Skip to content

Commit c19dadb

Browse files
vadim4ovadim4o
committed
Runtime btree recovery is now ON by default.
1 parent 66decbf commit c19dadb

File tree

3 files changed

+55
-53
lines changed

3 files changed

+55
-53
lines changed

‎src/backend/access/nbtree/nbtinsert.c‎

Lines changed: 49 additions & 47 deletions
Original file line numberDiff line numberDiff line change
@@ -8,7 +8,7 @@
88
*
99
*
1010
* IDENTIFICATION
11-
* $Header: /cvsroot/pgsql/src/backend/access/nbtree/nbtinsert.c,v 1.80 2001/02/02 19:49:15 vadim Exp $
11+
* $Header: /cvsroot/pgsql/src/backend/access/nbtree/nbtinsert.c,v 1.81 2001/02/07 23:35:33 vadim Exp $
1212
*
1313
*-------------------------------------------------------------------------
1414
*/
@@ -494,12 +494,13 @@ _bt_insertonpg(Relation rel,
494494
* then old root' btpo_parent still points to metapage.
495495
* We have to fix root page in this case.
496496
*/
497-
if (lpageop->btpo_parent == BTREE_METAPAGE)
497+
if (BTreeInvalidParent(lpageop))
498498
{
499499
if (!FixBTree)
500-
elog(ERROR, "bt_insertonpg: no root page found");
500+
elog(ERROR, "bt_insertonpg[%s]: no root page found", RelationGetRelationName(rel));
501501
_bt_wrtbuf(rel, rbuf);
502502
_bt_wrtnorelbuf(rel, buf);
503+
elog(NOTICE, "bt_insertonpg[%s]: root page unfound - fixing upper levels", RelationGetRelationName(rel));
503504
_bt_fixup(rel, buf);
504505
goto formres;
505506
}
@@ -549,10 +550,10 @@ _bt_insertonpg(Relation rel,
549550
elog(ERROR, "_bt_getstackbuf: my bits moved right off the end of the world!"
550551
"\n\tRecreate index %s.", RelationGetRelationName(rel));
551552
pfree(new_item);
553+
elog(NOTICE, "bt_insertonpg[%s]: parent page unfound - fixing branch", RelationGetRelationName(rel));
552554
_bt_fixbranch(rel, bknum, rbknum, stack);
553555
goto formres;
554556
}
555-
556557
/* Recursively update the parent */
557558
newres = _bt_insertonpg(rel, pbuf, stack->bts_parent,
558559
0, NULL, new_item, stack->bts_offset);
@@ -1313,6 +1314,11 @@ _bt_newroot(Relation rel, Buffer lbuf, Buffer rbuf)
13131314
PageSetLSN(metapg, recptr);
13141315
PageSetSUI(metapg, ThisStartUpID);
13151316

1317+
/* we changed their btpo_parent */
1318+
PageSetLSN(lpage, recptr);
1319+
PageSetSUI(lpage, ThisStartUpID);
1320+
PageSetLSN(rpage, recptr);
1321+
PageSetSUI(rpage, ThisStartUpID);
13161322
}
13171323
END_CRIT_SECTION();
13181324

@@ -1359,22 +1365,6 @@ _bt_fixroot(Relation rel, Buffer oldrootbuf, bool release)
13591365
rootLSN = PageGetLSN(rootpage);
13601366
rootblk = BufferGetBlockNumber(rootbuf);
13611367

1362-
/*
1363-
* Update LSN & StartUpID of old root buffer and its neighbor to
1364-
* ensure that they will be written on disk after logging new
1365-
* root creation. Unfortunately, for the moment (?) we do not
1366-
* log this operation and so possibly break our rule to log entire
1367-
* page content of first after checkpoint modification.
1368-
*/
1369-
HOLD_INTERRUPTS();
1370-
oldrootopaque->btpo_parent = rootblk;
1371-
leftopaque->btpo_parent = rootblk;
1372-
PageSetLSN(oldrootpage, rootLSN);
1373-
PageSetSUI(oldrootpage, ThisStartUpID);
1374-
PageSetLSN(leftpage, rootLSN);
1375-
PageSetSUI(leftpage, ThisStartUpID);
1376-
RESUME_INTERRUPTS();
1377-
13781368
/* parent page where to insert pointers */
13791369
buf = rootbuf;
13801370
page = BufferGetPage(buf);
@@ -1401,7 +1391,13 @@ _bt_fixroot(Relation rel, Buffer oldrootbuf, bool release)
14011391
rightpage = BufferGetPage(rightbuf);
14021392
rightopaque = (BTPageOpaque) PageGetSpecialPointer(rightpage);
14031393

