Details
-
Type:
Bug
-
Status: Resolved
-
Priority:
Major - P3
-
Resolution: Fixed
-
Affects Version/s: None
-
Fix Version/s: WT2.7.0
-
Labels:None
Description
The page-split cleanup needs review and cleanup, there are a few different obvious problems:
- wt_split_rewrite isn't discarding the newly created page on error
- split_multi_inmem discards the replacement disk image which means on failure, the reconciliation state is corrupted
- wt_split_insert doesn't get the accounting correct if the page-split fails.
This is exacerbated by the new code that relies on split_parent locking the page (using the reconciliation lock), and returning EBUSY if the lock isn't available and the file is being checkpointed, to avoid deadlock. This makes it likely that a busy system can trigger one of these failures.
Issue Links
- is depended on by
-
SERVER-20408
WiredTiger changes for MongoDB 3.1.9
-
- Closed
-
-
SERVER-20479
WiredTiger changes for MongoDB 3.0.7
-
- Closed
-
-
SERVER-20771
WiredTiger changes for MongoDB 3.2.0-rc0
-
- Closed
-
- links to
Activity
Author:
{u'username': u'michaelcahill', u'name': u'Michael Cahill', u'email': u'michael.cahill@mongodb.com'}Message: Merge pull request #2249 from wiredtiger/wt-2126-panic-on-deepen
WT-2126: __split_deepen() doesn't really need to panic
Branch: develop
https://github.com/wiredtiger/wiredtiger/commit/0766b6d7251310f6a0d8a98a191d66c6b8618e65
Author:
{u'username': u'keithbostic', u'name': u'Keith Bostic', u'email': u'keith@wiredtiger.com'}Message: WT-2126: panic if we see an error after possibly moving update lists
from the original page to a newly instantiated page, we've corrupted
the original page and cannot continue.
Branch: develop
https://github.com/wiredtiger/wiredtiger/commit/92006f5118ea47b6bd6dd84d46f8cec791d5d950
Author:
{u'username': u'keithbostic', u'name': u'Keith Bostic', u'email': u'keith@wiredtiger.com'}Message: WT-2126: instead of panicing on failure after moving the update chains
to the new page, break the move into two parts: first reference the
update chains from the new pages, and then once that's all done and
the parent page has split, discard the original page's references.
Branch: develop
https://github.com/wiredtiger/wiredtiger/commit/c6be1a08767f4a5caafc7b8cba0f213592ab8342
Author:
{u'username': u'keithbostic', u'name': u'Keith Bostic', u'email': u'keith@wiredtiger.com'}Message: WT-2126: Instead of leaking memory when a multi-page-split fails, set
a flag on the newly created page, allowing us to discard it without
freeing the updates (which are referenced by the original page).
Branch: develop
https://github.com/wiredtiger/wiredtiger/commit/9aee40b1e9cf4c9db0b8fc8f8579d5fe7ed81259
Author:
{u'username': u'michaelcahill', u'name': u'Michael Cahill', u'email': u'michael.cahill@mongodb.com'}Message: Merge pull request #2241 from wiredtiger/wt-2126-panic-on-failure
WT-2126: clean up if there is an error during splits
Branch: develop
https://github.com/wiredtiger/wiredtiger/commit/346ad40308d143269884d47b67ad4519aa952da5
I think the first step here should be testing: can we induce failures at the right points, so we see these things happening in automated tests?