database/sqlite
1
0
Fork 0
mirror of https://github.com/sqlite/sqlite.git synced 2025-11-18 10:21:03 +03:00
Code Activity

Deploy heuristics (well-commented) to better estimate how much unindexed terms

Browse Source 
in the WHERE clause filter the number of output rows from a single table.

FossilOrigin-Name: 221659945c3f78d3b6789bfe8fdeb8d3ee1fa038
This commit is contained in:
drh
2014-11-22 18:50:44 +00:00
parent a3898250d4
commit 7a1bca7e74
4 changed files with 48 additions and 22 deletions
Download Patch File Download Diff File Expand all files Collapse all files

52
src/where.c
View File

@@ -4291,10 +4291,30 @@ static int whereLoopInsert(WhereLoopBuilder *pBuilder, WhereLoop *pTemplate){
** Adjust the WhereLoop.nOut value downward to account for terms of the
** WHERE clause that reference the loop but which are not used by an
** index.
*
** For every WHERE clause term that is not used by the index
** and which has a truth probability assigned by one of the likelihood(),
** likely(), or unlikely() SQL functions, reduce the estimated number
** of output rows by the probability specified.
**
** In the current implementation, the first extra WHERE clause term reduces
** the number of output rows by a factor of 10 and each additional term
** reduces the number of output rows by sqrt(2).
** TUNING: For every WHERE clause term that is not used by the index
** and which does not have an assigned truth probability, heuristics
** described below are used to try to estimate the truth probability.
** TODO --> Perhaps this is something that could be improved by better
** table statistics.
**
** Heuristic 1: Estimate the truth probability as 6.25%. The 6.25%
** value corresponds to 1 in LogEst notation, so this means decrement
** the WhereLoop.nOut field for every such WHERE clause term.
**
** Heuristic 2: If there exists one or more WHERE clause terms of the
** form "x==EXPR" and EXPR is not a constant 0 or 1, then make sure the
** final output row estimate is no greater than 1/4 of the total number
** of rows in the table. In other words, assume that x==EXPR will filter
** out at least 3 out of 4 rows. If EXPR is -1 or 0 or 1, then maybe the
** "x" column is boolean or else -1 or 0 or 1 is a common default value
** on the "x" column and so in that case only cap the output row estimate
** at 1/2 instead of 1/4.
*/
static void whereLoopOutputAdjust(
WhereClause *pWC, /* The WHERE clause */
@@ -4303,8 +4323,8 @@ static void whereLoopOutputAdjust(
){
WhereTerm *pTerm, *pX;
Bitmask notAllowed = ~(pLoop->prereq|pLoop->maskSelf);
int i, j;
int nEq = 0; /* Number of = constraints not within likely()/unlikely() */
int i, j, k;
LogEst iReduce = 0; /* pLoop->nOut should not exceed nRow-iReduce */
assert( (pLoop->wsFlags & WHERE_AUTO_INDEX)==0 );
for(i=pWC->nTerm, pTerm=pWC->a; i>0; i--, pTerm++){
@@ -4319,20 +4339,26 @@ static void whereLoopOutputAdjust(
}
if( j<0 ){
if( pTerm->truthProb<=0 ){
/* If a truth probability is specified using the likelihood() hints,
** then use the probability provided by the application. */
pLoop->nOut += pTerm->truthProb;
}else{
/* In the absence of explicit truth probabilities, use heuristics to
** guess a reasonable truth probability. */
pLoop->nOut--;
if( pTerm->eOperator&WO_EQ ) nEq++;
if( pTerm->eOperator&WO_EQ ){
Expr *pRight = pTerm->pExpr->pRight;
if( sqlite3ExprIsInteger(pRight, &k) && k>=(-1) && k<=1 ){
k = 10;
}else{
k = 20;
}
if( iReduce<k ) iReduce = k;
}
}
}
}
/* TUNING: If there is at least one equality constraint in the WHERE
** clause that does not have a likelihood() explicitly assigned to it
** then do not let the estimated number of output rows exceed half
** the number of rows in the table. */
if( nEq && pLoop->nOut>nRow-10 ){
pLoop->nOut = nRow - 10;
}
if( pLoop->nOut > nRow-iReduce ) pLoop->nOut = nRow - iReduce;
}
/*