Movatterモバイル変換

[0]ホーム
URL:

Skip to content

Navigation Menu

Sign in
Appearance settings

Search code, repositories, users, issues, pull requests...

Provide feedback

We read every piece of feedback, and take your input very seriously.

Saved searches

Use saved searches to filter your results more quickly

 Sign up
Appearance settings

Commitce2cee0

Browse files
Fix nbtree kill_prior_tuple posting list assert.
An assertion added by commit0d861bb checked that _bt_killitems() onlyprocesses a BTScanPosItem whose heap TID is contained in a posting listtuple when its page offset number still matches what is on the page(i.e. when it matches the posting list tuple's current offset number).This was only correct in the common case where the page can't havechanged since we first read it. It was not correct in cases where wedon't drop the buffer pin (and don't need to verify the page hasn'tchanged using its LSN). The latter category includes scans involvingunlogged tables, and scans that use a non-MVCC snapshot, per the logicoriginally introduced by commit2ed5b87.The assertion still seems helpful. Fix it by taking cases where thepage may have been concurrently modified into account.Reported-By: Anastasia Lubennikova, Alexander LakhinDiscussion:https://postgr.es/m/c4e38e9a-0f9c-8e53-e639-adf343f94472@postgrespro.ru
1 parent7d6d82a commitce2cee0

File tree

1 file changed

+21
-5
lines changed

1 file changed

+21
-5
lines changed

‎src/backend/access/nbtree/nbtutils.c

Lines changed: 21 additions & 5 deletions
Original file line numberDiff line numberDiff line change
@@ -1725,6 +1725,7 @@ _bt_killitems(IndexScanDesc scan)
17251725
inti;
17261726
intnumKilled=so->numKilled;
17271727
boolkilledsomething= false;
1728+
booldroppedpinPG_USED_FOR_ASSERTS_ONLY;
17281729

17291730
Assert(BTScanPosIsValid(so->currPos));
17301731

@@ -1742,6 +1743,7 @@ _bt_killitems(IndexScanDesc scan)
17421743
* re-use of any TID on the page, so there is no need to check the
17431744
* LSN.
17441745
*/
1746+
droppedpin= false;
17451747
LockBuffer(so->currPos.buf,BT_READ);
17461748

17471749
page=BufferGetPage(so->currPos.buf);
@@ -1750,6 +1752,7 @@ _bt_killitems(IndexScanDesc scan)
17501752
{
17511753
Bufferbuf;
17521754

1755+
droppedpin= true;
17531756
/* Attempt to re-read the buffer, getting pin and lock. */
17541757
buf=_bt_getbuf(scan->indexRelation,so->currPos.currPage,BT_READ);
17551758

@@ -1795,9 +1798,18 @@ _bt_killitems(IndexScanDesc scan)
17951798
intj;
17961799

17971800
/*
1798-
* Note that we rely on the assumption that heap TIDs in the
1799-
* scanpos items array are always in ascending heap TID order
1800-
* within a posting list
1801+
* We rely on the convention that heap TIDs in the scanpos
1802+
* items array are stored in ascending heap TID order for a
1803+
* group of TIDs that originally came from a posting list
1804+
* tuple. This convention even applies during backwards
1805+
* scans, where returning the TIDs in descending order might
1806+
* seem more natural. This is about effectiveness, not
1807+
* correctness.
1808+
*
1809+
* Note that the page may have been modified in almost any way
1810+
* since we first read it (in the !droppedpin case), so it's
1811+
* possible that this posting list tuple wasn't a posting list
1812+
* tuple when we first encountered its heap TIDs.
18011813
*/
18021814
for (j=0;j<nposting;j++)
18031815
{
@@ -1806,8 +1818,12 @@ _bt_killitems(IndexScanDesc scan)
18061818
if (!ItemPointerEquals(item,&kitem->heapTid))
18071819
break;/* out of posting list loop */
18081820

1809-
/* kitem must have matching offnum when heap TIDs match */
1810-
Assert(kitem->indexOffset==offnum);
1821+
/*
1822+
* kitem must have matching offnum when heap TIDs match,
1823+
* though only in the common case where the page can't
1824+
* have been concurrently modified
1825+
*/
1826+
Assert(kitem->indexOffset==offnum|| !droppedpin);
18111827

18121828
/*
18131829
* Read-ahead to later kitems here.

0 commit comments

Comments
 (0)
[8]ページ先頭
©2009-2025 Movatter.jp