1404-
/* Update LSN & StartUpID (see comments above) */
1394+
/*
1395+
* Update LSN & StartUpID of child page buffer to ensure that
1396+
* it will be written on disk after flushing log record for new
1397+
* root creation. Unfortunately, for the moment (?) we do not
1398+
* log this operation and so possibly break our rule to log entire
1399+
* page content on first after checkpoint modification.
1400+
*/
14051401
HOLD_INTERRUPTS();
14061402
rightopaque->btpo_parent = rootblk;
14071403
if (XLByteLT(PageGetLSN(rightpage), rootLSN))
@@ -1442,15 +1438,15 @@ _bt_fixroot(Relation rel, Buffer oldrootbuf, bool release)
14421438
_bt_insertuple(rel, buf, itemsz, btitem, newitemoff);
14431439

14441440
/* give up left buffer */
1445-
_bt_relbuf(rel, leftbuf, BT_WRITE);
1441+
_bt_wrtbuf(rel, leftbuf);
14461442
pfree(btitem);
14471443
leftbuf = rightbuf;
14481444
leftpage = rightpage;
14491445
leftopaque = rightopaque;
14501446
}
14511447

14521448
/* give up rightmost page buffer */
1453-
_bt_relbuf(rel, leftbuf, BT_WRITE);
1449+
_bt_wrtbuf(rel, leftbuf);
14541450

14551451
/*
14561452
* Here we hold locks on old root buffer, new root buffer we've
@@ -1460,11 +1456,11 @@ _bt_fixroot(Relation rel, Buffer oldrootbuf, bool release)
14601456
* then we give up oldrootbuf.
14611457
*/
14621458
if (release)
1463-
_bt_relbuf(rel, oldrootbuf, BT_WRITE);
1459+
_bt_wrtbuf(rel, oldrootbuf);
14641460

14651461
if (rootbuf != buf)
14661462
{
1467-
_bt_relbuf(rel, buf, BT_WRITE);
1463+
_bt_wrtbuf(rel, buf);
14681464
return(_bt_fixroot(rel, rootbuf, true));
14691465
}
14701466

@@ -1483,15 +1479,13 @@ _bt_fixtree(Relation rel, BlockNumber blkno)
14831479
BTPageOpaque opaque;
14841480
BlockNumber pblkno;
14851481

1486-
elog(ERROR, "bt_fixtree: unimplemented , yet (need to recreate index)");
1487-
14881482
for ( ; ; )
14891483
{
14901484
buf = _bt_getbuf(rel, blkno, BT_READ);
14911485
page = BufferGetPage(buf);
14921486
opaque = (BTPageOpaque) PageGetSpecialPointer(page);
14931487
if (! P_LEFTMOST(opaque) || P_ISLEAF(opaque))
1494-
elog(ERROR, "bt_fixtree: invalid start page (need to recreate index)");
1488+
elog(ERROR, "bt_fixtree[%s]: invalid start page (need to recreate index)", RelationGetRelationName(rel));
14951489
pblkno = opaque->btpo_parent;
14961490

14971491
/* check/fix entire level */
@@ -1500,8 +1494,10 @@ _bt_fixtree(Relation rel, BlockNumber blkno)
15001494
/*
15011495
* No pins/locks are held here. Re-read start page if its
15021496
* btpo_parent pointed to meta page else go up one level.
1497+
*
1498+
* XXX have to catch InvalidBlockNumber at the moment -:(
15031499
*/
1504-
if (pblkno == BTREE_METAPAGE)
1500+
if (pblkno == BTREE_METAPAGE || pblkno == InvalidBlockNumber)
15051501
{
15061502
buf = _bt_getbuf(rel, blkno, BT_WRITE);
15071503
page = BufferGetPage(buf);
@@ -1512,18 +1508,19 @@ _bt_fixtree(Relation rel, BlockNumber blkno)
15121508
_bt_relbuf(rel, buf, BT_WRITE);
15131509
return;
15141510
}
1515-
pblkno = opaque->btpo_parent;
15161511
/* Call _bt_fixroot() if there is no upper level */
1517-
if (pblkno == BTREE_METAPAGE)
1512+
if (BTreeInvalidParent(opaque))
15181513
{
1514+
elog(NOTICE, "bt_fixtree[%s]: fixing root page", RelationGetRelationName(rel));
15191515
buf = _bt_fixroot(rel, buf, true);
15201516
_bt_relbuf(rel, buf, BT_WRITE);
15211517
return;
15221518
}
15231519
/* Have to go up one level */
1520+
pblkno = opaque->btpo_parent;
15241521
_bt_relbuf(rel, buf, BT_WRITE);
1525-
blkno = pblkno;
15261522
}
1523+
blkno = pblkno;
15271524
}
15281525

