mirrors/postgres
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

Fix thinko in hash cost estimation: average frequency

Browse Source 
should be computed from total number of distinct values in whole
relation, not # distinct values we expect to have after restriction
clauses are applied.
This commit is contained in:
Tom Lane 2001-06-10 02:59:35 +00:00
parent 26c94c5d9c
commit a8fe109ac1
1 changed files with 6 additions and 16 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

22
src/backend/optimizer/path/costsize.c
View File

@ -42,7 +42,7 @@
* Portions Copyright (c) 1994, Regents of the University of California * Portions Copyright (c) 1994, Regents of the University of California
* *
* IDENTIFICATION * IDENTIFICATION
* $Header: /cvsroot/pgsql/src/backend/optimizer/path/costsize.c,v 1.75 2001/06/05 05:26:04 tgl Exp $ * $Header: /cvsroot/pgsql/src/backend/optimizer/path/costsize.c,v 1.76 2001/06/10 02:59:35 tgl Exp $
* *
*------------------------------------------------------------------------- *-------------------------------------------------------------------------
*/ */
@ -873,6 +873,9 @@ estimate_hash_bucketsize(Query *root, Var *var)
if (ndistinct < 0.0) if (ndistinct < 0.0)
ndistinct = -ndistinct * rel->tuples; ndistinct = -ndistinct * rel->tuples;
/* Also compute avg freq of all distinct data values in raw relation */
avgfreq = (1.0 - stats->stanullfrac) / ndistinct;
/* /*
* Adjust ndistinct to account for restriction clauses. Observe we are * Adjust ndistinct to account for restriction clauses. Observe we are
* assuming that the data distribution is affected uniformly by the * assuming that the data distribution is affected uniformly by the
@ -883,17 +886,6 @@ estimate_hash_bucketsize(Query *root, Var *var)
*/ */
ndistinct *= rel->rows / rel->tuples; ndistinct *= rel->rows / rel->tuples;
/*
* Discourage use of hash join if there seem not to be very many distinct
* data values. The threshold here is somewhat arbitrary, as is the
* fraction used to "discourage" the choice.
*/
if (ndistinct < 50.0)
{
ReleaseSysCache(tuple);
return 0.5;
}
/* /*
* Form initial estimate of bucketsize fraction. Here we use rel->rows, * Form initial estimate of bucketsize fraction. Here we use rel->rows,
* ie the number of rows after applying restriction clauses, because * ie the number of rows after applying restriction clauses, because
@ -903,8 +895,8 @@ estimate_hash_bucketsize(Query *root, Var *var)
estfract = (double) NTUP_PER_BUCKET / rel->rows; estfract = (double) NTUP_PER_BUCKET / rel->rows;
/* /*
* Adjust estimated bucketsize if too few distinct values to fill * Adjust estimated bucketsize if too few distinct values (after
* all the buckets. * restriction clauses) to fill all the buckets.
*/ */
needdistinct = rel->rows / (double) NTUP_PER_BUCKET; needdistinct = rel->rows / (double) NTUP_PER_BUCKET;
if (ndistinct < needdistinct) if (ndistinct < needdistinct)
@ -931,8 +923,6 @@ estimate_hash_bucketsize(Query *root, Var *var)
/* /*
* Adjust estimated bucketsize upward to account for skewed distribution. * Adjust estimated bucketsize upward to account for skewed distribution.
*/ */
avgfreq = (1.0 - stats->stanullfrac) / ndistinct;
if (avgfreq > 0.0 && mcvfreq > avgfreq) if (avgfreq > 0.0 && mcvfreq > avgfreq)
estfract *= mcvfreq / avgfreq; estfract *= mcvfreq / avgfreq;