OSDN Git Service

Adjust scan plan nodes to avoid getting an extra AccessShareLock on a
authorTom Lane <tgl@sss.pgh.pa.us>
Fri, 2 Dec 2005 20:03:42 +0000 (20:03 +0000)
committerTom Lane <tgl@sss.pgh.pa.us>
Fri, 2 Dec 2005 20:03:42 +0000 (20:03 +0000)
relation if it's already been locked by execMain.c as either a result
relation or a FOR UPDATE/SHARE relation.  This avoids an extra trip to
the shared lock manager state.  Per my suggestion yesterday.

src/backend/executor/execMain.c
src/backend/executor/execUtils.c
src/backend/executor/nodeBitmapHeapscan.c
src/backend/executor/nodeBitmapIndexscan.c
src/backend/executor/nodeIndexscan.c
src/backend/executor/nodeSeqscan.c
src/backend/executor/nodeTidscan.c
src/include/executor/executor.h
src/include/nodes/execnodes.h

index 4228a8f..22ba788 100644 (file)
@@ -26,7 +26,7 @@
  *
  *
  * IDENTIFICATION
- *       $PostgreSQL: pgsql/src/backend/executor/execMain.c,v 1.261 2005/11/22 18:17:10 momjian Exp $
+ *       $PostgreSQL: pgsql/src/backend/executor/execMain.c,v 1.262 2005/12/02 20:03:40 tgl Exp $
  *
  *-------------------------------------------------------------------------
  */
 #include "utils/memutils.h"
 
 
-typedef struct execRowMark
-{
-       Relation        relation;
-       Index           rti;
-       char            resname[32];
-} execRowMark;
-
 typedef struct evalPlanQual
 {
        Index           rti;
@@ -567,10 +560,10 @@ InitPlan(QueryDesc *queryDesc, bool explainOnly)
                        Index           rti = lfirst_int(l);
                        Oid                     relid = getrelid(rti, rangeTable);
                        Relation        relation;
-                       execRowMark *erm;
+                       ExecRowMark *erm;
 
                        relation = heap_open(relid, RowShareLock);
-                       erm = (execRowMark *) palloc(sizeof(execRowMark));
+                       erm = (ExecRowMark *) palloc(sizeof(ExecRowMark));
                        erm->relation = relation;
                        erm->rti = rti;
                        snprintf(erm->resname, sizeof(erm->resname), "ctid%u", rti);
@@ -1020,7 +1013,7 @@ ExecEndPlan(PlanState *planstate, EState *estate)
         */
        foreach(l, estate->es_rowMarks)
        {
-               execRowMark *erm = lfirst(l);
+               ExecRowMark *erm = lfirst(l);
 
                heap_close(erm->relation, NoLock);
        }
@@ -1165,7 +1158,7 @@ lnext:    ;
                lmark:  ;
                                foreach(l, estate->es_rowMarks)
                                {
-                                       execRowMark *erm = lfirst(l);
+                                       ExecRowMark *erm = lfirst(l);
                                        HeapTupleData tuple;
                                        Buffer          buffer;
                                        ItemPointerData update_ctid;
@@ -1859,9 +1852,9 @@ EvalPlanQual(EState *estate, Index rti,
                relation = NULL;
                foreach(l, estate->es_rowMarks)
                {
-                       if (((execRowMark *) lfirst(l))->rti == rti)
+                       if (((ExecRowMark *) lfirst(l))->rti == rti)
                        {
-                               relation = ((execRowMark *) lfirst(l))->relation;
+                               relation = ((ExecRowMark *) lfirst(l))->relation;
                                break;
                        }
                }
index 5221624..c814b97 100644 (file)
@@ -8,7 +8,7 @@
  *
  *
  * IDENTIFICATION
- *       $PostgreSQL: pgsql/src/backend/executor/execUtils.c,v 1.129 2005/11/23 20:27:57 tgl Exp $
+ *       $PostgreSQL: pgsql/src/backend/executor/execUtils.c,v 1.130 2005/12/02 20:03:40 tgl Exp $
  *
  *-------------------------------------------------------------------------
  */
@@ -24,6 +24,9 @@
  *             ExecAssignResultType
  *             etc
  *
+ *             ExecOpenScanRelation    Common code for scan node init routines.
+ *             ExecCloseScanRelation
+ *
  *             ExecOpenIndices                 \
  *             ExecCloseIndices                 | referenced by InitPlan, EndPlan,
  *             ExecInsertIndexTuples   /  ExecInsert, ExecUpdate
@@ -45,6 +48,7 @@
 #include "catalog/pg_index.h"
 #include "executor/execdebug.h"
 #include "miscadmin.h"
+#include "parser/parsetree.h"
 #include "utils/builtins.h"
 #include "utils/fmgroids.h"
 #include "utils/memutils.h"
@@ -685,6 +689,90 @@ ExecAssignScanTypeFromOuterPlan(ScanState *scanstate)
 
