database/sqlite
1
0
Fork 0
mirror of https://github.com/sqlite/sqlite.git synced 2025-08-05 15:55:57 +03:00
Code Activity

Enhance the b-tree page sorting code to ensure that sqlite3PagerRekey() never

Browse Source 
overloads a page number and uses only the PENDING_BYTE page for temporary
storage.

FossilOrigin-Name: 5007742886bd20de20be3973737cf46b010359911615eb3da69cd262bd9a2435
This commit is contained in:
drh
2022-08-31 15:04:42 +00:00
parent 78d15f097a
commit 9c3a114ca0
4 changed files with 38 additions and 41 deletions
Download Patch File Download Diff File Expand all files Collapse all files

58
src/btree.c
View File

@@ -7893,8 +7893,6 @@ static int balance_nonroot(
Pgno pgno; /* Temp var to store a page number in */
u8 abDone[NB+2]; /* True after i'th new page is populated */
Pgno aPgno[NB+2]; /* Page numbers of new pages before shuffling */
Pgno aPgOrder[NB+2]; /* Copy of aPgno[] used for sorting pages */
u16 aPgFlags[NB+2]; /* flags field of new pages before shuffling */
CellArray b; /* Parsed information on cells being balanced */
memset(abDone, 0, sizeof(abDone));
@@ -8318,43 +8316,39 @@ static int balance_nonroot(
** of the table is closer to a linear scan through the file. That in turn
** helps the operating system to deliver pages from the disk more rapidly.
**
** An O(n^2) insertion sort algorithm is used, but since n is never more
** than (NB+2) (a small constant), that should not be a problem.
** An O(N*N) sort algorithm is used, but since N is never more than NB+2
** (5), that is not a performance concern.
**
** When NB==3, this one optimization makes the database about 25% faster
** for large insertions and deletions.
*/
for(i=0; i<nNew; i++){
aPgOrder[i] = aPgno[i] = apNew[i]->pgno;
aPgFlags[i] = apNew[i]->pDbPage->flags;
for(j=0; j<i; j++){
if( NEVER(aPgno[j]==aPgno[i]) ){
/* This branch is taken if the set of sibling pages somehow contains
** duplicate entries. This can happen if the database is corrupt.
** It would be simpler to detect this as part of the loop below, but
** we do the detection here in order to avoid populating the pager
** cache with two separate objects associated with the same
** page number. */
assert( CORRUPT_DB );
rc = SQLITE_CORRUPT_BKPT;
goto balance_cleanup;
}
}
aPgno[i] = apNew[i]->pgno;
assert( apNew[i]->pDbPage->flags & PGHDR_WRITEABLE );
assert( apNew[i]->pDbPage->flags & PGHDR_DIRTY );
}
for(i=0; i<nNew; i++){
int iBest = 0; /* aPgno[] index of page number to use */
for(j=1; j<nNew; j++){
if( aPgOrder[j]<aPgOrder[iBest] ) iBest = j;
for(i=0; i<nNew-1; i++){
int iB = i;
for(j=i+1; j<nNew; j++){
if( apNew[j]->pgno < apNew[iB]->pgno ) iB = j;
}
pgno = aPgOrder[iBest];
aPgOrder[iBest] = 0xffffffff;
if( iBest!=i ){
if( iBest>i ){
sqlite3PagerRekey(apNew[iBest]->pDbPage, pBt->nPage+iBest+1,
aPgFlags[iBest]);
}
sqlite3PagerRekey(apNew[i]->pDbPage, pgno, aPgFlags[iBest]);
apNew[i]->pgno = pgno;
/* If apNew[i] has a page number that is bigger than any of the
** subsequence apNew[i] entries, then swap apNew[i] with the subsequent
** entry that has the smallest page number (which we know to be
** entry apNew[iB]).
*/
if( iB!=i ){
Pgno pgnoA = apNew[i]->pgno;
Pgno pgnoB = apNew[iB]->pgno;
Pgno pgnoTemp = (PENDING_BYTE/pBt->pageSize)+1;
u16 fgA = apNew[i]->pDbPage->flags;
u16 fgB = apNew[iB]->pDbPage->flags;
sqlite3PagerRekey(apNew[i]->pDbPage, pgnoTemp, fgB);
sqlite3PagerRekey(apNew[iB]->pDbPage, pgnoA, fgA);
sqlite3PagerRekey(apNew[i]->pDbPage, pgnoB, fgB);
apNew[i]->pgno = pgnoB;
apNew[iB]->pgno = pgnoA;
}
}