15291526
}
@@ -1561,17 +1558,17 @@ _bt_fixlevel(Relation rel, Buffer buf, BlockNumber limit)
15611558
/* Initialize first child data */
15621559
coff[0] = P_FIRSTDATAKEY(opaque);
15631560
if (coff[0] > PageGetMaxOffsetNumber(page))
1564-
elog(ERROR, "bt_fixlevel: invalid maxoff on start page (need to recreate index)");
1561+
elog(ERROR, "bt_fixlevel[%s]: invalid maxoff on start page (need to recreate index)", RelationGetRelationName(rel));
15651562
btitem = (BTItem) PageGetItem(page, PageGetItemId(page, coff[0]));
15661563
cblkno[0] = ItemPointerGetBlockNumber(&(btitem->bti_itup.t_tid));
15671564
cbuf[0] = _bt_getbuf(rel, cblkno[0], BT_READ);
15681565
cpage[0] = BufferGetPage(cbuf[0]);
15691566
copaque[0] = (BTPageOpaque) PageGetSpecialPointer(cpage[0]);
15701567
if (P_LEFTMOST(opaque) && ! P_LEFTMOST(copaque[0]))
1571-
elog(ERROR, "bt_fixtlevel: non-leftmost child page of leftmost parent (need to recreate index)");
1568+
elog(ERROR, "bt_fixtlevel[%s]: non-leftmost child page of leftmost parent (need to recreate index)", RelationGetRelationName(rel));
15721569
/* caller should take care and avoid this */
15731570
if (P_RIGHTMOST(copaque[0]))
1574-
elog(ERROR, "bt_fixtlevel: invalid start child (need to recreate index)");
1571+
elog(ERROR, "bt_fixtlevel[%s]: invalid start child (need to recreate index)", RelationGetRelationName(rel));
15751572

15761573
for ( ; ; )
15771574
{
@@ -1597,7 +1594,7 @@ _bt_fixlevel(Relation rel, Buffer buf, BlockNumber limit)
15971594
if (coff[i] == InvalidOffsetNumber)
15981595
continue;
15991596
if (coff[cidx] != coff[i] + 1)
1600-
elog(ERROR, "bt_fixlevel: invalid item order(1) (need to recreate index)");
1597+
elog(ERROR, "bt_fixlevel[%s]: invalid item order(1) (need to recreate index)", RelationGetRelationName(rel));
16011598
break;
16021599
}
16031600
}
@@ -1629,7 +1626,7 @@ _bt_fixlevel(Relation rel, Buffer buf, BlockNumber limit)
16291626

16301627
buf = _bt_getstackbuf(rel, &stack, BT_WRITE);
16311628
if (buf == InvalidBuffer)
1632-
elog(ERROR, "bt_fixlevel: pointer disappeared (need to recreate index)");
1629+
elog(ERROR, "bt_fixlevel[%s]: pointer disappeared (need to recreate index)", RelationGetRelationName(rel));
16331630

