTL(&mtstate->ps, &TTSOpsVirtual);
         slot = mtstate->ps.ps_ResultTupleSlot;
 
         /* Need an econtext too */
         if (mtstate->ps.ps_ExprContext == NULL)
             ExecAssignExprContext(estate, &mtstate->ps);
         econtext = mtstate->ps.ps_ExprContext;
 
         /*
          * Build a projection for each result rel.
          */
         resultRelInfo = mtstate->resultRelInfo;
         foreach(l, node->returningLists)
         {
             List       *rlist = (List *) lfirst(l);
 
             resultRelInfo->ri_returningList = rlist;
             resultRelInfo->ri_projectReturning =
                 ExecBuildProjectionInfo(rlist, econtext, slot, &mtstate->ps,
                                         resultRelInfo->ri_RelationDesc->rd_att);
             resultRelInfo++;
         }
     }
     else
     {
         /*
          * We still must construct a dummy result tuple type, because InitPlan
          * expects one (maybe should change that?).
          */
         mtstate->ps.plan->targetlist = NIL;
         ExecInitResultTypeTL(&mtstate->ps);
 
         mtstate->ps.ps_ExprContext = NULL;
     }
 
     /* Set the list of arbiter indexes if needed for ON CONFLICT */
     resultRelInfo = mtstate->resultRelInfo;
     if (node->onConflictAction != ONCONFLICT_NONE)
     {
         /* insert may only have one relation, inheritance is not expanded */
         Assert(nrels == 1);
         resultRelInfo->ri_onConflictArbiterIndexes = node->arbiterIndexes;
     }
 
     /*
      * If needed, Initialize target list, projection and qual for ON CONFLICT
      * DO UPDATE.
      */
     if (node->onConflictAction == ONCONFLICT_UPDATE)
     {
         OnConflictSetState *onconfl = makeNode(OnConflictSetState);
         ExprContext *econtext;
         TupleDesc   relationDesc;
 
         /* already exists if created by RETURNING processing above */
         if (mtstate->ps.ps_ExprContext == NULL)
             ExecAssignExprContext(estate, &mtstate->ps);
 
         econtext = mtstate->ps.ps_ExprContext;
         relationDesc = resultRelInfo->ri_RelationDesc->rd_att;
 
         /* create state for DO UPDATE SET operation */
         resultRelInfo->ri_onConflict = onconfl;
 
         /* initialize slot for the existing tuple */
         onconfl->oc_Existing =
             table_slot_create(resultRelInfo->ri_RelationDesc,
                               &mtstate->ps.state->es_tupleTable);
 
         /*
          * Create the tuple slot for the UPDATE SET projection. We want a slot
          * of the table's type here, because the slot will be used to insert
          * into the table, and for RETURNING processing - which may access
          * system attributes.
          */
         onconfl->oc_ProjSlot =
             table_slot_create(resultRelInfo->ri_RelationDesc,
                               &mtstate->ps.state->es_tupleTable);
 
         /* build UPDATE SET projection state */
         onconfl->oc_ProjInfo =
             ExecBuildUpdateProjection(node->onConflictSet,
                                       true,
                                       node->onConflictCols,
                                       relationDesc,
                                       econtext,
                                       onconfl->oc_ProjSlot,
                                       &mtstate->ps);
 
         /* initialize state to evaluate the WHERE clause, if any */
         if (node->onConflictWhere)
         {
             ExprState  *qualexpr;
 
             qualexpr = ExecInitQual((List *) node->onConflictWhere,
                                     &mtstate->ps);
             onconfl->oc_WhereClause = qualexpr;
         }
     }
 
     /*
      * If we have any secondary relations in an UPDATE or DELETE, they need to
      * be treated like non-locked relations in SELECT FOR UPDATE, ie, the
      * EvalPlanQual mechanism needs to be told about them.  Locate the
      * relevant ExecRowMarks.
      */
     arowmarks = NIL;
     foreach(l, node->rowMarks)
     {
         PlanRowMark *rc = lfirst_node(PlanRowMark, l);
         ExecRowMark *erm;
         ExecAuxRowMark *aerm;
 
         /* ignore "parent" rowmarks; they are irrelevant at runtime */
         if (rc->isParent)
             continue;
 
         /* Find ExecRowMark and build ExecAuxRowMark */
         erm = ExecFindRowMark(estate, rc->rti, false);
         aerm = ExecBuildAuxRowMark(erm, subplan->targetlist);
         arowmarks = lappend(arowmarks, aerm);
     }
 
     EvalPlanQualSetPlan(&mtstate->mt_epqstate, subplan, arowmarks);
 
     /*
      * If there are a lot of result relations, use a hash table to speed the
      * lookups.  If there are not a lot, a simple linear search is faster.
      *
      * It's not clear where the threshold is, but try 64 for starters.  In a
      * debugging build, use a small threshold so that we get some test
      * coverage of both code paths.
      */
 #ifdef USE_ASSERT_CHECKING
 #define MT_NRELS_HASH 4
 #else
 #define MT_NRELS_HASH 64
 #endif
     if (nrels >= MT_NRELS_HASH)
     {
         HASHCTL     hash_ctl;
 
         hash_ctl.keysize = sizeof(Oid);
         hash_ctl.entrysize = sizeof(MTTargetRelLookup);
         hash_ctl.hcxt = CurrentMemoryContext;
         mtstate->mt_resultOidHash =
             hash_create("ModifyTable target hash",
                         nrels, &hash_ctl,
                         HASH_ELEM | HASH_BLOBS | HASH_CONTEXT);
         for (i = 0; i < nrels; i++)
         {
             Oid         hashkey;
             MTTargetRelLookup *mtlookup;
             bool        found;
 
             resultRelInfo = &mtstate->resultRelInfo[i];
             hashkey = RelationGetRelid(resultRelInfo->ri_RelationDesc);
             mtlookup = (MTTargetRelLookup *)
                 hash_search(mtstate->mt_resultOidHash, &hashkey,
                             HASH_ENTER, &found);
             Assert(!found);
             mtlookup->relationIndex = i;
         }
     }
     else
         mtstate->mt_resultOidHash = NULL;
 
     /*
      * Determine if the FDW supports batch insert and determine the batch size
      * (a FDW may support batching, but it may be disabled for the
      * server/table).
      *
      * We only do this for INSERT, so that for UPDATE/DELETE the batch size
      * remains set to 0.
      */
     if (operation == CMD_INSERT)
     {
         /* insert may only have one relation, inheritance is not expanded */
         Assert(nrels == 1);
         resultRelInfo = mtstate->resultRelInfo;
         if (!resultRelInfo->ri_usesFdwDirectModify &&
             resultRelInfo->ri_FdwRoutine != NULL &&
             resultRelInfo->ri_FdwRoutine->GetForeignModifyBatchSize &&
             resultRelInfo->ri_FdwRoutine->ExecForeignBatchInsert)
         {
             resultRelInfo->ri_BatchSize =
                 resultRelInfo->ri_FdwRoutine->GetForeignModifyBatchSize(resultRelInfo);
             Assert(resultRelInfo->ri_BatchSize >= 1);
         }
         else
             resultRelInfo->ri_BatchSize = 1;
     }
 
     /*
      * Lastly, if this is not the primary (canSetTag) ModifyTable node, add it
      * to estate->es_auxmodifytables so that it will be run to completion by
      * ExecPostprocessPlan.  (It'd actually work fine to add the primary
      * ModifyTable node too, but there's no need.)  Note the use of lcons not
      * lappend: we need later-initialized ModifyTable nodes to be shut down
      * before earlier ones.  This ensures that we don't throw away RETURNING
      * rows that need to be seen by a later CTE subplan.
      */
     if (!mtstate->canSetTag)
         estate->es_auxmodifytables = lcons(mtstate,
                                            estate->es_auxmodifytables);
 
     return mtstate;
 }

◆ ExecInitUpdateProjection()

static void ExecInitUpdateProjection	(	ModifyTableState *	mtstate,
		ResultRelInfo *	resultRelInfo
	)

static

Definition at line 454 of file nodeModifyTable.c.

References Assert, EState::es_tupleTable, ExecAssignExprContext(), ExecBuildUpdateProjection(), list_nth(), ModifyTableState::mt_lastResultIndex, outerPlan, PlanState::plan, ModifyTableState::ps, PlanState::ps_ExprContext, RelationGetDescr, ModifyTableState::resultRelInfo, ResultRelInfo::ri_newTupleSlot, ResultRelInfo::ri_oldTupleSlot, ResultRelInfo::ri_projectNew, ResultRelInfo::ri_projectNewInfoValid, ResultRelInfo::ri_RelationDesc, PlanState::state, table_slot_create(), and ModifyTable::updateColnosLists.

Referenced by ExecCrossPartitionUpdate(), ExecModifyTable(), and ExecUpdate().

