Catcaches can now store negative entries as well as positive ones, to

speed up repetitive failed searches; per pghackers discussion in late
January.  inval.c logic substantially simplified, since we can now treat
inserts and deletes alike as far as inval events are concerned.  Some
repair work needed in heap_create_with_catalog, which turns out to have
been doing CommandCounterIncrement at a point where the new relation has
non-self-consistent catalog entries.  With the new inval code, that
resulted in assert failures during a relcache entry rebuild.

src/backend/access/heap/heapam.c

@@ -8,7 +8,7 @@
  *
  *
  * IDENTIFICATION
- *	  $Header: /cvsroot/pgsql/src/backend/access/heap/heapam.c,v 1.130 2002/03/02 21:39:17 momjian Exp $
+ *	  $Header: /cvsroot/pgsql/src/backend/access/heap/heapam.c,v 1.131 2002/03/03 17:47:53 tgl Exp $
  *
  *
  * INTERFACE ROUTINES
@@ -1131,12 +1131,12 @@ heap_insert(Relation relation, HeapTuple tup)
 	WriteBuffer(buffer);
 
 	/*
-	 * If tuple is cachable, mark it for rollback from the caches in case
+	 * If tuple is cachable, mark it for invalidation from the caches in case
 	 * we abort.  Note it is OK to do this after WriteBuffer releases the
 	 * buffer, because the "tup" data structure is all in local memory,
 	 * not in the shared buffer.
 	 */
-	RelationMark4RollbackHeapTuple(relation, tup);
+	CacheInvalidateHeapTuple(relation, tup);
 
 	return tup->t_data->t_oid;
 }
@@ -1278,7 +1278,7 @@ l1:
 	 * look at the contents of the tuple, so we need to hold our refcount
 	 * on the buffer.
 	 */
-	RelationInvalidateHeapTuple(relation, &tp);
+	CacheInvalidateHeapTuple(relation, &tp);
 
 	WriteBuffer(buffer);
 
@@ -1585,19 +1585,19 @@ l2:
 	 * boundary. We have to do this before WriteBuffer because we need to
 	 * look at the contents of the tuple, so we need to hold our refcount.
 	 */
-	RelationInvalidateHeapTuple(relation, &oldtup);
+	CacheInvalidateHeapTuple(relation, &oldtup);
 
 	if (newbuf != buffer)
 		WriteBuffer(newbuf);
 	WriteBuffer(buffer);
 
 	/*
-	 * If new tuple is cachable, mark it for rollback from the caches in
+	 * If new tuple is cachable, mark it for invalidation from the caches in
 	 * case we abort.  Note it is OK to do this after WriteBuffer releases
 	 * the buffer, because the "newtup" data structure is all in local
 	 * memory, not in the shared buffer.
 	 */
-	RelationMark4RollbackHeapTuple(relation, newtup);
+	CacheInvalidateHeapTuple(relation, newtup);
 
 	return HeapTupleMayBeUpdated;
 }

src/backend/catalog/index.c

@@ -8,7 +8,7 @@
  *
  *
  * IDENTIFICATION
- *	  $Header: /cvsroot/pgsql/src/backend/catalog/index.c,v 1.172 2002/02/19 20:11:11 tgl Exp $
+ *	  $Header: /cvsroot/pgsql/src/backend/catalog/index.c,v 1.173 2002/03/03 17:47:54 tgl Exp $
  *
  *
  * INTERFACE ROUTINES
@@ -800,19 +800,6 @@ index_drop(Oid indexId)
 	simple_heap_delete(relationRelation, &tuple->t_self);
 	heap_freetuple(tuple);
 
-	/*
-	 * Update the pg_class tuple for the owning relation.  We are
-	 * presently too lazy to attempt to compute the new correct value of
-	 * relhasindex (the next VACUUM will fix it if necessary).	But we
-	 * must send out a shared-cache-inval notice on the owning relation to
-	 * ensure other backends update their relcache lists of indexes.  So,
-	 * unconditionally do setRelhasindex(true).
-	 *
-	 * Possible future improvement: skip the physical tuple update and just
-	 * send out an invalidation message.
-	 */
-	setRelhasindex(heapId, true, false, InvalidOid);
-
 	heap_close(relationRelation, RowExclusiveLock);
 
 	/*
@@ -858,6 +845,15 @@ index_drop(Oid indexId)
 
 	smgrunlink(DEFAULT_SMGR, userIndexRelation);
 