 
 /* ----------------------------------------------------------------
+ *                               Scan node support
+ * ----------------------------------------------------------------
+ */
+
+/* ----------------------------------------------------------------
+ *             ExecOpenScanRelation
+ *
+ *             Open the heap relation to be scanned by a base-level scan plan node.
+ *             This should be called during the node's ExecInit routine.
+ *
+ * By default, this acquires AccessShareLock on the relation.  However,
+ * if the relation was already locked by InitPlan, we don't need to acquire
+ * any additional lock.  This saves trips to the shared lock manager.
+ * ----------------------------------------------------------------
+ */
+Relation
+ExecOpenScanRelation(EState *estate, Index scanrelid)
+{
+       RangeTblEntry *rtentry;
+       Oid                     reloid;
+       LOCKMODE        lockmode;
+       ResultRelInfo *resultRelInfos;
+       int                     i;
+
+       /*
+        * First determine the lock type we need.  Scan to see if target relation
+        * is either a result relation or a FOR UPDATE/FOR SHARE relation.
+        */
+       lockmode = AccessShareLock;
+       resultRelInfos = estate->es_result_relations;
+       for (i = 0; i < estate->es_num_result_relations; i++)
+       {
+               if (resultRelInfos[i].ri_RangeTableIndex == scanrelid)
+               {
+                       lockmode = NoLock;
+                       break;
+               }
+       }
+
+       if (lockmode == AccessShareLock)
+       {
+               ListCell   *l;
+
+               foreach(l, estate->es_rowMarks)
+               {
+                       ExecRowMark *erm = lfirst(l);
+
+                       if (erm->rti == scanrelid)
+                       {
+                               lockmode = NoLock;
+                               break;
+                       }
+               }
+       }
+
+       /* OK, open the relation and acquire lock as needed */
+       rtentry = rt_fetch(scanrelid, estate->es_range_table);
+       reloid = rtentry->relid;
+
+       return heap_open(reloid, lockmode);
+}
+
+/* ----------------------------------------------------------------
+ *             ExecCloseScanRelation
+ *
+ *             Close the heap relation scanned by a base-level scan plan node.
+ *             This should be called during the node's ExecEnd routine.
+ *
+ * Currently, we do not release the lock acquired by ExecOpenScanRelation.
+ * This lock should be held till end of transaction.  (There is a faction
+ * that considers this too much locking, however.)
+ *
+ * If we did want to release the lock, we'd have to repeat the logic in
+ * ExecOpenScanRelation in order to figure out what to release.
+ * ----------------------------------------------------------------
+ */
+void
+ExecCloseScanRelation(Relation scanrel)
+{
+       heap_close(scanrel, NoLock);
+}
+
+
+/* ----------------------------------------------------------------
  *                               ExecInsertIndexTuples support
  * ----------------------------------------------------------------
  */
@@ -760,7 +848,7 @@ ExecOpenIndices(ResultRelInfo *resultRelInfo)
                 *
                 * If the index AM is not safe for concurrent updates, obtain an
                 * exclusive lock on the index to lock out other updaters as well as
-                * readers (index_beginscan places AccessShareLock).
+                * readers (index_beginscan places AccessShareLock on the index).
                 *
                 * If there are multiple not-concurrent-safe indexes, all backends
                 * must lock the indexes in the same order or we will get deadlocks
index 37bd42d..959b559 100644 (file)
@@ -21,7 +21,7 @@
  *
  *
  * IDENTIFICATION
- *       $PostgreSQL: pgsql/src/backend/executor/nodeBitmapHeapscan.c,v 1.7 2005/12/02 01:29:55 tgl Exp $
+ *       $PostgreSQL: pgsql/src/backend/executor/nodeBitmapHeapscan.c,v 1.8 2005/12/02 20:03:40 tgl Exp $
  *
  *-------------------------------------------------------------------------
  */
@@ -141,9 +141,9 @@ BitmapHeapNext(BitmapHeapScanState *node)
 