 {
     ModifyTable *node = (ModifyTable *) mtstate->ps.plan;
     Plan       *subplan = outerPlan(node);
     EState     *estate = mtstate->ps.state;
     TupleDesc   relDesc = RelationGetDescr(resultRelInfo->ri_RelationDesc);
     int         whichrel;
     List       *updateColnos;
 
     /*
      * Usually, mt_lastResultIndex matches the target rel.  If it happens not
      * to, we can get the index the hard way with an integer division.
      */
     whichrel = mtstate->mt_lastResultIndex;
     if (resultRelInfo != mtstate->resultRelInfo + whichrel)
     {
         whichrel = resultRelInfo - mtstate->resultRelInfo;
         Assert(whichrel >= 0 && whichrel < mtstate->mt_nrels);
     }
 
     updateColnos = (List *) list_nth(node->updateColnosLists, whichrel);
 
     /*
      * For UPDATE, we use the old tuple to fill up missing values in the tuple
      * produced by the subplan to get the new tuple.  We need two slots, both
      * matching the table's desired format.
      */
     resultRelInfo->ri_oldTupleSlot =
         table_slot_create(resultRelInfo->ri_RelationDesc,
                           &estate->es_tupleTable);
     resultRelInfo->ri_newTupleSlot =
         table_slot_create(resultRelInfo->ri_RelationDesc,
                           &estate->es_tupleTable);
 
     /* need an expression context to do the projection */
     if (mtstate->ps.ps_ExprContext == NULL)
         ExecAssignExprContext(estate, &mtstate->ps);
 
     resultRelInfo->ri_projectNew =
         ExecBuildUpdateProjection(subplan->targetlist,
                                   false,    /* subplan did the evaluation */
                                   updateColnos,
                                   relDesc,
                                   mtstate->ps.ps_ExprContext,
                                   resultRelInfo->ri_newTupleSlot,
                                   &mtstate->ps);
 
     resultRelInfo->ri_projectNewInfoValid = true;
 }

◆ ExecInsert()

static TupleTableSlot* ExecInsert	(	ModifyTableState *	mtstate,
		ResultRelInfo *	resultRelInfo,
		TupleTableSlot *	slot,
		TupleTableSlot *	planSlot,
		EState *	estate,
		bool	canSetTag
	)

static

Definition at line 594 of file nodeModifyTable.c.

Referenced by ExecCrossPartitionUpdate(), and ExecModifyTable().

 {
     Relation    resultRelationDesc;
     List       *recheckIndexes = NIL;
     TupleTableSlot *result = NULL;
     TransitionCaptureState *ar_insert_trig_tcs;
     ModifyTable *node = (ModifyTable *) mtstate->ps.plan;
     OnConflictAction onconflict = node->onConflictAction;
     PartitionTupleRouting *proute = mtstate->mt_partition_tuple_routing;
     MemoryContext oldContext;
 
     /*
      * If the input result relation is a partitioned table, find the leaf
      * partition to insert the tuple into.
      */
     if (proute)
     {
         ResultRelInfo *partRelInfo;
 
         slot = ExecPrepareTupleRouting(mtstate, estate, proute,
                                        resultRelInfo, slot,
                                        &partRelInfo);
         resultRelInfo = partRelInfo;
     }
 
     ExecMaterializeSlot(slot);
 
     resultRelationDesc = resultRelInfo->ri_RelationDesc;
 
     /*
      * Open the table's indexes, if we have not done so already, so that we
      * can add new index entries for the inserted tuple.
      */
     if (resultRelationDesc->rd_rel->relhasindex &&
         resultRelInfo->ri_IndexRelationDescs == NULL)
         ExecOpenIndices(resultRelInfo, onconflict != ONCONFLICT_NONE);
 
     /*
      * BEFORE ROW INSERT Triggers.
      *
      * Note: We fire BEFORE ROW TRIGGERS for every attempted insertion in an
      * INSERT ... ON CONFLICT statement.  We cannot check for constraint
      * violations before firing these triggers, because they can change the
      * values to insert.  Also, they can run arbitrary user-defined code with
      * side-effects that we can't cancel by just not inserting the tuple.
      */
     if (resultRelInfo->ri_TrigDesc &&
         resultRelInfo->ri_TrigDesc->trig_insert_before_row)
     {
         if (!ExecBRInsertTriggers(estate, resultRelInfo, slot))
             return NULL;        /* "do nothing" */
     }
 
     /* INSTEAD OF ROW INSERT Triggers */
     if (resultRelInfo->ri_TrigDesc &&
         resultRelInfo->ri_TrigDesc->trig_insert_instead_row)
     {
         if (!ExecIRInsertTriggers(estate, resultRelInfo, slot))
             return NULL;        /* "do nothing" */
     }
     else if (resultRelInfo->ri_FdwRoutine)
     {
         /*
          * GENERATED expressions might reference the tableoid column, so
          * (re-)initialize tts_tableOid before evaluating them.
          */
         slot->tts_tableOid = RelationGetRelid(resultRelInfo->ri_RelationDesc);
 
         /*
          * Compute stored generated columns
          */
         if (resultRelationDesc->rd_att->constr &&
             resultRelationDesc->rd_att->constr->has_generated_stored)
             ExecComputeStoredGenerated(resultRelInfo, estate, slot,
                                        CMD_INSERT);
 
         /*
          * If the FDW supports batching, and batching is requested, accumulate
          * rows and insert them in batches. Otherwise use the per-row inserts.
          */
         if (resultRelInfo->ri_BatchSize > 1)
         {
             /*
              * If a certain number of tuples have already been accumulated, or
              * a tuple has come for a different relation than that for the
              * accumulated tuples, perform the batch insert
              */
             if (resultRelInfo->ri_NumSlots == resultRelInfo->ri_BatchSize)
             {
                 ExecBatchInsert(mtstate, resultRelInfo,
                                 resultRelInfo->ri_Slots,
                                 resultRelInfo->ri_PlanSlots,
                                 resultRelInfo->ri_NumSlots,
                                 estate, canSetTag);
                 resultRelInfo->ri_NumSlots = 0;
             }
 
             oldContext = MemoryContextSwitchTo(estate->es_query_cxt);
 
             if (resultRelInfo->ri_Slots == NULL)
             {
                 resultRelInfo->ri_Slots = palloc(sizeof(TupleTableSlot *) *
                                                  resultRelInfo->ri_BatchSize);
                 resultRelInfo->ri_PlanSlots = palloc(sizeof(TupleTableSlot *) *
                                                      resultRelInfo->ri_BatchSize);
             }
 
             /*
              * Initialize the batch slots. We don't know how many slots will
              * be needed, so we initialize them as the batch grows, and we
              * keep them across batches. To mitigate an inefficiency in how
              * resource owner handles objects with many references (as with
              * many slots all referencing the same tuple descriptor) we copy
              * the appropriate tuple descriptor for each slot.
              */
             if (resultRelInfo->ri_NumSlots >= resultRelInfo->ri_NumSlotsInitialized)
             {
                 TupleDesc   tdesc = CreateTupleDescCopy(slot->tts_tupleDescriptor);
                 TupleDesc   plan_tdesc =
                     CreateTupleDescCopy(planSlot->tts_tupleDescriptor);
 
                 resultRelInfo->ri_Slots[resultRelInfo->ri_NumSlots] =
                     MakeSingleTupleTableSlot(tdesc, slot->tts_ops);
 
                 resultRelInfo->ri_PlanSlots[resultRelInfo->ri_NumSlots] =
                     MakeSingleTupleTableSlot(plan_tdesc, planSlot->tts_ops);
 
                 /* remember how many batch slots we initialized */
                 resultRelInfo->ri_NumSlotsInitialized++;
             }
 
             ExecCopySlot(resultRelInfo->ri_Slots[resultRelInfo->ri_NumSlots],
                          slot);
 
             ExecCopySlot(resultRelInfo->ri_PlanSlots[resultRelInfo->ri_NumSlots],
                          planSlot);
 
             resultRelInfo->ri_NumSlots++;
 
             MemoryContextSwitchTo(oldContext);
 
             return NULL;
         }
 
         /*
          * insert into foreign table: let the FDW do it
          */
         slot = resultRelInfo->ri_FdwRoutine->ExecForeignInsert(estate,
                                                                resultRelInfo,
                                                                slot,
                                                                planSlot);
 
         if (slot == NULL)       /* "do nothing" */
             return NULL;
 
         /*
          * AFTER ROW Triggers or RETURNING expressions might reference the
          * tableoid column, so (re-)initialize tts_tableOid before evaluating
          * them.  (This covers the case where the FDW replaced the slot.)
          */
         slot->tts_tableOid = RelationGetRelid(resultRelInfo->ri_RelationDesc);
     }
     else
     {
         WCOKind     wco_kind;
 
         /*
          * Constraints and GENERATED expressions might reference the tableoid
          * column, so (re-)initialize tts_tableOid before evaluating them.
          */
         slot->tts_tableOid = RelationGetRelid(resultRelationDesc);
 