+	/*
+	 * We are presently too lazy to attempt to compute the new correct value
+	 * of relhasindex (the next VACUUM will fix it if necessary).  So there is
+	 * no need to update the pg_class tuple for the owning relation.
+	 * But we must send out a shared-cache-inval notice on the owning relation
+	 * to ensure other backends update their relcache lists of indexes.
+	 */
+	CacheInvalidateRelcache(heapId);
+
 	/*
 	 * Close rels, but keep locks
 	 */
@@ -1076,7 +1072,7 @@ LockClassinfoForUpdate(Oid relid, HeapTuple rtup,
 		}
 		break;
 	}
-	RelationInvalidateHeapTuple(relationRelation, rtup);
+	CacheInvalidateHeapTuple(relationRelation, rtup);
 	if (confirmCommitted)
 	{
 		HeapTupleHeader th = rtup->t_data;
@@ -1137,10 +1133,8 @@ IndexesAreActive(Oid relid, bool confirmCommitted)
  *
  * NOTE: an important side-effect of this operation is that an SI invalidation
  * message is sent out to all backends --- including me --- causing relcache
- * entries to be flushed or updated with the new hasindex data.
- * Therefore, we execute the update even if relhasindex has the right value
- * already.  Possible future improvement: skip the disk update and just send
- * an SI message in that case.
+ * entries to be flushed or updated with the new hasindex data.  This must
+ * happen even if we find that no change is needed in the pg_class row.
  * ----------------
  */
 void
@@ -1149,6 +1143,7 @@ setRelhasindex(Oid relid, bool hasindex, bool isprimary, Oid reltoastidxid)
 	Relation	pg_class;
 	HeapTuple	tuple;
 	Form_pg_class classtuple;
+	bool		dirty = false;
 	HeapScanDesc pg_class_scan = NULL;
 
 	/*
@@ -1192,13 +1187,28 @@ setRelhasindex(Oid relid, bool hasindex, bool isprimary, Oid reltoastidxid)
 		LockBuffer(pg_class_scan->rs_cbuf, BUFFER_LOCK_EXCLUSIVE);
 
 	classtuple = (Form_pg_class) GETSTRUCT(tuple);
+
+	if (classtuple->relhasindex != hasindex)
+	{
 		classtuple->relhasindex = hasindex;
+		dirty = true;
+	}
 	if (isprimary)
+	{
+		if (!classtuple->relhaspkey)
+		{
 			classtuple->relhaspkey = true;
+			dirty = true;
+		}
+	}
 	if (OidIsValid(reltoastidxid))
 	{
 		Assert(classtuple->relkind == RELKIND_TOASTVALUE);
+		if (classtuple->reltoastidxid != reltoastidxid)
+		{
 			classtuple->reltoastidxid = reltoastidxid;
+			dirty = true;
+		}
 	}
 
 	if (pg_class_scan)
@@ -1210,10 +1220,10 @@ setRelhasindex(Oid relid, bool hasindex, bool isprimary, Oid reltoastidxid)
 		WriteNoReleaseBuffer(pg_class_scan->rs_cbuf);
 		/* Send out shared cache inval if necessary */
 		if (!IsBootstrapProcessingMode())
-			RelationInvalidateHeapTuple(pg_class, tuple);
+			CacheInvalidateHeapTuple(pg_class, tuple);
 		BufferSync();
 	}
-	else
+	else if (dirty)
 	{
 		simple_heap_update(pg_class, &tuple->t_self, tuple);
 
@@ -1228,6 +1238,11 @@ setRelhasindex(Oid relid, bool hasindex, bool isprimary, Oid reltoastidxid)
 			CatalogCloseIndices(Num_pg_class_indices, idescs);
 		}
 	}
+	else
+	{
+		/* no need to change tuple, but force relcache rebuild anyway */
+		CacheInvalidateRelcache(relid);
+	}
 
