From 3e2c198e06a5c67f8adbcaf53c7d65ed504c72dd Mon Sep 17 00:00:00 2001 From: Tom Lane Date: Fri, 1 Jun 2007 15:58:02 +0000 Subject: [PATCH] Fix performance problems in multi-batch hash joins by ensuring that we select a well-randomized batch number even when given a poorly-randomized hash value. This is a bit inefficient but seems the only practical solution given the constraint that we can't change the hash functions in released branches. Per report from Joseph Shraibman. Applied to 8.1 and 8.2 only --- HEAD is getting a cleaner fix, and 8.0 and before use different coding that seems less vulnerable. --- src/backend/access/hash/hashfunc.c | 27 ++++++++++++++++++++++++++- src/backend/executor/nodeHash.c | 13 ++++++++----- src/include/access/hash.h | 3 ++- 3 files changed, 36 insertions(+), 7 deletions(-) diff --git a/src/backend/access/hash/hashfunc.c b/src/backend/access/hash/hashfunc.c index 1e2d779a14c..1f9813277a3 100644 --- a/src/backend/access/hash/hashfunc.c +++ b/src/backend/access/hash/hashfunc.c @@ -8,7 +8,7 @@ * * * IDENTIFICATION - * $PostgreSQL: pgsql/src/backend/access/hash/hashfunc.c,v 1.48 2006/10/04 00:29:48 momjian Exp $ + * $PostgreSQL: pgsql/src/backend/access/hash/hashfunc.c,v 1.48.2.1 2007/06/01 15:58:01 tgl Exp $ * * NOTES * These functions are stored in pg_amproc. For each operator class @@ -267,6 +267,31 @@ hash_any(register const unsigned char *k, register int keylen) /* case 0: nothing left to add */ } mix(a, b, c); + + /* report the result */ + return UInt32GetDatum(c); +} + +/* + * hash_uint32() -- hash a 32-bit value + * + * This has the same result (at least on little-endian machines) as + * hash_any(&k, sizeof(uint32)) + * but is faster and doesn't force the caller to store k into memory. + */ +Datum +hash_uint32(uint32 k) +{ + register uint32 a, + b, + c; + + a = 0x9e3779b9 + k; + b = 0x9e3779b9; + c = 3923095 + (uint32) sizeof(uint32); + + mix(a, b, c); + /* report the result */ return UInt32GetDatum(c); } diff --git a/src/backend/executor/nodeHash.c b/src/backend/executor/nodeHash.c index 171cfb58f28..b36832b8384 100644 --- a/src/backend/executor/nodeHash.c +++ b/src/backend/executor/nodeHash.c @@ -8,7 +8,7 @@ * * * IDENTIFICATION - * $PostgreSQL: pgsql/src/backend/executor/nodeHash.c,v 1.107 2006/07/14 04:44:46 momjian Exp $ + * $PostgreSQL: pgsql/src/backend/executor/nodeHash.c,v 1.107.2.1 2007/06/01 15:58:01 tgl Exp $ * *------------------------------------------------------------------------- */ @@ -24,6 +24,7 @@ #include #include +#include "access/hash.h" #include "executor/execdebug.h" #include "executor/hashjoin.h" #include "executor/instrument.h" @@ -719,9 +720,11 @@ ExecHashGetHashValue(HashJoinTable hashtable, * chains), and must only cause the batch number to remain the same or * increase. Our algorithm is * bucketno = hashvalue MOD nbuckets - * batchno = (hashvalue DIV nbuckets) MOD nbatch - * where nbuckets should preferably be prime so that all bits of the - * hash value can affect both bucketno and batchno. + * batchno = hash_uint32(hashvalue) MOD nbatch + * which gives reasonably independent bucket and batch numbers in the face + * of some rather poorly-implemented hash functions in hashfunc.c. (This + * will change in PG 8.3.) + * * nbuckets doesn't change over the course of the join. * * nbatch is always a power of 2; we increase it only by doubling it. This @@ -740,7 +743,7 @@ ExecHashGetBucketAndBatch(HashJoinTable hashtable, { *bucketno = hashvalue % nbuckets; /* since nbatch is a power of 2, can do MOD by masking */ - *batchno = (hashvalue / nbuckets) & (nbatch - 1); + *batchno = hash_uint32(hashvalue) & (nbatch - 1); } else { diff --git a/src/include/access/hash.h b/src/include/access/hash.h index 016e51d09f9..4b8ce788928 100644 --- a/src/include/access/hash.h +++ b/src/include/access/hash.h @@ -7,7 +7,7 @@ * Portions Copyright (c) 1996-2006, PostgreSQL Global Development Group * Portions Copyright (c) 1994, Regents of the University of California * - * $PostgreSQL: pgsql/src/include/access/hash.h,v 1.73.2.1 2007/04/19 20:24:10 tgl Exp $ + * $PostgreSQL: pgsql/src/include/access/hash.h,v 1.73.2.2 2007/06/01 15:58:02 tgl Exp $ * * NOTES * modeled after Margo Seltzer's hash implementation for unix. @@ -262,6 +262,7 @@ extern Datum hashname(PG_FUNCTION_ARGS); extern Datum hashtext(PG_FUNCTION_ARGS); extern Datum hashvarlena(PG_FUNCTION_ARGS); extern Datum hash_any(register const unsigned char *k, register int keylen); +extern Datum hash_uint32(uint32 k); /* private routines */