         /*
          * Compute stored generated columns
          */
         if (resultRelationDesc->rd_att->constr &&
             resultRelationDesc->rd_att->constr->has_generated_stored)
             ExecComputeStoredGenerated(resultRelInfo, estate, slot,
                                        CMD_INSERT);
 
         /*
          * Check any RLS WITH CHECK policies.
          *
          * Normally we should check INSERT policies. But if the insert is the
          * result of a partition key update that moved the tuple to a new
          * partition, we should instead check UPDATE policies, because we are
          * executing policies defined on the target table, and not those
          * defined on the child partitions.
          */
         wco_kind = (mtstate->operation == CMD_UPDATE) ?
             WCO_RLS_UPDATE_CHECK : WCO_RLS_INSERT_CHECK;
 
         /*
          * ExecWithCheckOptions() will skip any WCOs which are not of the kind
          * we are looking for at this point.
          */
         if (resultRelInfo->ri_WithCheckOptions != NIL)
             ExecWithCheckOptions(wco_kind, resultRelInfo, slot, estate);
 
         /*
          * Check the constraints of the tuple.
          */
         if (resultRelationDesc->rd_att->constr)
             ExecConstraints(resultRelInfo, slot, estate);
 
         /*
          * Also check the tuple against the partition constraint, if there is
          * one; except that if we got here via tuple-routing, we don't need to
          * if there's no BR trigger defined on the partition.
          */
         if (resultRelationDesc->rd_rel->relispartition &&
             (resultRelInfo->ri_RootResultRelInfo == NULL ||
              (resultRelInfo->ri_TrigDesc &&
               resultRelInfo->ri_TrigDesc->trig_insert_before_row)))
             ExecPartitionCheck(resultRelInfo, slot, estate, true);
 
         if (onconflict != ONCONFLICT_NONE && resultRelInfo->ri_NumIndices > 0)
         {
             /* Perform a speculative insertion. */
             uint32      specToken;
             ItemPointerData conflictTid;
             bool        specConflict;
             List       *arbiterIndexes;
 
             arbiterIndexes = resultRelInfo->ri_onConflictArbiterIndexes;
 
             /*
              * Do a non-conclusive check for conflicts first.
              *
              * We're not holding any locks yet, so this doesn't guarantee that
              * the later insert won't conflict.  But it avoids leaving behind
              * a lot of canceled speculative insertions, if you run a lot of
              * INSERT ON CONFLICT statements that do conflict.
              *
              * We loop back here if we find a conflict below, either during
              * the pre-check, or when we re-check after inserting the tuple
              * speculatively.
              */
     vlock:
             specConflict = false;
             if (!ExecCheckIndexConstraints(resultRelInfo, slot, estate,
                                            &conflictTid, arbiterIndexes))
             {
                 /* committed conflict tuple found */
                 if (onconflict == ONCONFLICT_UPDATE)
                 {
                     /*
                      * In case of ON CONFLICT DO UPDATE, execute the UPDATE
                      * part.  Be prepared to retry if the UPDATE fails because
                      * of another concurrent UPDATE/DELETE to the conflict
                      * tuple.
                      */
                     TupleTableSlot *returning = NULL;
 
                     if (ExecOnConflictUpdate(mtstate, resultRelInfo,
                                              &conflictTid, planSlot, slot,
                                              estate, canSetTag, &returning))
                     {
                         InstrCountTuples2(&mtstate->ps, 1);
                         return returning;
                     }
                     else
                         goto vlock;
                 }
                 else
                 {
                     /*
                      * In case of ON CONFLICT DO NOTHING, do nothing. However,
                      * verify that the tuple is visible to the executor's MVCC
                      * snapshot at higher isolation levels.
                      *
                      * Using ExecGetReturningSlot() to store the tuple for the
                      * recheck isn't that pretty, but we can't trivially use
                      * the input slot, because it might not be of a compatible
                      * type. As there's no conflicting usage of
                      * ExecGetReturningSlot() in the DO NOTHING case...
                      */
                     Assert(onconflict == ONCONFLICT_NOTHING);
                     ExecCheckTIDVisible(estate, resultRelInfo, &conflictTid,
                                         ExecGetReturningSlot(estate, resultRelInfo));
                     InstrCountTuples2(&mtstate->ps, 1);
                     return NULL;
                 }
             }
 
             /*
              * Before we start insertion proper, acquire our "speculative
              * insertion lock".  Others can use that to wait for us to decide
              * if we're going to go ahead with the insertion, instead of
              * waiting for the whole transaction to complete.
              */
             specToken = SpeculativeInsertionLockAcquire(GetCurrentTransactionId());
 
             /* insert the tuple, with the speculative token */
             table_tuple_insert_speculative(resultRelationDesc, slot,
                                            estate->es_output_cid,
                                            0,
                                            NULL,
                                            specToken);
 
             /* insert index entries for tuple */
             recheckIndexes = ExecInsertIndexTuples(resultRelInfo,
                                                    slot, estate, false, true,
                                                    &specConflict,
                                                    arbiterIndexes);
 
             /* adjust the tuple's state accordingly */
             table_tuple_complete_speculative(resultRelationDesc, slot,
                                              specToken, !specConflict);
 
             /*
              * Wake up anyone waiting for our decision.  They will re-check
              * the tuple, see that it's no longer speculative, and wait on our
              * XID as if this was a regularly inserted tuple all along.  Or if
              * we killed the tuple, they will see it's dead, and proceed as if
              * the tuple never existed.
              */
             SpeculativeInsertionLockRelease(GetCurrentTransactionId());
 
             /*
              * If there was a conflict, start from the beginning.  We'll do
              * the pre-check again, which will now find the conflicting tuple
              * (unless it aborts before we get there).
              */
             if (specConflict)
             {
                 list_free(recheckIndexes);
                 goto vlock;
             }
 
             /* Since there was no insertion conflict, we're done */
         }
         else
         {
             /* insert the tuple normally */
             table_tuple_insert(resultRelationDesc, slot,
                                estate->es_output_cid,
                                0, NULL);
 
             /* insert index entries for tuple */
             if (resultRelInfo->ri_NumIndices > 0)
                 recheckIndexes = ExecInsertIndexTuples(resultRelInfo,
                                                        slot, estate, false,
                                                        false, NULL, NIL);
         }
     }
 
     if (canSetTag)
         (estate->es_processed)++;
 
     /*
      * If this insert is the result of a partition key update that moved the
      * tuple to a new partition, put this row into the transition NEW TABLE,
      * if there is one. We need to do this separately for DELETE and INSERT
      * because they happen on different tables.
      */
     ar_insert_trig_tcs = mtstate->mt_transition_capture;
     if (mtstate->operation == CMD_UPDATE && mtstate->mt_transition_capture
         && mtstate->mt_transition_capture->tcs_update_new_table)
     {
         ExecARUpdateTriggers(estate, resultRelInfo, NULL,
                              NULL,
                              slot,
                              NULL,
                              mtstate->mt_transition_capture);
 
         /*
          * We've already captured the NEW TABLE row, so make sure any AR
          * INSERT trigger fired below doesn't capture it again.
          */
         ar_insert_trig_tcs = NULL;
     }
 
     /* AFTER ROW INSERT Triggers */
     ExecARInsertTriggers(estate, resultRelInfo, slot, recheckIndexes,
                          ar_insert_trig_tcs);
 
     list_free(recheckIndexes);
 
     /*
      * Check any WITH CHECK OPTION constraints from parent views.  We are
      * required to do this after testing all constraints and uniqueness
      * violations per the SQL spec, so we do it after actually inserting the
      * record into the heap and all indexes.
      *
      * ExecWithCheckOptions will elog(ERROR) if a violation is found, so the
      * tuple will never be seen, if it violates the WITH CHECK OPTION.
      *
      * ExecWithCheckOptions() will skip any WCOs which are not of the kind we
      * are looking for at this point.
      */
     if (resultRelInfo->ri_WithCheckOptions != NIL)
         ExecWithCheckOptions(WCO_VIEW_CHECK, resultRelInfo, slot, estate);
 
     /* Process RETURNING if present */
     if (resultRelInfo->ri_projectReturning)
         result = ExecProcessReturning(resultRelInfo, slot, planSlot);
 
     return result;
 }

◆ ExecLookupResultRelByOid()

ResultRelInfo* ExecLookupResultRelByOid	(	ModifyTableState *	node,
		Oid	resultoid,
		bool	missing_ok,
		bool	update_cache
	)

Definition at line 2659 of file nodeModifyTable.c.

References elog, ERROR, HASH_FIND, hash_search(), ModifyTableState::mt_lastResultIndex, ModifyTableState::mt_lastResultOid, ModifyTableState::mt_nrels, ModifyTableState::mt_resultOidHash, RelationGetRelid, MTTargetRelLookup::relationIndex, ModifyTableState::resultRelInfo, and ResultRelInfo::ri_RelationDesc.

Referenced by exec_rt_fetch(), ExecFindPartition(), and ExecModifyTable().