 	if (!pg_class_scan)
 		heap_freetuple(tuple);
@@ -1280,7 +1295,7 @@ setNewRelfilenode(Relation relation)
 		classTuple = &lockTupleData;
 		/* Send out shared cache inval if necessary */
 		if (!IsBootstrapProcessingMode())
-			RelationInvalidateHeapTuple(pg_class, classTuple);
+			CacheInvalidateHeapTuple(pg_class, classTuple);
 		/* Update the buffer in-place */
 		LockBuffer(buffer, BUFFER_LOCK_EXCLUSIVE);
 		((Form_pg_class) GETSTRUCT(classTuple))->relfilenode = newrelfilenode;
@@ -1442,7 +1457,7 @@ UpdateStats(Oid relid, double reltuples)
 		LockBuffer(pg_class_scan->rs_cbuf, BUFFER_LOCK_UNLOCK);
 		WriteNoReleaseBuffer(pg_class_scan->rs_cbuf);
 		if (!IsBootstrapProcessingMode())
-			RelationInvalidateHeapTuple(pg_class, tuple);
+			CacheInvalidateHeapTuple(pg_class, tuple);
 	}
 	else
 	{

src/backend/commands/vacuum.c

@@ -13,7 +13,7 @@
  *
  *
  * IDENTIFICATION
- *	  $Header: /cvsroot/pgsql/src/backend/commands/vacuum.c,v 1.215 2002/03/02 21:39:23 momjian Exp $
+ *	  $Header: /cvsroot/pgsql/src/backend/commands/vacuum.c,v 1.216 2002/03/03 17:47:54 tgl Exp $
  *
  *-------------------------------------------------------------------------
  */
@@ -527,7 +527,7 @@ vac_update_relstats(Oid relid, BlockNumber num_pages, double num_tuples,
 	 * no flush will occur, but no great harm is done since there are no
 	 * noncritical state updates here.)
 	 */
-	RelationInvalidateHeapTuple(rd, &rtup);
+	CacheInvalidateHeapTuple(rd, &rtup);
 
 	/* Write the buffer */
 	WriteBuffer(buffer);
@@ -583,7 +583,7 @@ vac_update_dbstats(Oid dbid,
 	dbform->datfrozenxid = frozenXID;
 
 	/* invalidate the tuple in the cache and write the buffer */
-	RelationInvalidateHeapTuple(relation, tuple);
+	CacheInvalidateHeapTuple(relation, tuple);
 	WriteNoReleaseBuffer(scan->rs_cbuf);
 
 	heap_endscan(scan);
@@ -1796,7 +1796,10 @@ repair_frag(VRelStats *vacrelstats, Relation onerel,
 					 */
 					heap_copytuple_with_tuple(&tuple, &newtup);
 
-					RelationInvalidateHeapTuple(onerel, &tuple);
+					/*
+					 * register invalidation of source tuple in catcaches.
+					 */
+					CacheInvalidateHeapTuple(onerel, &tuple);
 
 					/* NO ELOG(ERROR) TILL CHANGES ARE LOGGED */
 					START_CRIT_SECTION();
@@ -1953,7 +1956,15 @@ repair_frag(VRelStats *vacrelstats, Relation onerel,
 			/* copy tuple */
 			heap_copytuple_with_tuple(&tuple, &newtup);
 
-			RelationInvalidateHeapTuple(onerel, &tuple);
+			/*
+			 * register invalidation of source tuple in catcaches.
+			 *
+			 * (Note: we do not need to register the copied tuple,
+			 * because we are not changing the tuple contents and
+			 * so there cannot be any need to flush negative
+			 * catcache entries.)
+			 */
+			CacheInvalidateHeapTuple(onerel, &tuple);
 
 			/* NO ELOG(ERROR) TILL CHANGES ARE LOGGED */
 			START_CRIT_SECTION();

src/backend/utils/cache/relcache.c

@@ -8,7 +8,7 @@
  *
  *
  * IDENTIFICATION
- *	  $Header: /cvsroot/pgsql/src/backend/utils/cache/relcache.c,v 1.152 2002/02/19 20:11:17 tgl Exp $
+ *	  $Header: /cvsroot/pgsql/src/backend/utils/cache/relcache.c,v 1.153 2002/03/03 17:47:55 tgl Exp $
  *
  *-------------------------------------------------------------------------
  */
@@ -2065,10 +2065,16 @@ RelationBuildLocalRelation(const char *relname,
 		rel->rd_isnailed = true;
 
 	/*
-	 * create a new tuple descriptor from the one passed in (we do this to
-	 * copy it into the cache context)
+	 * create a new tuple descriptor from the one passed in.  We do this
+	 * partly to copy it into the cache context, and partly because the
+	 * new relation can't have any defaults or constraints yet; they
+	 * have to be added in later steps, because they require additions
+	 * to multiple system catalogs.  We can copy attnotnull constraints
+	 * here, however.
 	 */
-	rel->rd_att = CreateTupleDescCopyConstr(tupDesc);
+	rel->rd_att = CreateTupleDescCopy(tupDesc);
+	for (i = 0; i < natts; i++)
+		rel->rd_att->attrs[i]->attnotnull = tupDesc->attrs[i]->attnotnull;
 
