database/sqlite
1
0
Fork 0
mirror of https://github.com/sqlite/sqlite.git synced 2025-11-16 23:02:26 +03:00
Code Activity

Consider doing a partial table scan to fulfill an IN operator rather

Browse Source 
than using an index.  Try to pick the plan with the lowest cost.

FossilOrigin-Name: 1fa40a78fef4516c39b217bff67efe7e7d2077cca00aae0ef5c2c9cff94f008b
This commit is contained in:
drh
2018-06-08 18:22:10 +00:00
parent 83193d0133
commit da4c409aea
3 changed files with 41 additions and 9 deletions
Download Patch File Download Diff File Expand all files Collapse all files

32
src/where.c
View File

@@ -2451,7 +2451,7 @@ static int whereLoopAddBtreeIndex(
if( eOp & WO_IN ){
Expr *pExpr = pTerm->pExpr;
pNew->wsFlags |= WHERE_COLUMN_IN;
LogEst M, logK;
if( ExprHasProperty(pExpr, EP_xIsSelect) ){
/* "x IN (SELECT ...)": TUNING: the SELECT returns 25 rows */
int i;
@@ -2471,6 +2471,36 @@ static int whereLoopAddBtreeIndex(
assert( nIn>0 ); /* RHS always has 2 or more terms... The parser
** changes "x IN (?)" into "x=?". */
}
/* Let:
** N = the total number of rows in the table
** K = the number of entries on the right-hand side of the IN operator
** M = the number of rows in the table that match terms to the
** to the left in the same index. If the IN operator is on
** the left-most index column, M==N.
**
** Given the definitions above, it is better to omit the IN operator
** from the index lookup and instead do a scan of the M elements,
** testing each scanned row against the IN operator separately, if:
**
** M*log(K) < K*log(N)
**
** Our estimates for M, K, and N might be inaccurate, so we build in
** a safety margin of 2 (LogEst: 10) that favors using the IN operator
** with the index, as using an index has better worst-case behavior.
*/
M = pProbe->aiRowLogEst[saved_nEq+1];
logK = sqlite3LogEst(nIn);
if( M + logK + 10 < nIn + rLogSize ){
WHERETRACE(0x40,
("IN operator costs more than scan on column %d of \"%s\" (%d<%d)\n",
saved_nEq, pProbe->zName, M+logK+10, nIn+rLogSize));
continue;
}else{
WHERETRACE(0x40,
("IN operator cheaper than scan on column %d of \"%s\" (%d>=%d)\n",
saved_nEq, pProbe->zName, M+logK+10, nIn+rLogSize));
}
pNew->wsFlags |= WHERE_COLUMN_IN;
}else if( eOp & (WO_EQ|WO_IS) ){
int iCol = pProbe->aiColumn[saved_nEq];
pNew->wsFlags |= WHERE_COLUMN_EQ;