Shared-memory hashtables have non-extensible directories, which means

it's a good idea to choose the directory size based on the expected
number of entries.  But ShmemInitHash was using a hard-wired constant.
Boo hiss.  This accounts for recent report of postmaster failure when
asking for 64K or more buffers.

src/backend/storage/ipc/shmem.c

@@ -8,7 +8,7 @@
  *
  *
  * IDENTIFICATION
- *	  $Header: /cvsroot/pgsql/src/backend/storage/ipc/shmem.c,v 1.48 2000/01/26 05:56:58 momjian Exp $
+ *	  $Header: /cvsroot/pgsql/src/backend/storage/ipc/shmem.c,v 1.49 2000/02/26 05:25:55 tgl Exp $
  *
  *-------------------------------------------------------------------------
  */
@@ -332,7 +332,7 @@ ShmemIsValid(unsigned long addr)
 HTAB *
 ShmemInitHash(char *name,		/* table string name for shmem index */
 			  long init_size,	/* initial table size */
-			  long max_size,	/* max size of the table (NOT USED) */
+			  long max_size,	/* max size of the table */
 			  HASHCTL *infoP,	/* info about key and bucket size */
 			  int hash_flags)	/* info about infoP */
 {
@@ -340,19 +340,21 @@ ShmemInitHash(char *name,		/* table string name for shmem index */
 	long	   *location;
 
 	/*
-	 * Hash tables allocated in shared memory have a fixed directory; it
+	 * Hash tables allocated in shared memory have a fixed directory;
-	 * can't grow or other backends wouldn't be able to find it. The
+	 * it can't grow or other backends wouldn't be able to find it.
-	 * segbase is for calculating pointer values. The shared memory
+	 * So, make sure we make it big enough to start with.
+	 *
+	 * The segbase is for calculating pointer values. The shared memory
 	 * allocator must be specified too.
 	 */
-	infoP->dsize = infoP->max_dsize = DEF_DIRSIZE;
+	infoP->dsize = infoP->max_dsize = hash_select_dirsize(max_size);
 	infoP->segbase = (long *) ShmemBase;
 	infoP->alloc = ShmemAlloc;
 	hash_flags |= HASH_SHARED_MEM | HASH_DIRSIZE;
 
 	/* look it up in the shmem index */
 	location = ShmemInitStruct(name,
-						 sizeof(HHDR) + DEF_DIRSIZE * sizeof(SEG_OFFSET),
+						 sizeof(HHDR) + infoP->dsize * sizeof(SEG_OFFSET),
 							   &found);
 
 	/*

src/backend/utils/hash/dynahash.c

@@ -8,7 +8,7 @@
  *
  *
  * IDENTIFICATION
- *	  $Header: /cvsroot/pgsql/src/backend/utils/hash/dynahash.c,v 1.28 2000/01/26 05:57:24 momjian Exp $
+ *	  $Header: /cvsroot/pgsql/src/backend/utils/hash/dynahash.c,v 1.29 2000/02/26 05:25:54 tgl Exp $
  *
  *-------------------------------------------------------------------------
  */
@@ -328,10 +328,7 @@ init_htab(HTAB *hashp, int nelem)
 	{
 		*segp = seg_alloc(hashp);
 		if (*segp == (SEG_OFFSET) 0)
-		{
+			return -1;
-			hash_destroy(hashp);
-			return 0;
-		}
 	}
 
 #if HASH_DEBUG
@@ -392,6 +389,34 @@ hash_estimate_size(long num_entries, long keysize, long datasize)
 	return size;
 }
 
+/*
+ * Select an appropriate directory size for a hashtable with the given
+ * maximum number of entries.
+ * This is only needed for hashtables in shared memory, whose directories
+ * cannot be expanded dynamically.
+ * NB: assumes that all hash structure parameters have default values!
+ *
+ * XXX this had better agree with the behavior of init_htab()...
+ */
+long
+hash_select_dirsize(long num_entries)
+{
+	long		nBuckets,
+				nSegments,
+				nDirEntries;
+
+	/* estimate number of buckets wanted */
+	nBuckets = 1L << my_log2((num_entries - 1) / DEF_FFACTOR + 1);
+	/* # of segments needed for nBuckets */
+	nSegments = 1L << my_log2((nBuckets - 1) / DEF_SEGSIZE + 1);
+	/* directory entries */
+	nDirEntries = DEF_DIRSIZE;
+	while (nDirEntries < nSegments)
+		nDirEntries <<= 1;		/* dir_alloc doubles dsize at each call */
+
+	return nDirEntries;
+}
+
 
