https://bugzilla.wikimedia.org/show_bug.cgi?id=37225
christ...@quelltextlich.at changed:
What |Removed |Added
----------------------------------------------------------------------------
CC| |christ...@quelltextlich.at
--- Comment #59 from Aaron Schulz <aschulz4...@gmail.com> 2012-07-09 17:13:26
UTC ---
(In reply to comment #58)
> What happens when the master can't be reached (in Revision::newFromConds())?
> Any chance we could get a null back instead of an exception?
Normally, no unless the ignoreErrors() function was called on the DB. Running
the sha1 script though, I noticed that sometimes the DB class forgets that the
connection was dropped, and I get raw mysql_query warnings. I think this
happens when DB errors are caught and then the DB is used further down the
line.
--- Comment #60 from christ...@quelltextlich.at 2012-07-09 23:51:48 UTC ---
After some playing around with xmldumps-test, I am able to reliably
reproduce the problem automatically: It's a race condition, that can
be triggered in the following way:
(1) Start with a page, whose last revision is from PersonA.
(2) PersonB starts uploading a new revision of the page.
(3) PersonC starts uploading a new revision of the page.
(4) Request from PersonC finishes
(5) Request from PersonB finishes
Between (2) and (5), PersonB's connection to the master database executes the
SQL sequence
[...]
SELECT /* User::loadGroups */ ug_group FROM `user_groups` WHERE ug_user =
'1'
SELECT /* Block::newLoad Root */
ipb_id,ipb_address,ipb_by,ipb_by_text,ipb_reason,ipb_timestamp,ipb_auto,ipb_anon_only,ipb_create_account,ipb_enable_autoblock,ipb_expiry,ipb_deleted,ipb_block_email,ipb_allow_usertalk,ipb_parent_block_id
FROM `ipblocks` WHERE ipb_address = 'Root'
SELECT /* WikiPage::pageData Root */
page_id,page_namespace,page_title,page_restrictions,page_counter,page_is_redirect,page_is_new,page_random,page_touched,page_latest,page_len
FROM `page` WHERE page_namespace = '0' AND page_title = 'PageZ' LIMIT 1
FOR UPDATE
[...]
If PersonC's request is dealt /between/ this given first and third
SELECT, both revisions (PersonB's and PersonC's) end up in the
database: The oldest version is PersonA's, followed by PersonC's, and
finally PersonB's. In the "Recent Changes", the PersonB's revision
lists its full size as +value.
No database error, connection loss, network problem, etc need to
occur.
Yes, it's only a rough description. But as I will not find time to
investigate the issue further in the next few days, I hope this
description and SQL sequence nevertheless helps others to quickly
reproduce the race condition as well and hunt it down. As the
setup is tedious, I attached some log files. ;-)
--
Configure bugmail: https://bugzilla.wikimedia.org/userprefs.cgi?tab=email
------- You are receiving this mail because: -------
You are on the CC list for the bug.
_______________________________________________
Wikibugs-l mailing list
Wikibugs-l@lists.wikimedia.org
https://lists.wikimedia.org/mailman/listinfo/wikibugs-l