                        /*
                         * Ignore any claimed entries past what we think is the end of the
-                        * relation.  (This is probably not necessary given that we got
-                        * AccessShareLock before performing any of the indexscans, but
-                        * let's be safe.)
+                        * relation.  (This is probably not necessary given that we got at
+                        * least AccessShareLock on the table before performing any of the
+                        * indexscans, but let's be safe.)
                         */
                        if (tbmres->blockno >= scan->rs_nblocks)
                        {
@@ -448,13 +448,8 @@ ExecEndBitmapHeapScan(BitmapHeapScanState *node)
 
        /*
         * close the heap relation.
-        *
-        * Currently, we do not release the AccessShareLock acquired by
-        * ExecInitBitmapHeapScan.      This lock should be held till end of
-        * transaction. (There is a faction that considers this too much locking,
-        * however.)
         */
-       heap_close(relation, NoLock);
+       ExecCloseScanRelation(relation);
 }
 
 /* ----------------------------------------------------------------
@@ -467,9 +462,6 @@ BitmapHeapScanState *
 ExecInitBitmapHeapScan(BitmapHeapScan *node, EState *estate)
 {
        BitmapHeapScanState *scanstate;
-       RangeTblEntry *rtentry;
-       Index           relid;
-       Oid                     reloid;
        Relation        currentRelation;
 
        /*
@@ -519,13 +511,9 @@ ExecInitBitmapHeapScan(BitmapHeapScan *node, EState *estate)
        CXT1_printf("ExecInitBitmapHeapScan: context is %d\n", CurrentMemoryContext);
 
        /*
-        * open the base relation and acquire AccessShareLock on it.
+        * open the base relation and acquire appropriate lock on it.
         */
-       relid = node->scan.scanrelid;
-       rtentry = rt_fetch(relid, estate->es_range_table);
-       reloid = rtentry->relid;
-
-       currentRelation = heap_open(reloid, AccessShareLock);
+       currentRelation = ExecOpenScanRelation(estate, node->scan.scanrelid);
 
        scanstate->ss.ss_currentRelation = currentRelation;
 
index 46e8c14..4217de3 100644 (file)
@@ -8,7 +8,7 @@
  *
  *
  * IDENTIFICATION
- *       $PostgreSQL: pgsql/src/backend/executor/nodeBitmapIndexscan.c,v 1.12 2005/11/25 19:47:49 tgl Exp $
+ *       $PostgreSQL: pgsql/src/backend/executor/nodeBitmapIndexscan.c,v 1.13 2005/12/02 20:03:40 tgl Exp $
  *
  *-------------------------------------------------------------------------
  */
@@ -286,8 +286,8 @@ ExecInitBitmapIndexScan(BitmapIndexScan *node, EState *estate)
 
        /*
         * We do not open or lock the base relation here.  We assume that an
-        * ancestor BitmapHeapScan node is holding AccessShareLock on the heap
-        * relation throughout the execution of the plan tree.
+        * ancestor BitmapHeapScan node is holding AccessShareLock (or better)
+        * on the heap relation throughout the execution of the plan tree.
         */
 
        indexstate->ss.ss_currentRelation = NULL;
index f34a14c..4beecfb 100644 (file)
@@ -8,7 +8,7 @@
  *
  *
  * IDENTIFICATION
- *       $PostgreSQL: pgsql/src/backend/executor/nodeIndexscan.c,v 1.107 2005/11/25 19:47:49 tgl Exp $
+ *       $PostgreSQL: pgsql/src/backend/executor/nodeIndexscan.c,v 1.108 2005/12/02 20:03:40 tgl Exp $
  *
  *-------------------------------------------------------------------------
  */
@@ -421,12 +421,8 @@ ExecEndIndexScan(IndexScanState *node)
 