 	/*
 	 * initialize relation tuple form (caller may add/override data later)
@@ -2082,8 +2088,6 @@ RelationBuildLocalRelation(const char *relname,
 	rel->rd_rel->relhasoids = true;
 	rel->rd_rel->relnatts = natts;
 	rel->rd_rel->reltype = InvalidOid;
-	if (tupDesc->constr)
-		rel->rd_rel->relchecks = tupDesc->constr->num_check;
 
 	/*
 	 * Insert relation OID and database/tablespace ID into the right

src/include/storage/sinval.h

@@ -7,7 +7,7 @@
  * Portions Copyright (c) 1996-2001, PostgreSQL Global Development Group
  * Portions Copyright (c) 1994, Regents of the University of California
  *
- * $Id: sinval.h,v 1.25 2001/11/05 17:46:35 momjian Exp $
+ * $Id: sinval.h,v 1.26 2002/03/03 17:47:56 tgl Exp $
  *
  *-------------------------------------------------------------------------
  */
@@ -28,22 +28,32 @@
  * are available to identify other inval message types.
  *
  * Shared-inval events are initially driven by detecting tuple inserts,
- * updates and deletions in system catalogs (see RelationInvalidateHeapTuple
- * and RelationMark4RollbackHeapTuple).  Note that some system catalogs have
- * multiple caches on them (with different indexes).  On detecting a tuple
- * invalidation in such a catalog, a separate catcache inval message must be
- * generated for each of its caches.  The catcache inval message carries the
- * hash index for the target tuple, so that the catcache only needs to search
- * one hash chain not all its chains.  Of course this assumes that all the
- * backends are using identical hashing code, but that should be OK.
+ * updates and deletions in system catalogs (see CacheInvalidateHeapTuple).
+ * An update generates two inval events, one for the old tuple and one for
+ * the new --- this is needed to get rid of both positive entries for the
+ * old tuple, and negative cache entries associated with the new tuple's
+ * cache key.  (This could perhaps be optimized down to one event when the
+ * cache key is not changing, but for now we don't bother to try.)  Note that
+ * the inval events themselves don't actually say whether the tuple is being
+ * inserted or deleted.
+ *
+ * Note that some system catalogs have multiple caches on them (with different
+ * indexes).  On detecting a tuple invalidation in such a catalog, separate
+ * catcache inval messages must be generated for each of its caches.  The
+ * catcache inval messages carry the hash value for the target tuple, so
+ * that the catcache only needs to search one hash chain not all its chains,
+ * and so that negative cache entries can be recognized with good accuracy.
+ * (Of course this assumes that all the backends are using identical hashing
+ * code, but that should be OK.)
  */
 
 typedef struct
 {
+	/* note: field layout chosen with an eye to alignment concerns */
 	int16		id;				/* cache ID --- must be first */
-	uint16		hashIndex;		/* hashchain index within this catcache */
-	Oid			dbId;			/* database ID, or 0 if a shared relation */
 	ItemPointerData tuplePtr;	/* tuple identifier in cached relation */
+	Oid			dbId;			/* database ID, or 0 if a shared relation */
+	uint32		hashValue;		/* hash value of key for this catcache */
 } SharedInvalCatcacheMsg;
 
 #define SHAREDINVALRELCACHE_ID	(-1)