 {
     if (node->mt_resultOidHash)
     {
         /* Use the pre-built hash table to locate the rel */
         MTTargetRelLookup *mtlookup;
 
         mtlookup = (MTTargetRelLookup *)
             hash_search(node->mt_resultOidHash, &resultoid, HASH_FIND, NULL);
         if (mtlookup)
         {
             if (update_cache)
             {
                 node->mt_lastResultOid = resultoid;
                 node->mt_lastResultIndex = mtlookup->relationIndex;
             }
             return node->resultRelInfo + mtlookup->relationIndex;
         }
     }
     else
     {
         /* With few target rels, just search the ResultRelInfo array */
         for (int ndx = 0; ndx < node->mt_nrels; ndx++)
         {
             ResultRelInfo *rInfo = node->resultRelInfo + ndx;
 
             if (RelationGetRelid(rInfo->ri_RelationDesc) == resultoid)
             {
                 if (update_cache)
                 {
                     node->mt_lastResultOid = resultoid;
                     node->mt_lastResultIndex = ndx;
                 }
                 return rInfo;
             }
         }
     }
 
     if (!missing_ok)
         elog(ERROR, "incorrect result relation OID %u", resultoid);
     return NULL;
 }

◆ ExecModifyTable()

static TupleTableSlot* ExecModifyTable ( PlanState * pstate )

static

Definition at line 2349 of file nodeModifyTable.c.

Referenced by ExecInitModifyTable().

 {
     ModifyTableState *node = castNode(ModifyTableState, pstate);
     EState     *estate = node->ps.state;
     CmdType     operation = node->operation;
     ResultRelInfo *resultRelInfo;
     PlanState  *subplanstate;
     TupleTableSlot *slot;
     TupleTableSlot *planSlot;
     TupleTableSlot *oldSlot;
     ItemPointer tupleid;
     ItemPointerData tuple_ctid;
     HeapTupleData oldtupdata;
     HeapTuple   oldtuple;
     PartitionTupleRouting *proute = node->mt_partition_tuple_routing;
     List       *relinfos = NIL;
     ListCell   *lc;
 
     CHECK_FOR_INTERRUPTS();
 
     /*
      * This should NOT get called during EvalPlanQual; we should have passed a
      * subplan tree to EvalPlanQual, instead.  Use a runtime test not just
      * Assert because this condition is easy to miss in testing.  (Note:
      * although ModifyTable should not get executed within an EvalPlanQual
      * operation, we do have to allow it to be initialized and shut down in
      * case it is within a CTE subplan.  Hence this test must be here, not in
      * ExecInitModifyTable.)
      */
     if (estate->es_epq_active != NULL)
         elog(ERROR, "ModifyTable should not be called during EvalPlanQual");
 
     /*
      * If we've already completed processing, don't try to do more.  We need
      * this test because ExecPostprocessPlan might call us an extra time, and
      * our subplan's nodes aren't necessarily robust against being called
      * extra times.
      */
     if (node->mt_done)
         return NULL;
 
     /*
      * On first call, fire BEFORE STATEMENT triggers before proceeding.
      */
     if (node->fireBSTriggers)
     {
         fireBSTriggers(node);
         node->fireBSTriggers = false;
     }
 
     /* Preload local variables */
     resultRelInfo = node->resultRelInfo + node->mt_lastResultIndex;
     subplanstate = outerPlanState(node);
 
     /*
      * Fetch rows from subplan, and execute the required table modification
      * for each row.
      */
     for (;;)
     {
         /*
          * Reset the per-output-tuple exprcontext.  This is needed because
          * triggers expect to use that context as workspace.  It's a bit ugly
          * to do this below the top level of the plan, however.  We might need
          * to rethink this later.
          */
         ResetPerTupleExprContext(estate);
 
         /*
          * Reset per-tuple memory context used for processing on conflict and
          * returning clauses, to free any expression evaluation storage
          * allocated in the previous cycle.
          */
         if (pstate->ps_ExprContext)
             ResetExprContext(pstate->ps_ExprContext);
 
         planSlot = ExecProcNode(subplanstate);
 
         /* No more tuples to process? */
         if (TupIsNull(planSlot))
             break;
 
         /*
          * When there are multiple result relations, each tuple contains a
          * junk column that gives the OID of the rel from which it came.
          * Extract it and select the correct result relation.
          */
         if (AttributeNumberIsValid(node->mt_resultOidAttno))
         {
             Datum       datum;
             bool        isNull;
             Oid         resultoid;
 
             datum = ExecGetJunkAttribute(planSlot, node->mt_resultOidAttno,
                                          &isNull);
             if (isNull)
                 elog(ERROR, "tableoid is NULL");
             resultoid = DatumGetObjectId(datum);
 
             /* If it's not the same as last time, we need to locate the rel */
             if (resultoid != node->mt_lastResultOid)
                 resultRelInfo = ExecLookupResultRelByOid(node, resultoid,
                                                          false, true);
         }
 
         /*
          * If resultRelInfo->ri_usesFdwDirectModify is true, all we need to do
          * here is compute the RETURNING expressions.
          */
         if (resultRelInfo->ri_usesFdwDirectModify)
         {
             Assert(resultRelInfo->ri_projectReturning);
 
             /*
              * A scan slot containing the data that was actually inserted,
              * updated or deleted has already been made available to
              * ExecProcessReturning by IterateDirectModify, so no need to
              * provide it here.
              */
             slot = ExecProcessReturning(resultRelInfo, NULL, planSlot);
 
             return slot;
         }
 
         EvalPlanQualSetSlot(&node->mt_epqstate, planSlot);
         slot = planSlot;
 
         tupleid = NULL;
         oldtuple = NULL;
 
         /*
          * For UPDATE/DELETE, fetch the row identity info for the tuple to be
          * updated/deleted.  For a heap relation, that's a TID; otherwise we
          * may have a wholerow junk attr that carries the old tuple in toto.
          * Keep this in step with the part of ExecInitModifyTable that sets up
          * ri_RowIdAttNo.
          */
         if (operation == CMD_UPDATE || operation == CMD_DELETE)
         {
             char        relkind;
             Datum       datum;
             bool        isNull;
 
             relkind = resultRelInfo->ri_RelationDesc->rd_rel->relkind;
             if (relkind == RELKIND_RELATION ||
                 relkind == RELKIND_MATVIEW ||
                 relkind == RELKIND_PARTITIONED_TABLE)
             {
                 /* ri_RowIdAttNo refers to a ctid attribute */
                 Assert(AttributeNumberIsValid(resultRelInfo->ri_RowIdAttNo));
                 datum = ExecGetJunkAttribute(slot,
                                              resultRelInfo->ri_RowIdAttNo,
                                              &isNull);
                 /* shouldn't ever get a null result... */
                 if (isNull)
                     elog(ERROR, "ctid is NULL");
 
                 tupleid = (ItemPointer) DatumGetPointer(datum);
                 tuple_ctid = *tupleid;  /* be sure we don't free ctid!! */
                 tupleid = &tuple_ctid;
             }
 
             /*
              * Use the wholerow attribute, when available, to reconstruct the
              * old relation tuple.  The old tuple serves one or both of two
              * purposes: 1) it serves as the OLD tuple for row triggers, 2) it
              * provides values for any unchanged columns for the NEW tuple of
              * an UPDATE, because the subplan does not produce all the columns
              * of the target table.
              *
              * Note that the wholerow attribute does not carry system columns,
              * so foreign table triggers miss seeing those, except that we
              * know enough here to set t_tableOid.  Quite separately from
              * this, the FDW may fetch its own junk attrs to identify the row.
              *
              * Other relevant relkinds, currently limited to views, always
              * have a wholerow attribute.
              */
             else if (AttributeNumberIsValid(resultRelInfo->ri_RowIdAttNo))
             {
                 datum = ExecGetJunkAttribute(slot,
                                              resultRelInfo->ri_RowIdAttNo,
                                              &isNull);
                 /* shouldn't ever get a null result... */
                 if (isNull)
                     elog(ERROR, "wholerow is NULL");
 
                 oldtupdata.t_data = DatumGetHeapTupleHeader(datum);
                 oldtupdata.t_len =
                     HeapTupleHeaderGetDatumLength(oldtupdata.t_data);
                 ItemPointerSetInvalid(&(oldtupdata.t_self));
                 /* Historically, view triggers see invalid t_tableOid. */
                 oldtupdata.t_tableOid =
                     (relkind == RELKIND_VIEW) ? InvalidOid :
                     RelationGetRelid(resultRelInfo->ri_RelationDesc);
 
                 oldtuple = &oldtupdata;
             }
             else
             {
                 /* Only foreign tables are allowed to omit a row-ID attr */
                 Assert(relkind == RELKIND_FOREIGN_TABLE);
             }
         }
 
         switch (operation)
         {
             case CMD_INSERT:
                 /* Initialize projection info if first time for this table */
                 if (unlikely(!resultRelInfo->ri_projectNewInfoValid))
                     ExecInitInsertProjection(node, resultRelInfo);
                 slot = ExecGetInsertNewTuple(resultRelInfo, planSlot);
                 slot = ExecInsert(node, resultRelInfo, slot, planSlot,
                                   estate, node->canSetTag);
                 break;
             case CMD_UPDATE:
                 /* Initialize projection info if first time for this table */
                 if (unlikely(!resultRelInfo->ri_projectNewInfoValid))
                     ExecInitUpdateProjection(node, resultRelInfo);
 