        /*
         * close the heap relation.
-        *
-        * Currently, we do not release the AccessShareLock acquired by
-        * ExecInitIndexScan.  This lock should be held till end of transaction.
-        * (There is a faction that considers this too much locking, however.)
         */
-       heap_close(relation, NoLock);
+       ExecCloseScanRelation(relation);
 }
 
 /* ----------------------------------------------------------------
@@ -464,9 +460,6 @@ IndexScanState *
 ExecInitIndexScan(IndexScan *node, EState *estate)
 {
        IndexScanState *indexstate;
-       RangeTblEntry *rtentry;
-       Index           relid;
-       Oid                     reloid;
        Relation        currentRelation;
 
        /*
@@ -551,13 +544,9 @@ ExecInitIndexScan(IndexScan *node, EState *estate)
        }
 
        /*
-        * open the base relation and acquire AccessShareLock on it.
+        * open the base relation and acquire appropriate lock on it.
         */
-       relid = node->scan.scanrelid;
-       rtentry = rt_fetch(relid, estate->es_range_table);
-       reloid = rtentry->relid;
-
-       currentRelation = heap_open(reloid, AccessShareLock);
+       currentRelation = ExecOpenScanRelation(estate, node->scan.scanrelid);
 
        indexstate->ss.ss_currentRelation = currentRelation;
        indexstate->ss.ss_currentScanDesc = NULL;       /* no heap scan here */
@@ -570,7 +559,7 @@ ExecInitIndexScan(IndexScan *node, EState *estate)
        /*
         * open the index relation and initialize relation and scan descriptors.
         * Note we acquire no locks here; the index machinery does its own locks
-        * and unlocks.  (We rely on having AccessShareLock on the parent table to
+        * and unlocks.  (We rely on having a lock on the parent table to
         * ensure the index won't go away!)
         */
        indexstate->iss_RelationDesc = index_open(node->indexid);
index 391fdf7..5e8ba95 100644 (file)
@@ -8,7 +8,7 @@
  *
  *
  * IDENTIFICATION
- *       $PostgreSQL: pgsql/src/backend/executor/nodeSeqscan.c,v 1.55 2005/11/25 04:24:48 tgl Exp $
+ *       $PostgreSQL: pgsql/src/backend/executor/nodeSeqscan.c,v 1.56 2005/12/02 20:03:40 tgl Exp $
  *
  *-------------------------------------------------------------------------
  */
@@ -141,25 +141,15 @@ ExecSeqScan(SeqScanState *node)
 static void
 InitScanRelation(SeqScanState *node, EState *estate)
 {
-       Index           relid;
-       List       *rangeTable;
-       RangeTblEntry *rtentry;
-       Oid                     reloid;
        Relation        currentRelation;
        HeapScanDesc currentScanDesc;
 
        /*
         * get the relation object id from the relid'th entry in the range table,
-        * open that relation and initialize the scan state.
-        *
-        * We acquire AccessShareLock for the duration of the scan.
+        * open that relation and acquire appropriate lock on it.
         */
-       relid = ((SeqScan *) node->ps.plan)->scanrelid;
-       rangeTable = estate->es_range_table;
-       rtentry = rt_fetch(relid, rangeTable);
-       reloid = rtentry->relid;
-
-       currentRelation = heap_open(reloid, AccessShareLock);
+       currentRelation = ExecOpenScanRelation(estate,
+                                                                                  ((SeqScan *) node->ps.plan)->scanrelid);
 
        currentScanDesc = heap_beginscan(currentRelation,
                                                                         estate->es_snapshot,
@@ -281,12 +271,8 @@ ExecEndSeqScan(SeqScanState *node)
 
        /*
         * close the heap relation.
-        *
-        * Currently, we do not release the AccessShareLock acquired by
-        * InitScanRelation.  This lock should be held till end of transaction.
-        * (There is a faction that considers this too much locking, however.)
         */
-       heap_close(relation, NoLock);
+       ExecCloseScanRelation(relation);
 }
 
