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

Commit3a01f68

Browse files
committed
Check for interrupts inside the nbtree page deletion code.
When deleting pages the nbtree code has to walk through siblings of atree node. When those sibling links are corrupted that can lead toendless loops - which are currently not interruptible. This isespecially problematic if autovacuum is repeatedly blocked on suchindexes, as it can be hard to get out of that situation withoutresorting to single user mode.Thus add interrupt checks to appropriate places in suchloops. Unfortunately in one of the cases it's it's not easy to do so.Between 9.3 and 9.4 the page deletion (and page split) code changedsignificantly. Before it was significantly less robust againstinterruptions. Therefore don't backpatch to 9.3.Author: Andres FreundDiscussion:https://postgr.es/m/20180627191629.wkunw2qbibnvlz53@alap3.anarazel.deBackpatch: 9.4-
1 parentb416691 commit3a01f68

File tree

1 file changed

+21
-0
lines changed

1 file changed

+21
-0
lines changed

‎src/backend/access/nbtree/nbtpage.c

Lines changed: 21 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -1443,6 +1443,7 @@ _bt_pagedel(Relation rel, Buffer buf)
14431443
rightsib_empty= false;
14441444
while (P_ISHALFDEAD(opaque))
14451445
{
1446+
/* will check for interrupts, once lock is released */
14461447
if (!_bt_unlink_halfdead_page(rel,buf,&rightsib_empty))
14471448
{
14481449
/* _bt_unlink_halfdead_page already released buffer */
@@ -1455,6 +1456,12 @@ _bt_pagedel(Relation rel, Buffer buf)
14551456

14561457
_bt_relbuf(rel,buf);
14571458

1459+
/*
1460+
* Check here, as calling loops will have locks held, preventing
1461+
* interrupts from being processed.
1462+
*/
1463+
CHECK_FOR_INTERRUPTS();
1464+
14581465
/*
14591466
* The page has now been deleted. If its right sibling is completely
14601467
* empty, it's possible that the reason we haven't deleted it earlier
@@ -1707,6 +1714,12 @@ _bt_unlink_halfdead_page(Relation rel, Buffer leafbuf, bool *rightsib_empty)
17071714

17081715
LockBuffer(leafbuf,BUFFER_LOCK_UNLOCK);
17091716

1717+
/*
1718+
* Check here, as calling loops will have locks held, preventing
1719+
* interrupts from being processed.
1720+
*/
1721+
CHECK_FOR_INTERRUPTS();
1722+
17101723
/*
17111724
* If the leaf page still has a parent pointing to it (or a chain of
17121725
* parents), we don't unlink the leaf page yet, but the topmost remaining
@@ -1766,6 +1779,14 @@ _bt_unlink_halfdead_page(Relation rel, Buffer leafbuf, bool *rightsib_empty)
17661779
/* step right one page */
17671780
leftsib=opaque->btpo_next;
17681781
_bt_relbuf(rel,lbuf);
1782+
1783+
/*
1784+
* It'd be good to check for interrupts here, but it's not easy to
1785+
* do so because a lock is always held. This block isn't
1786+
* frequently reached, so hopefully the consequences of not
1787+
* checking interrupts aren't too bad.
1788+
*/
1789+
17691790
if (leftsib==P_NONE)
17701791
{
17711792
elog(LOG,"no left sibling (concurrent deletion?) of block %u in \"%s\"",

0 commit comments

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