Further twiddling of nodeHash.c hashtable sizing calculation.

tglsfdc · tglsfdc · commit 4075fc4b97f9 · 2015-10-04T15:55:07.000-04:00
On reflection, the submitted patch didn't really work to prevent the
request size from exceeding MaxAllocSize, because of the fact that we'd
happily round nbuckets up to the next power of 2 after we'd limited it to
max_pointers.  The simplest way to enforce the limit correctly is to
round max_pointers down to a power of 2 when it isn't one already.

(Note that the constraint to INT_MAX / 2, if it were doing anything useful
at all, is properly applied after that.)
diff --git a/src/backend/executor/nodeHash.c b/src/backend/executor/nodeHash.c
@@ -397,6 +397,7 @@ ExecChooseHashTableSize(double ntuples, int tupwidth, bool useskew,
 	long		hash_table_bytes;
 	long		skew_table_bytes;
 	long		max_pointers;
+	long		mppow2;
 	int			nbatch;
 	int			nbuckets;
 	int			i;
@@ -464,7 +465,12 @@ ExecChooseHashTableSize(double ntuples, int tupwidth, bool useskew,
 	 */
 	max_pointers = (work_mem * 1024L) / sizeof(HashJoinTuple);
 	max_pointers = Min(max_pointers, MaxAllocSize / sizeof(HashJoinTuple));
-	/* also ensure we avoid integer overflow in nbatch and nbuckets */
+	/* If max_pointers isn't a power of 2, must round it down to one */
+	mppow2 = 1L << my_log2(max_pointers);
+	if (max_pointers != mppow2)
+		max_pointers = mppow2 / 2;
+
+	/* Also ensure we avoid integer overflow in nbatch and nbuckets */
 	/* (this step is redundant given the current value of MaxAllocSize) */
 	max_pointers = Min(max_pointers, INT_MAX / 2);
 