16341631
page = BufferGetPage(buf);
16351632
opaque = (BTPageOpaque) PageGetSpecialPointer(page);
@@ -1648,7 +1645,7 @@ _bt_fixlevel(Relation rel, Buffer buf, BlockNumber limit)
16481645
{
16491646
if (parblk[i] == parblk[i - 1] &&
16501647
coff[i] != coff[i - 1] + 1)
1651-
elog(ERROR, "bt_fixlevel: invalid item order(2) (need to recreate index)");
1648+
elog(ERROR, "bt_fixlevel[%s]: invalid item order(2) (need to recreate index)", RelationGetRelationName(rel));
16521649
continue;
16531650
}
16541651
/* Have to check next page ? */
@@ -1662,7 +1659,7 @@ _bt_fixlevel(Relation rel, Buffer buf, BlockNumber limit)
16621659
if (coff[i] != InvalidOffsetNumber) /* found ! */
16631660
{
16641661
if (coff[i] != P_FIRSTDATAKEY(newopaque))
1665-
elog(ERROR, "bt_fixlevel: invalid item order(3) (need to recreate index)");
1662+
elog(ERROR, "bt_fixlevel[%s]: invalid item order(3) (need to recreate index)", RelationGetRelationName(rel));
16661663
_bt_relbuf(rel, buf, BT_WRITE);
16671664
buf = newbuf;
16681665
page = newpage;
@@ -1791,28 +1788,31 @@ _bt_fixbranch(Relation rel, BlockNumber lblkno,
17911788
ItemPointerSet(&(stack.bts_btitem.bti_itup.t_tid), lblkno, P_HIKEY);
17921789
buf = _bt_getstackbuf(rel, &stack, BT_READ);
17931790
if (buf == InvalidBuffer)
1794-
elog(ERROR, "bt_fixbranch: left pointer unfound (need to recreate index)");
1791+
elog(ERROR, "bt_fixbranch[%s]: left pointer unfound (need to recreate index)", RelationGetRelationName(rel));
17951792
page = BufferGetPage(buf);
17961793
offnum = _bt_getoff(page, rblkno);
17971794

17981795
if (offnum != InvalidOffsetNumber) /* right pointer found */
17991796
{
18001797
if (offnum <= stack.bts_offset)
1801-
elog(ERROR, "bt_fixbranch: invalid item order (need to recreate index)");
1798+
elog(ERROR, "bt_fixbranch[%s]: invalid item order (need to recreate index)", RelationGetRelationName(rel));
18021799
_bt_relbuf(rel, buf, BT_READ);
18031800
return;
18041801
}
18051802

18061803
/* Pointers are on different parent pages - find right one */
18071804
lblkno = BufferGetBlockNumber(buf);
1805+
opaque = (BTPageOpaque) PageGetSpecialPointer(page);
1806+
if (P_RIGHTMOST(opaque))
1807+
elog(ERROR, "bt_fixbranch[%s]: right pointer unfound(1) (need to recreate index)", RelationGetRelationName(rel));
18081808

18091809
stack.bts_parent = NULL;
1810-
stack.bts_blkno = lblkno;
1810+
stack.bts_blkno = opaque->btpo_next;
18111811
stack.bts_offset = InvalidOffsetNumber;
18121812
ItemPointerSet(&(stack.bts_btitem.bti_itup.t_tid), rblkno, P_HIKEY);
18131813
rbuf = _bt_getstackbuf(rel, &stack, BT_READ);
18141814
if (rbuf == InvalidBuffer)
1815-
elog(ERROR, "bt_fixbranch: right pointer unfound (need to recreate index)");
1815+
elog(ERROR, "bt_fixbranch[%s]: right pointer unfound(2) (need to recreate index)", RelationGetRelationName(rel));
18161816
rblkno = BufferGetBlockNumber(rbuf);
18171817
_bt_relbuf(rel, rbuf, BT_READ);
18181818

@@ -1834,8 +1834,7 @@ _bt_fixbranch(Relation rel, BlockNumber lblkno,
18341834
* then we'll use it to continue, else we'll fix/restore upper
18351835
* levels entirely.
18361836
*/
1837-
opaque = (BTPageOpaque) PageGetSpecialPointer(page);
1838-
if (opaque->btpo_parent != BTREE_METAPAGE)
1837+
if (!BTreeInvalidParent(opaque))
18391838
{
18401839
blkno = opaque->btpo_parent;
18411840
_bt_relbuf(rel, buf, BT_READ);
@@ -1847,7 +1846,7 @@ _bt_fixbranch(Relation rel, BlockNumber lblkno,
18471846
buf = _bt_getbuf(rel, blkno, BT_WRITE);
18481847
page = BufferGetPage(buf);
18491848
opaque = (BTPageOpaque) PageGetSpecialPointer(page);
1850-
if (opaque->btpo_parent != BTREE_METAPAGE)
1849+
if (!BTreeInvalidParent(opaque))
18511850
{
18521851
blkno = opaque->btpo_parent;
18531852
_bt_relbuf(rel, buf, BT_WRITE);
@@ -1861,6 +1860,7 @@ _bt_fixbranch(Relation rel, BlockNumber lblkno,
18611860
break;
18621861
}
18631862