                 /*
                  * Make the new tuple by combining plan's output tuple with
                  * the old tuple being updated.
                  */
                 oldSlot = resultRelInfo->ri_oldTupleSlot;
                 if (oldtuple != NULL)
                 {
                     /* Use the wholerow junk attr as the old tuple. */
                     ExecForceStoreHeapTuple(oldtuple, oldSlot, false);
                 }
                 else
                 {
                     /* Fetch the most recent version of old tuple. */
                     Relation    relation = resultRelInfo->ri_RelationDesc;
 
                     Assert(tupleid != NULL);
                     if (!table_tuple_fetch_row_version(relation, tupleid,
                                                        SnapshotAny,
                                                        oldSlot))
                         elog(ERROR, "failed to fetch tuple being updated");
                 }
                 slot = ExecGetUpdateNewTuple(resultRelInfo, planSlot,
                                              oldSlot);
 
                 /* Now apply the update. */
                 slot = ExecUpdate(node, resultRelInfo, tupleid, oldtuple, slot,
                                   planSlot, &node->mt_epqstate, estate,
                                   node->canSetTag);
                 break;
             case CMD_DELETE:
                 slot = ExecDelete(node, resultRelInfo, tupleid, oldtuple,
                                   planSlot, &node->mt_epqstate, estate,
                                   true, /* processReturning */
                                   node->canSetTag,
                                   false,    /* changingPart */
                                   NULL, NULL);
                 break;
             default:
                 elog(ERROR, "unknown operation");
                 break;
         }
 
         /*
          * If we got a RETURNING result, return it to caller.  We'll continue
          * the work on next call.
          */
         if (slot)
             return slot;
     }
 
     /*
      * Insert remaining tuples for batch insert.
      */
     if (proute)
         relinfos = estate->es_tuple_routing_result_relations;
     else
         relinfos = estate->es_opened_result_relations;
 
     foreach(lc, relinfos)
     {
         resultRelInfo = lfirst(lc);
         if (resultRelInfo->ri_NumSlots > 0)
             ExecBatchInsert(node, resultRelInfo,
                             resultRelInfo->ri_Slots,
                             resultRelInfo->ri_PlanSlots,
                             resultRelInfo->ri_NumSlots,
                             estate, node->canSetTag);
     }
 
     /*
      * We're done, but fire AFTER STATEMENT triggers before exiting.
      */
     fireASTriggers(node);
 
     node->mt_done = true;
 
     return NULL;
 }

◆ ExecOnConflictUpdate()

static bool ExecOnConflictUpdate	(	ModifyTableState *	mtstate,
		ResultRelInfo *	resultRelInfo,
		ItemPointer	conflictTid,
		TupleTableSlot *	planSlot,
		TupleTableSlot *	excludedSlot,
		EState *	estate,
		bool	canSetTag,
		TupleTableSlot **	returning
	)

static

Definition at line 1984 of file nodeModifyTable.c.

References Assert, TM_FailureData::ctid, DatumGetTransactionId, ExprContext::ecxt_innertuple, ExprContext::ecxt_outertuple, ExprContext::ecxt_scantuple, elog, ereport, errcode(), errhint(), errmsg(), ERROR, EState::es_output_cid, EState::es_snapshot, ExecCheckTupleVisible(), ExecClearTuple(), ExecProject(), ExecQual(), ExecUpdate(), ExecUpdateLockMode(), ExecWithCheckOptions(), InstrCountFiltered1, IsolationUsesXactSnapshot, ItemPointerIndicatesMovedPartitions, LockWaitBlock, MinTransactionIdAttributeNumber, ModifyTableState::mt_epqstate, NIL, OnConflictSetState::oc_Existing, OnConflictSetState::oc_ProjInfo, OnConflictSetState::oc_ProjSlot, OnConflictSetState::oc_WhereClause, ModifyTableState::ps, PlanState::ps_ExprContext, ResultRelInfo::ri_onConflict, ResultRelInfo::ri_RelationDesc, ResultRelInfo::ri_WithCheckOptions, slot_getsysattr(), PlanState::state, table_tuple_lock(), test(), TM_Deleted, TM_Invisible, TM_Ok, TM_SelfModified, TM_Updated, TransactionIdIsCurrentTransactionId(), and WCO_RLS_CONFLICT_CHECK.

Referenced by ExecInsert().

 {
     ExprContext *econtext = mtstate->ps.ps_ExprContext;
     Relation    relation = resultRelInfo->ri_RelationDesc;
     ExprState  *onConflictSetWhere = resultRelInfo->ri_onConflict->oc_WhereClause;
     TupleTableSlot *existing = resultRelInfo->ri_onConflict->oc_Existing;
     TM_FailureData tmfd;
     LockTupleMode lockmode;
     TM_Result   test;
     Datum       xminDatum;
     TransactionId xmin;
     bool        isnull;
 
     /* Determine lock mode to use */
     lockmode = ExecUpdateLockMode(estate, resultRelInfo);
 
     /*
      * Lock tuple for update.  Don't follow updates when tuple cannot be
      * locked without doing so.  A row locking conflict here means our
      * previous conclusion that the tuple is conclusively committed is not
      * true anymore.
      */
     test = table_tuple_lock(relation, conflictTid,
                             estate->es_snapshot,
                             existing, estate->es_output_cid,
                             lockmode, LockWaitBlock, 0,
                             &tmfd);
     switch (test)
     {
         case TM_Ok:
             /* success! */
             break;
 
         case TM_Invisible:
 
             /*
              * This can occur when a just inserted tuple is updated again in
              * the same command. E.g. because multiple rows with the same
              * conflicting key values are inserted.
              *
              * This is somewhat similar to the ExecUpdate() TM_SelfModified
              * case.  We do not want to proceed because it would lead to the
              * same row being updated a second time in some unspecified order,
              * and in contrast to plain UPDATEs there's no historical behavior
              * to break.
              *
              * It is the user's responsibility to prevent this situation from
              * occurring.  These problems are why SQL-2003 similarly specifies
              * that for SQL MERGE, an exception must be raised in the event of
              * an attempt to update the same row twice.
              */
             xminDatum = slot_getsysattr(existing,
                                         MinTransactionIdAttributeNumber,
                                         &isnull);
             Assert(!isnull);
             xmin = DatumGetTransactionId(xminDatum);
 
             if (TransactionIdIsCurrentTransactionId(xmin))
                 ereport(ERROR,
                         (errcode(ERRCODE_CARDINALITY_VIOLATION),
                          errmsg("ON CONFLICT DO UPDATE command cannot affect row a second time"),
                          errhint("Ensure that no rows proposed for insertion within the same command have duplicate constrained values.")));
 
             /* This shouldn't happen */
             elog(ERROR, "attempted to lock invisible tuple");
             break;
 
         case TM_SelfModified:
 
             /*
              * This state should never be reached. As a dirty snapshot is used
              * to find conflicting tuples, speculative insertion wouldn't have
              * seen this row to conflict with.
              */
             elog(ERROR, "unexpected self-updated tuple");
             break;
 
         case TM_Updated:
             if (IsolationUsesXactSnapshot())
                 ereport(ERROR,
                         (errcode(ERRCODE_T_R_SERIALIZATION_FAILURE),
                          errmsg("could not serialize access due to concurrent update")));
 
             /*
              * As long as we don't support an UPDATE of INSERT ON CONFLICT for
              * a partitioned table we shouldn't reach to a case where tuple to
              * be lock is moved to another partition due to concurrent update
              * of the partition key.
              */
             Assert(!ItemPointerIndicatesMovedPartitions(&tmfd.ctid));
 
             /*
              * Tell caller to try again from the very start.
              *
              * It does not make sense to use the usual EvalPlanQual() style
              * loop here, as the new version of the row might not conflict
              * anymore, or the conflicting tuple has actually been deleted.
              */
             ExecClearTuple(existing);
             return false;
 
         case TM_Deleted:
             if (IsolationUsesXactSnapshot())
                 ereport(ERROR,
                         (errcode(ERRCODE_T_R_SERIALIZATION_FAILURE),
                          errmsg("could not serialize access due to concurrent delete")));
 
             /* see TM_Updated case */
             Assert(!ItemPointerIndicatesMovedPartitions(&tmfd.ctid));
             ExecClearTuple(existing);
             return false;
 
         default:
             elog(ERROR, "unrecognized table_tuple_lock status: %u", test);
     }
 