src/include/utils/catcache.h

@@ -13,7 +13,7 @@
  * Portions Copyright (c) 1996-2001, PostgreSQL Global Development Group
  * Portions Copyright (c) 1994, Regents of the University of California
  *
- * $Id: catcache.h,v 1.38 2002/02/19 20:11:19 tgl Exp $
+ * $Id: catcache.h,v 1.39 2002/03/03 17:47:56 tgl Exp $
  *
  *-------------------------------------------------------------------------
  */
@@ -35,20 +35,28 @@ typedef struct catcache
 	struct catcache *cc_next;	/* link to next catcache */
 	char	   *cc_relname;		/* name of relation the tuples come from */
 	char	   *cc_indname;		/* name of index matching cache keys */
-	int			cc_reloidattr;	/* AttrNumber of relation OID, or 0 */
+	Oid			cc_reloid;		/* OID of relation the tuples come from */
 	bool		cc_relisshared; /* is relation shared? */
 	TupleDesc	cc_tupdesc;		/* tuple descriptor (copied from reldesc) */
+	int			cc_reloidattr;	/* AttrNumber of relation OID attr, or 0 */
 	int			cc_ntup;		/* # of tuples currently in this cache */
 	int			cc_size;		/* # of hash buckets in this cache */
 	int			cc_nkeys;		/* number of keys (1..4) */
 	int			cc_key[4];		/* AttrNumber of each key */
 	PGFunction	cc_hashfunc[4]; /* hash function to use for each key */
 	ScanKeyData cc_skey[4];		/* precomputed key info for heap scans */
+	bool		cc_isname[4];	/* flag key columns that are NAMEs */
 #ifdef CATCACHE_STATS
 	long		cc_searches;	/* total # searches against this cache */
 	long		cc_hits;		/* # of matches against existing entry */
+	long		cc_neg_hits;	/* # of matches against negative entry */
 	long		cc_newloads;	/* # of successful loads of new entry */
-	/* cc_searches - (cc_hits + cc_newloads) is # of failed searches */
+	/*
+	 * cc_searches - (cc_hits + cc_neg_hits + cc_newloads) is number of
+	 * failed searches, each of which will result in loading a negative entry
+	 */
+	long		cc_invals;		/* # of entries invalidated from cache */
+	long		cc_discards;	/* # of entries discarded due to overflow */
 #endif
 	Dllist		cc_bucket[1];	/* hash buckets --- VARIABLE LENGTH ARRAY */
 } CatCache;						/* VARIABLE LENGTH STRUCT */
@@ -68,11 +76,18 @@ typedef struct catctup
 	 * A tuple marked "dead" must not be returned by subsequent searches.
 	 * However, it won't be physically deleted from the cache until its
 	 * refcount goes to zero.
+	 *
+	 * A negative cache entry is an assertion that there is no tuple
+	 * matching a particular key.  This is just as useful as a normal entry
+	 * so far as avoiding catalog searches is concerned.  Management of
+	 * positive and negative entries is identical.
 	 */
 	Dlelem		lrulist_elem;	/* list member of global LRU list */
 	Dlelem		cache_elem;		/* list member of per-bucket list */
 	int			refcount;		/* number of active references */
 	bool		dead;			/* dead but not yet removed? */
+	bool		negative;		/* negative cache entry? */
+	uint32		hash_value;		/* hash value for this tuple's keys */
 	HeapTupleData tuple;		/* tuple management header */
 } CatCTup;
 
@@ -104,10 +119,10 @@ extern void ReleaseCatCache(HeapTuple tuple);
 
 extern void ResetCatalogCaches(void);
 extern void CatalogCacheFlushRelation(Oid relId);
-extern void CatalogCacheIdInvalidate(int cacheId, Index hashIndex,
+extern void CatalogCacheIdInvalidate(int cacheId, uint32 hashValue,
 						 ItemPointer pointer);
 extern void PrepareToInvalidateCacheTuple(Relation relation,
 							  HeapTuple tuple,
-						void (*function) (int, Index, ItemPointer, Oid));
+						void (*function) (int, uint32, ItemPointer, Oid));
 
 #endif   /* CATCACHE_H */

src/include/utils/inval.h

@@ -7,7 +7,7 @@
  * Portions Copyright (c) 1996-2001, PostgreSQL Global Development Group
  * Portions Copyright (c) 1994, Regents of the University of California
  *
- * $Id: inval.h,v 1.23 2001/11/05 17:46:36 momjian Exp $
+ * $Id: inval.h,v 1.24 2002/03/03 17:47:56 tgl Exp $
  *
  *-------------------------------------------------------------------------
  */
@@ -23,8 +23,8 @@ extern void AtEOXactInvalidationMessages(bool isCommit);
 
 extern void CommandEndInvalidationMessages(bool isCommit);
 
-extern void RelationInvalidateHeapTuple(Relation relation, HeapTuple tuple);
+extern void CacheInvalidateHeapTuple(Relation relation, HeapTuple tuple);
 
-extern void RelationMark4RollbackHeapTuple(Relation relation, HeapTuple tuple);
+extern void CacheInvalidateRelcache(Oid relationId);
 
 #endif   /* INVAL_H */