 /********************** DESTROY ROUTINES ************************/
 

src/include/utils/hsearch.h

@@ -1,13 +1,13 @@
 /*-------------------------------------------------------------------------
  *
  * hsearch.h
- *	  for hashing in the new buffer manager
+ *	  for hash tables, particularly hash tables in shared memory
  *
  *
  * Portions Copyright (c) 1996-2000, PostgreSQL, Inc
  * Portions Copyright (c) 1994, Regents of the University of California
  *
- * $Id: hsearch.h,v 1.13 2000/01/26 05:58:38 momjian Exp $
+ * $Id: hsearch.h,v 1.14 2000/02/26 05:25:53 tgl Exp $
  *
  *-------------------------------------------------------------------------
  */
@@ -25,12 +25,15 @@
  * whole lot of records per bucket or performance goes down.
  *
  * In a hash table allocated in shared memory, the directory cannot be
- * expanded because it must stay at a fixed address.
+ * expanded because it must stay at a fixed address.  The directory size
+ * should be selected using hash_select_dirsize (and you'd better have
+ * a good idea of the maximum number of entries!).  For non-shared hash
+ * tables, the initial directory size can be left at the default.
  */
 #define DEF_SEGSIZE			   256
-#define DEF_SEGSIZE_SHIFT	   8/* log2(SEGSIZE)  */
+#define DEF_SEGSIZE_SHIFT	   8		/* must be log2(DEF_SEGSIZE) */
 #define DEF_DIRSIZE			   256
-#define DEF_FFACTOR			   1/* default fill factor */
+#define DEF_FFACTOR			   1		/* default fill factor */
 
 #define PRIME1				   37		/* for the hash function */
 #define PRIME2				   1048583
@@ -42,13 +45,13 @@
  */
 typedef struct element
 {
-	unsigned long next;			/* secret from user		 */
+	unsigned long next;			/* secret from user */
 	long		key;
 } ELEMENT;
 
 typedef unsigned long BUCKET_INDEX;
 
-/* segment is an array of bucket pointers  */
+/* segment is an array of bucket pointers */
 typedef BUCKET_INDEX *SEGMENT;
 typedef unsigned long SEG_OFFSET;
 
@@ -65,10 +68,8 @@ typedef struct hashhdr
 	long		nsegs;			/* Number of allocated segments */
 	long		keysize;		/* hash key length in bytes */
 	long		datasize;		/* elem data length in bytes */
-	long		max_dsize;		/* 'dsize' limit if directory is fixed
+	long		max_dsize;		/* 'dsize' limit if directory is fixed size */
-								 * size */
+	BUCKET_INDEX freeBucketIndex;	/* index of first free bucket */
-	BUCKET_INDEX freeBucketIndex;
-	/* index of first free bucket */
 #ifdef HASH_STATISTICS
 	long		accesses;
 	long		collisions;
@@ -84,7 +85,6 @@ typedef struct htab
 	SEG_OFFSET *dir;			/* 'directory' of segm starts */
 	long	   *(*alloc) ();	/* memory allocator (long * for alignment
 								 * reasons) */
-
 } HTAB;
 
 typedef struct hashctl
@@ -115,7 +115,7 @@ typedef struct hashctl
 #define HASH_ALLOC		0x100	/* Setting memory allocator */
 
 
-/* seg_alloc assumes that INVALID_INDEX is 0*/
+/* seg_alloc assumes that INVALID_INDEX is 0 */
 #define INVALID_INDEX			(0)
 #define NO_MAX_DSIZE			(-1)
 /* number of hash buckets allocated at once */
@@ -141,6 +141,7 @@ extern long *hash_search(HTAB *hashp, char *keyPtr, HASHACTION action,
 			bool *foundPtr);
 extern long *hash_seq(HTAB *hashp);
 extern long hash_estimate_size(long num_entries, long keysize, long datasize);
+extern long hash_select_dirsize(long num_entries);
 
 /*
  * prototypes from functions in hashfn.c