     /* Success, the tuple is locked. */
 
     /*
      * Verify that the tuple is visible to our MVCC snapshot if the current
      * isolation level mandates that.
      *
      * It's not sufficient to rely on the check within ExecUpdate() as e.g.
      * CONFLICT ... WHERE clause may prevent us from reaching that.
      *
      * This means we only ever continue when a new command in the current
      * transaction could see the row, even though in READ COMMITTED mode the
      * tuple will not be visible according to the current statement's
      * snapshot.  This is in line with the way UPDATE deals with newer tuple
      * versions.
      */
     ExecCheckTupleVisible(estate, relation, existing);
 
     /*
      * Make tuple and any needed join variables available to ExecQual and
      * ExecProject.  The EXCLUDED tuple is installed in ecxt_innertuple, while
      * the target's existing tuple is installed in the scantuple.  EXCLUDED
      * has been made to reference INNER_VAR in setrefs.c, but there is no
      * other redirection.
      */
     econtext->ecxt_scantuple = existing;
     econtext->ecxt_innertuple = excludedSlot;
     econtext->ecxt_outertuple = NULL;
 
     if (!ExecQual(onConflictSetWhere, econtext))
     {
         ExecClearTuple(existing);   /* see return below */
         InstrCountFiltered1(&mtstate->ps, 1);
         return true;            /* done with the tuple */
     }
 
     if (resultRelInfo->ri_WithCheckOptions != NIL)
     {
         /*
          * Check target's existing tuple against UPDATE-applicable USING
          * security barrier quals (if any), enforced here as RLS checks/WCOs.
          *
          * The rewriter creates UPDATE RLS checks/WCOs for UPDATE security
          * quals, and stores them as WCOs of "kind" WCO_RLS_CONFLICT_CHECK,
          * but that's almost the extent of its special handling for ON
          * CONFLICT DO UPDATE.
          *
          * The rewriter will also have associated UPDATE applicable straight
          * RLS checks/WCOs for the benefit of the ExecUpdate() call that
          * follows.  INSERTs and UPDATEs naturally have mutually exclusive WCO
          * kinds, so there is no danger of spurious over-enforcement in the
          * INSERT or UPDATE path.
          */
         ExecWithCheckOptions(WCO_RLS_CONFLICT_CHECK, resultRelInfo,
                              existing,
                              mtstate->ps.state);
     }
 
     /* Project the new tuple version */
     ExecProject(resultRelInfo->ri_onConflict->oc_ProjInfo);
 
     /*
      * Note that it is possible that the target tuple has been modified in
      * this session, after the above table_tuple_lock. We choose to not error
      * out in that case, in line with ExecUpdate's treatment of similar cases.
      * This can happen if an UPDATE is triggered from within ExecQual(),
      * ExecWithCheckOptions() or ExecProject() above, e.g. by selecting from a
      * wCTE in the ON CONFLICT's SET.
      */
 
     /* Execute UPDATE with projection */
     *returning = ExecUpdate(mtstate, resultRelInfo, conflictTid, NULL,
                             resultRelInfo->ri_onConflict->oc_ProjSlot,
                             planSlot,
                             &mtstate->mt_epqstate, mtstate->ps.state,
                             canSetTag);
 
     /*
      * Clear out existing tuple, as there might not be another conflict among
      * the next input rows. Don't want to hold resources till the end of the
      * query.
      */
     ExecClearTuple(existing);
     return true;
 }

◆ ExecPrepareTupleRouting()

static TupleTableSlot * ExecPrepareTupleRouting	(	ModifyTableState *	mtstate,
		EState *	estate,
		PartitionTupleRouting *	proute,
		ResultRelInfo *	targetRelInfo,
		TupleTableSlot *	slot,
		ResultRelInfo **	partRelInfo
	)

static

Definition at line 2289 of file nodeModifyTable.c.

References TupleConversionMap::attrMap, ExecFindPartition(), execute_attr_map_slot(), ModifyTableState::mt_transition_capture, ResultRelInfo::ri_PartitionTupleSlot, ResultRelInfo::ri_RootToPartitionMap, ResultRelInfo::ri_TrigDesc, TransitionCaptureState::tcs_original_insert_tuple, and TriggerDesc::trig_insert_before_row.

Referenced by ExecInsert().

 {
     ResultRelInfo *partrel;
     TupleConversionMap *map;
 
     /*
      * Lookup the target partition's ResultRelInfo.  If ExecFindPartition does
      * not find a valid partition for the tuple in 'slot' then an error is
      * raised.  An error may also be raised if the found partition is not a
      * valid target for INSERTs.  This is required since a partitioned table
      * UPDATE to another partition becomes a DELETE+INSERT.
      */
     partrel = ExecFindPartition(mtstate, targetRelInfo, proute, slot, estate);
 
     /*
      * If we're capturing transition tuples, we might need to convert from the
      * partition rowtype to root partitioned table's rowtype.  But if there
      * are no BEFORE triggers on the partition that could change the tuple, we
      * can just remember the original unconverted tuple to avoid a needless
      * round trip conversion.
      */
     if (mtstate->mt_transition_capture != NULL)
     {
         bool        has_before_insert_row_trig;
 
         has_before_insert_row_trig = (partrel->ri_TrigDesc &&
                                       partrel->ri_TrigDesc->trig_insert_before_row);
 
         mtstate->mt_transition_capture->tcs_original_insert_tuple =
             !has_before_insert_row_trig ? slot : NULL;
     }
 
     /*
      * Convert the tuple, if necessary.
      */
     map = partrel->ri_RootToPartitionMap;
     if (map != NULL)
     {
         TupleTableSlot *new_slot = partrel->ri_PartitionTupleSlot;
 
         slot = execute_attr_map_slot(map->attrMap, slot, new_slot);
     }
 
     *partRelInfo = partrel;
     return slot;
 }

◆ ExecProcessReturning()

static TupleTableSlot* ExecProcessReturning	(	ResultRelInfo *	resultRelInfo,
		TupleTableSlot *	tupleSlot,
		TupleTableSlot *	planSlot
	)

static

Definition at line 170 of file nodeModifyTable.c.

References ExprContext::ecxt_outertuple, ExprContext::ecxt_scantuple, ExecProject(), ProjectionInfo::pi_exprContext, RelationGetRelid, ResultRelInfo::ri_projectReturning, ResultRelInfo::ri_RelationDesc, and TupleTableSlot::tts_tableOid.

Referenced by ExecDelete(), ExecInsert(), ExecModifyTable(), and ExecUpdate().

 {
     ProjectionInfo *projectReturning = resultRelInfo->ri_projectReturning;
     ExprContext *econtext = projectReturning->pi_exprContext;
 
     /* Make tuple and any needed join variables available to ExecProject */
     if (tupleSlot)
         econtext->ecxt_scantuple = tupleSlot;
     econtext->ecxt_outertuple = planSlot;
 
     /*
      * RETURNING expressions might reference the tableoid column, so
      * reinitialize tts_tableOid before evaluating them.
      */
     econtext->ecxt_scantuple->tts_tableOid =
         RelationGetRelid(resultRelInfo->ri_RelationDesc);
 
     /* Compute the RETURNING expressions */
     return ExecProject(projectReturning);
 }

◆ ExecReScanModifyTable()

void ExecReScanModifyTable ( ModifyTableState * node )

Definition at line 3234 of file nodeModifyTable.c.

References elog, and ERROR.

Referenced by ExecReScan().

 {
     /*
      * Currently, we don't need to support rescan on ModifyTable nodes. The
      * semantics of that would be a bit debatable anyway.
      */
     elog(ERROR, "ExecReScanModifyTable is not implemented");
 }

◆ ExecSetupTransitionCaptureState()

static void ExecSetupTransitionCaptureState	(	ModifyTableState *	mtstate,
		EState *	estate
	)

static

Definition at line 2260 of file nodeModifyTable.c.

References CMD_INSERT, CMD_UPDATE, MakeTransitionCaptureState(), ModifyTableState::mt_oc_transition_capture, ModifyTableState::mt_transition_capture, ONCONFLICT_UPDATE, ModifyTable::onConflictAction, ModifyTable::operation, ModifyTableState::operation, PlanState::plan, ModifyTableState::ps, RelationGetRelid, and ModifyTableState::rootResultRelInfo.

Referenced by ExecInitModifyTable().