1863+
elog(NOTICE, "bt_fixbranch[%s]: fixing upper levels", RelationGetRelationName(rel));
18641864
_bt_fixup(rel, buf);
18651865

18661866
return;
@@ -1887,10 +1887,11 @@ _bt_fixup(Relation rel, Buffer buf)
18871887
* then it's time for _bt_fixtree() to check upper
18881888
* levels and fix them, if required.
18891889
*/
1890-
if (opaque->btpo_parent != BTREE_METAPAGE)
1890+
if (!BTreeInvalidParent(opaque))
18911891
{
18921892
blkno = opaque->btpo_parent;
18931893
_bt_relbuf(rel, buf, BT_WRITE);
1894+
elog(NOTICE, "bt_fixup[%s]: checking/fixing upper levels", RelationGetRelationName(rel));
18941895
_bt_fixtree(rel, blkno);
18951896
return;
18961897
}
@@ -1907,6 +1908,7 @@ _bt_fixup(Relation rel, Buffer buf)
19071908
* by us and its btpo_parent points to meta page - time
19081909
* for _bt_fixroot().
19091910
*/
1911+
elog(NOTICE, "bt_fixup[%s]: fixing root page", RelationGetRelationName(rel));
19101912
buf = _bt_fixroot(rel, buf, true);
19111913
_bt_relbuf(rel, buf, BT_WRITE);
19121914

‎src/backend/access/nbtree/nbtpage.c‎

Lines changed: 4 additions & 4 deletions
Original file line numberDiff line numberDiff line change
@@ -9,7 +9,7 @@
99
*
1010
*
1111
* IDENTIFICATION
12-
* $Header: /cvsroot/pgsql/src/backend/access/nbtree/nbtpage.c,v 1.49 2001/01/29 07:28:17 vadim Exp $
12+
* $Header: /cvsroot/pgsql/src/backend/access/nbtree/nbtpage.c,v 1.50 2001/02/07 23:35:33 vadim Exp $
1313
*
1414
* NOTES
1515
* Postgres btree pages look like ordinary relation pages. The opaque
@@ -249,15 +249,15 @@ _bt_getroot(Relation rel, int access)
249249
Buffer newrootbuf;
250250

251251
check_parent:;
252-
if (rootopaque->btpo_parent == BTREE_METAPAGE) /* unupdated! */
252+
if (BTreeInvalidParent(rootopaque)) /* unupdated! */
253253
{
254254
LockBuffer(rootbuf, BUFFER_LOCK_UNLOCK);
255255
LockBuffer(rootbuf, BT_WRITE);
256256

257257
/* handle concurrent fix of root page */
258-
if (rootopaque->btpo_parent == BTREE_METAPAGE) /* unupdated! */
258+
if (BTreeInvalidParent(rootopaque)) /* unupdated! */
259259
{
260-
elog(NOTICE, "bt_getroot: fixing root page");
260+
elog(NOTICE, "bt_getroot[%s]: fixing root page", RelationGetRelationName(rel));
261261
newrootbuf = _bt_fixroot(rel, rootbuf, true);
262262
LockBuffer(newrootbuf, BUFFER_LOCK_UNLOCK);
263263
LockBuffer(newrootbuf, BT_READ);

‎src/backend/access/nbtree/nbtree.c‎

Lines changed: 2 additions & 2 deletions
Original file line numberDiff line numberDiff line change
@@ -12,7 +12,7 @@
1212
* Portions Copyright (c) 1994, Regents of the University of California
1313
*
1414
* IDENTIFICATION
15-
* $Header: /cvsroot/pgsql/src/backend/access/nbtree/nbtree.c,v 1.77 2001/01/29 00:39:14 tgl Exp $
15+
* $Header: /cvsroot/pgsql/src/backend/access/nbtree/nbtree.c,v 1.78 2001/02/07 23:35:33 vadim Exp $
1616
*
1717
*-------------------------------------------------------------------------
1818
*/
@@ -37,7 +37,7 @@ bool FastBuild = true; /* use sort/build instead */
3737
* TEMPORARY FLAG FOR TESTING NEW FIX TREE
3838
* CODE WITHOUT AFFECTING ANYONE ELSE
3939
*/
40-
bool FixBTree = false;
40+
bool FixBTree = true;
4141

4242
static void _bt_restscan(IndexScanDesc scan);
4343

0 commit comments

Comments
 (0)