 /* ----------------------------------------------------------------
index ba4b7f3..e6d61c7 100644 (file)
@@ -8,7 +8,7 @@
  *
  *
  * IDENTIFICATION
- *       $PostgreSQL: pgsql/src/backend/executor/nodeTidscan.c,v 1.45 2005/11/26 22:14:56 tgl Exp $
+ *       $PostgreSQL: pgsql/src/backend/executor/nodeTidscan.c,v 1.46 2005/12/02 20:03:41 tgl Exp $
  *
  *-------------------------------------------------------------------------
  */
@@ -421,12 +421,8 @@ ExecEndTidScan(TidScanState *node)
 
        /*
         * close the heap relation.
-        *
-        * Currently, we do not release the AccessShareLock acquired by
-        * ExecInitTidScan.  This lock should be held till end of transaction.
-        * (There is a faction that considers this too much locking, however.)
         */
-       heap_close(node->ss.ss_currentRelation, NoLock);
+       ExecCloseScanRelation(node->ss.ss_currentRelation);
 }
 
 /* ----------------------------------------------------------------
@@ -472,9 +468,6 @@ TidScanState *
 ExecInitTidScan(TidScan *node, EState *estate)
 {
        TidScanState *tidstate;
-       RangeTblEntry *rtentry;
-       Oid                     relid;
-       Oid                     reloid;
        Relation        currentRelation;
 
        /*
@@ -521,15 +514,9 @@ ExecInitTidScan(TidScan *node, EState *estate)
        tidstate->tss_TidPtr = -1;
 
        /*
-        * open the base relation
-        *
-        * We acquire AccessShareLock for the duration of the scan.
+        * open the base relation and acquire appropriate lock on it.
         */
-       relid = node->scan.scanrelid;
-       rtentry = rt_fetch(relid, estate->es_range_table);
-       reloid = rtentry->relid;
-
-       currentRelation = heap_open(reloid, AccessShareLock);
+       currentRelation = ExecOpenScanRelation(estate, node->scan.scanrelid);
 
        tidstate->ss.ss_currentRelation = currentRelation;
        tidstate->ss.ss_currentScanDesc = NULL;         /* no heap scan here */
index 39c76c6..f1d4e37 100644 (file)
@@ -7,7 +7,7 @@
  * Portions Copyright (c) 1996-2005, PostgreSQL Global Development Group
  * Portions Copyright (c) 1994, Regents of the University of California
  *
- * $PostgreSQL: pgsql/src/include/executor/executor.h,v 1.121 2005/11/23 20:27:58 tgl Exp $
+ * $PostgreSQL: pgsql/src/include/executor/executor.h,v 1.122 2005/12/02 20:03:42 tgl Exp $
  *
  *-------------------------------------------------------------------------
  */
@@ -230,6 +230,9 @@ extern void ExecAssignScanType(ScanState *scanstate,
                                   TupleDesc tupDesc, bool shouldFree);
 extern void ExecAssignScanTypeFromOuterPlan(ScanState *scanstate);
 
+extern Relation ExecOpenScanRelation(EState *estate, Index scanrelid);
+extern void ExecCloseScanRelation(Relation scanrel);
+
 extern void ExecOpenIndices(ResultRelInfo *resultRelInfo);
 extern void ExecCloseIndices(ResultRelInfo *resultRelInfo);
 extern void ExecInsertIndexTuples(TupleTableSlot *slot, ItemPointer tupleid,
index 7371f95..062985a 100644 (file)
@@ -7,7 +7,7 @@
  * Portions Copyright (c) 1996-2005, PostgreSQL Global Development Group
  * Portions Copyright (c) 1994, Regents of the University of California
  *
- * $PostgreSQL: pgsql/src/include/nodes/execnodes.h,v 1.145 2005/11/28 23:46:03 tgl Exp $
+ * $PostgreSQL: pgsql/src/include/nodes/execnodes.h,v 1.146 2005/12/02 20:03:42 tgl Exp $
  *
  *-------------------------------------------------------------------------
  */
@@ -346,6 +346,15 @@ typedef struct EState
 } EState;
 
 
+/* es_rowMarks is a list of these structs: */
+typedef struct ExecRowMark
+{
+       Relation        relation;               /* opened and RowShareLock'd relation */
+       Index           rti;                    /* its range table index */
+       char            resname[32];    /* name for its ctid junk attribute */
+} ExecRowMark;
+
+
 /* ----------------------------------------------------------------
  *                              Tuple Hash Tables
  *