 {
     ModifyTable *plan = (ModifyTable *) mtstate->ps.plan;
     ResultRelInfo *targetRelInfo = mtstate->rootResultRelInfo;
 
     /* Check for transition tables on the directly targeted relation. */
     mtstate->mt_transition_capture =
         MakeTransitionCaptureState(targetRelInfo->ri_TrigDesc,
                                    RelationGetRelid(targetRelInfo->ri_RelationDesc),
                                    mtstate->operation);
     if (plan->operation == CMD_INSERT &&
         plan->onConflictAction == ONCONFLICT_UPDATE)
         mtstate->mt_oc_transition_capture =
             MakeTransitionCaptureState(targetRelInfo->ri_TrigDesc,
                                        RelationGetRelid(targetRelInfo->ri_RelationDesc),
                                        CMD_UPDATE);
 }

◆ ExecUpdate()

static TupleTableSlot* ExecUpdate	(	ModifyTableState *	mtstate,
		ResultRelInfo *	resultRelInfo,
		ItemPointer	tupleid,
		HeapTuple	oldtuple,
		TupleTableSlot *	slot,
		TupleTableSlot *	planSlot,
		EPQState *	epqstate,
		EState *	estate,
		bool	canSetTag
	)

static

Definition at line 1601 of file nodeModifyTable.c.

Referenced by ExecModifyTable(), and ExecOnConflictUpdate().

 {
     Relation    resultRelationDesc = resultRelInfo->ri_RelationDesc;
     TM_Result   result;
     TM_FailureData tmfd;
     List       *recheckIndexes = NIL;
 
     /*
      * abort the operation if not running transactions
      */
     if (IsBootstrapProcessingMode())
         elog(ERROR, "cannot UPDATE during bootstrap");
 
     ExecMaterializeSlot(slot);
 
     /*
      * Open the table's indexes, if we have not done so already, so that we
      * can add new index entries for the updated tuple.
      */
     if (resultRelationDesc->rd_rel->relhasindex &&
         resultRelInfo->ri_IndexRelationDescs == NULL)
         ExecOpenIndices(resultRelInfo, false);
 
     /* BEFORE ROW UPDATE Triggers */
     if (resultRelInfo->ri_TrigDesc &&
         resultRelInfo->ri_TrigDesc->trig_update_before_row)
     {
         if (!ExecBRUpdateTriggers(estate, epqstate, resultRelInfo,
                                   tupleid, oldtuple, slot))
             return NULL;        /* "do nothing" */
     }
 
     /* INSTEAD OF ROW UPDATE Triggers */
     if (resultRelInfo->ri_TrigDesc &&
         resultRelInfo->ri_TrigDesc->trig_update_instead_row)
     {
         if (!ExecIRUpdateTriggers(estate, resultRelInfo,
                                   oldtuple, slot))
             return NULL;        /* "do nothing" */
     }
     else if (resultRelInfo->ri_FdwRoutine)
     {
         /*
          * GENERATED expressions might reference the tableoid column, so
          * (re-)initialize tts_tableOid before evaluating them.
          */
         slot->tts_tableOid = RelationGetRelid(resultRelInfo->ri_RelationDesc);
 
         /*
          * Compute stored generated columns
          */
         if (resultRelationDesc->rd_att->constr &&
             resultRelationDesc->rd_att->constr->has_generated_stored)
             ExecComputeStoredGenerated(resultRelInfo, estate, slot,
                                        CMD_UPDATE);
 
         /*
          * update in foreign table: let the FDW do it
          */
         slot = resultRelInfo->ri_FdwRoutine->ExecForeignUpdate(estate,
                                                                resultRelInfo,
                                                                slot,
                                                                planSlot);
 
         if (slot == NULL)       /* "do nothing" */
             return NULL;
 
         /*
          * AFTER ROW Triggers or RETURNING expressions might reference the
          * tableoid column, so (re-)initialize tts_tableOid before evaluating
          * them.  (This covers the case where the FDW replaced the slot.)
          */
         slot->tts_tableOid = RelationGetRelid(resultRelationDesc);
     }
     else
     {
         LockTupleMode lockmode;
         bool        partition_constraint_failed;
         bool        update_indexes;
 
         /*
          * Constraints and GENERATED expressions might reference the tableoid
          * column, so (re-)initialize tts_tableOid before evaluating them.
          */
         slot->tts_tableOid = RelationGetRelid(resultRelationDesc);
 
         /*
          * Compute stored generated columns
          */
         if (resultRelationDesc->rd_att->constr &&
             resultRelationDesc->rd_att->constr->has_generated_stored)
             ExecComputeStoredGenerated(resultRelInfo, estate, slot,
                                        CMD_UPDATE);
 
         /*
          * Check any RLS UPDATE WITH CHECK policies
          *
          * If we generate a new candidate tuple after EvalPlanQual testing, we
          * must loop back here and recheck any RLS policies and constraints.
          * (We don't need to redo triggers, however.  If there are any BEFORE
          * triggers then trigger.c will have done table_tuple_lock to lock the
          * correct tuple, so there's no need to do them again.)
          */
 lreplace:;
 
         /* ensure slot is independent, consider e.g. EPQ */
         ExecMaterializeSlot(slot);
 
         /*
          * If partition constraint fails, this row might get moved to another
          * partition, in which case we should check the RLS CHECK policy just
          * before inserting into the new partition, rather than doing it here.
          * This is because a trigger on that partition might again change the
          * row.  So skip the WCO checks if the partition constraint fails.
          */
         partition_constraint_failed =
             resultRelationDesc->rd_rel->relispartition &&
             !ExecPartitionCheck(resultRelInfo, slot, estate, false);
 
         if (!partition_constraint_failed &&
             resultRelInfo->ri_WithCheckOptions != NIL)
         {
             /*
              * ExecWithCheckOptions() will skip any WCOs which are not of the
              * kind we are looking for at this point.
              */
             ExecWithCheckOptions(WCO_RLS_UPDATE_CHECK,
                                  resultRelInfo, slot, estate);
         }
 
         /*
          * If a partition check failed, try to move the row into the right
          * partition.
          */
         if (partition_constraint_failed)
         {
             TupleTableSlot *inserted_tuple,
                        *retry_slot;
             bool        retry;
 
             /*
              * ExecCrossPartitionUpdate will first DELETE the row from the
              * partition it's currently in and then insert it back into the
              * root table, which will re-route it to the correct partition.
              * The first part may have to be repeated if it is detected that
              * the tuple we're trying to move has been concurrently updated.
              */
             retry = !ExecCrossPartitionUpdate(mtstate, resultRelInfo, tupleid,
                                               oldtuple, slot, planSlot,
                                               epqstate, canSetTag,
                                               &retry_slot, &inserted_tuple);
             if (retry)
             {
                 slot = retry_slot;
                 goto lreplace;
             }
 
             return inserted_tuple;
         }
 
         /*
          * Check the constraints of the tuple.  We've already checked the
          * partition constraint above; however, we must still ensure the tuple
          * passes all other constraints, so we will call ExecConstraints() and
          * have it validate all remaining checks.
          */
         if (resultRelationDesc->rd_att->constr)
             ExecConstraints(resultRelInfo, slot, estate);
 
         /*
          * replace the heap tuple
          *
          * Note: if es_crosscheck_snapshot isn't InvalidSnapshot, we check
          * that the row to be updated is visible to that snapshot, and throw a
          * can't-serialize error if not. This is a special-case behavior
          * needed for referential integrity updates in transaction-snapshot
          * mode transactions.
          */
         result = table_tuple_update(resultRelationDesc, tupleid, slot,
                                     estate->es_output_cid,
                                     estate->es_snapshot,
                                     estate->es_crosscheck_snapshot,
                                     true /* wait for commit */ ,
                                     &tmfd, &lockmode, &update_indexes);
 
         switch (result)
         {
             case TM_SelfModified:
 
                 /*
                  * The target tuple was already updated or deleted by the
                  * current command, or by a later command in the current
                  * transaction.  The former case is possible in a join UPDATE
                  * where multiple tuples join to the same target tuple. This
                  * is pretty questionable, but Postgres has always allowed it:
                  * we just execute the first update action and ignore
                  * additional update attempts.
                  *
                  * The latter case arises if the tuple is modified by a
                  * command in a BEFORE trigger, or perhaps by a command in a
                  * volatile function used in the query.  In such situations we
                  * should not ignore the update, but it is equally unsafe to
                  * proceed.  We don't want to discard the original UPDATE
                  * while keeping the triggered actions based on it; and we
                  * have no principled way to merge this update with the
                  * previous ones.  So throwing an error is the only safe
                  * course.
                  *
                  * If a trigger actually intends this type of interaction, it
                  * can re-execute the UPDATE (assuming it can figure out how)
                  * and then return NULL to cancel the outer update.
                  */
                 if (tmfd.cmax != estate->es_output_cid)
                     ereport(ERROR,
                             (errcode(ERRCODE_TRIGGERED_DATA_CHANGE_VIOLATION),
                              errmsg("tuple to be updated was already modified by an operation triggered by the current command"),
                              errhint("Consider using an AFTER trigger instead of a BEFORE trigger to propagate changes to other rows.")));
 
                 /* Else, already updated by self; nothing to do */
                 return NULL;
 
             case TM_Ok:
                 break;
 
             case TM_Updated:
                 {
                     TupleTableSlot *inputslot;
                     TupleTableSlot *epqslot;
                     TupleTableSlot *oldSlot;
 
                     if (IsolationUsesXactSnapshot())
                         ereport(ERROR,
                                 (errcode(ERRCODE_T_R_SERIALIZATION_FAILURE),
                                  errmsg("could not serialize access due to concurrent update")));
 
                     /*
                      * Already know that we're going to need to do EPQ, so
                      * fetch tuple directly into the right slot.
                      */
                     inputslot = EvalPlanQualSlot(epqstate, resultRelationDesc,
                                                  resultRelInfo->ri_RangeTableIndex);
 
                     result = table_tuple_lock(resultRelationDesc, tupleid,
                                               estate->es_snapshot,
                                               inputslot, estate->es_output_cid,
                                               lockmode, LockWaitBlock,
                                               TUPLE_LOCK_FLAG_FIND_LAST_VERSION,
                                               &tmfd);
 
                     switch (result)
                     {
                         case TM_Ok:
                             Assert(tmfd.traversed);
 
                             epqslot = EvalPlanQual(epqstate,
                                                    resultRelationDesc,
                                                    resultRelInfo->ri_RangeTableIndex,
                                                    inputslot);
                             if (TupIsNull(epqslot))
                                 /* Tuple not passing quals anymore, exiting... */
                                 return NULL;
 
                             /* Make sure ri_oldTupleSlot is initialized. */
                             if (unlikely(!resultRelInfo->ri_projectNewInfoValid))
                                 ExecInitUpdateProjection(mtstate, resultRelInfo);
 
                             /* Fetch the most recent version of old tuple. */
                             oldSlot = resultRelInfo->ri_oldTupleSlot;
                             if (!table_tuple_fetch_row_version(resultRelationDesc,
                                                                tupleid,
                                                                SnapshotAny,
                                                                oldSlot))
                                 elog(ERROR, "failed to fetch tuple being updated");
                             slot = ExecGetUpdateNewTuple(resultRelInfo,
                                                          epqslot, oldSlot);
                             goto lreplace;
 
                         case TM_Deleted:
                             /* tuple already deleted; nothing to do */
                             return NULL;
 
                         case TM_SelfModified:
 
                             /*
                              * This can be reached when following an update
                              * chain from a tuple updated by another session,
                              * reaching a tuple that was already updated in
                              * this transaction. If previously modified by
                              * this command, ignore the redundant update,
                              * otherwise error out.
                              *
                              * See also TM_SelfModified response to
                              * table_tuple_update() above.
                              */
                             if (tmfd.cmax != estate->es_output_cid)
                                 ereport(ERROR,
                                         (errcode(ERRCODE_TRIGGERED_DATA_CHANGE_VIOLATION),
                                          errmsg("tuple to be updated was already modified by an operation triggered by the current command"),
                                          errhint("Consider using an AFTER trigger instead of a BEFORE trigger to propagate changes to other rows.")));
                             return NULL;
 
                         default:
                             /* see table_tuple_lock call in ExecDelete() */
                             elog(ERROR, "unexpected table_tuple_lock status: %u",
                                  result);
                             return NULL;
                     }
                 }
 
                 break;
 
             case TM_Deleted:
                 if (IsolationUsesXactSnapshot())
                     ereport(ERROR,
                             (errcode(ERRCODE_T_R_SERIALIZATION_FAILURE),
                              errmsg("could not serialize access due to concurrent delete")));
                 /* tuple already deleted; nothing to do */
                 return NULL;
 
             default:
                 elog(ERROR, "unrecognized table_tuple_update status: %u",
                      result);
                 return NULL;
         }
 
         /* insert index entries for tuple if necessary */
         if (resultRelInfo->ri_NumIndices > 0 && update_indexes)
             recheckIndexes = ExecInsertIndexTuples(resultRelInfo,
                                                    slot, estate, true, false,
                                                    NULL, NIL);
     }
 
     if (canSetTag)
         (estate->es_processed)++;
 
     /* AFTER ROW UPDATE Triggers */
     ExecARUpdateTriggers(estate, resultRelInfo, tupleid, oldtuple, slot,
                          recheckIndexes,
                          mtstate->operation == CMD_INSERT ?
                          mtstate->mt_oc_transition_capture :
                          mtstate->mt_transition_capture);
 
     list_free(recheckIndexes);
 
     /*
      * Check any WITH CHECK OPTION constraints from parent views.  We are
      * required to do this after testing all constraints and uniqueness
      * violations per the SQL spec, so we do it after actually updating the
      * record in the heap and all indexes.
      *
      * ExecWithCheckOptions() will skip any WCOs which are not of the kind we
      * are looking for at this point.
      */
     if (resultRelInfo->ri_WithCheckOptions != NIL)
         ExecWithCheckOptions(WCO_VIEW_CHECK, resultRelInfo, slot, estate);
 
     /* Process RETURNING if present */
     if (resultRelInfo->ri_projectReturning)
         return ExecProcessReturning(resultRelInfo, slot, planSlot);
 
     return NULL;
 }

◆ fireASTriggers()

static void fireASTriggers ( ModifyTableState * node )

static

Definition at line 2226 of file nodeModifyTable.c.

References CMD_DELETE, CMD_INSERT, CMD_UPDATE, elog, ERROR, ExecASDeleteTriggers(), ExecASInsertTriggers(), ExecASUpdateTriggers(), ModifyTableState::mt_oc_transition_capture, ModifyTableState::mt_transition_capture, ONCONFLICT_UPDATE, ModifyTable::onConflictAction, ModifyTableState::operation, PlanState::plan, ModifyTableState::ps, ModifyTableState::rootResultRelInfo, and PlanState::state.

Referenced by ExecModifyTable().

 {
     ModifyTable *plan = (ModifyTable *) node->ps.plan;
     ResultRelInfo *resultRelInfo = node->rootResultRelInfo;
 
     switch (node->operation)
     {
         case CMD_INSERT:
             if (plan->onConflictAction == ONCONFLICT_UPDATE)
                 ExecASUpdateTriggers(node->ps.state,
                                      resultRelInfo,
                                      node->mt_oc_transition_capture);
             ExecASInsertTriggers(node->ps.state, resultRelInfo,
                                  node->mt_transition_capture);
             break;
         case CMD_UPDATE:
             ExecASUpdateTriggers(node->ps.state, resultRelInfo,
                                  node->mt_transition_capture);
             break;
         case CMD_DELETE:
             ExecASDeleteTriggers(node->ps.state, resultRelInfo,
                                  node->mt_transition_capture);
             break;
         default:
             elog(ERROR, "unknown operation");
             break;
     }
 }

◆ fireBSTriggers()

static void fireBSTriggers ( ModifyTableState * node )

static

Definition at line 2197 of file nodeModifyTable.c.

References CMD_DELETE, CMD_INSERT, CMD_UPDATE, elog, ERROR, ExecBSDeleteTriggers(), ExecBSInsertTriggers(), ExecBSUpdateTriggers(), ONCONFLICT_UPDATE, ModifyTable::onConflictAction, ModifyTableState::operation, PlanState::plan, ModifyTableState::ps, ModifyTableState::rootResultRelInfo, and PlanState::state.

Referenced by ExecModifyTable().

 {
     ModifyTable *plan = (ModifyTable *) node->ps.plan;
     ResultRelInfo *resultRelInfo = node->rootResultRelInfo;
 
     switch (node->operation)
     {
         case CMD_INSERT:
             ExecBSInsertTriggers(node->ps.state, resultRelInfo);
             if (plan->onConflictAction == ONCONFLICT_UPDATE)
                 ExecBSUpdateTriggers(node->ps.state,
                                      resultRelInfo);
             break;
         case CMD_UPDATE:
             ExecBSUpdateTriggers(node->ps.state, resultRelInfo);
             break;
         case CMD_DELETE:
             ExecBSDeleteTriggers(node->ps.state, resultRelInfo);
             break;
         default:
             elog(ERROR, "unknown operation");
             break;
     }
 }

Data Structures

Macros

Typedefs

Functions

Macro Definition Documentation

◆ MT_NRELS_HASH

Typedef Documentation

◆ MTTargetRelLookup

Function Documentation

◆ ExecBatchInsert()

◆ ExecCheckPlanOutput()

◆ ExecCheckTIDVisible()

◆ ExecCheckTupleVisible()

◆ ExecComputeStoredGenerated()

◆ ExecCrossPartitionUpdate()

◆ ExecDelete()

◆ ExecEndModifyTable()

◆ ExecGetInsertNewTuple()

◆ ExecGetUpdateNewTuple()

◆ ExecInitInsertProjection()

◆ ExecInitModifyTable()

◆ ExecInitUpdateProjection()

◆ ExecInsert()

◆ ExecLookupResultRelByOid()

◆ ExecModifyTable()

◆ ExecOnConflictUpdate()

◆ ExecPrepareTupleRouting()

◆ ExecProcessReturning()

◆ ExecReScanModifyTable()

◆ ExecSetupTransitionCaptureState()

◆ ExecUpdate()

◆ fireASTriggers()

◆ fireBSTriggers()