Last modified: 2013-04-19 05:25:00 UTC
Category sorting on pt.wikipedia.org is broken due to ICU sort key interpretation changes between Lucid and Precise. I am depooling Precise appservers and job queue runners temporarily. "updateCollation.php --force" will have to be run from a Lucid server to fix the database. A test case: <https://pt.wikipedia.org/wiki/Categoria:Engenheiros_da_Alemanha> It may fix itself if the article in question is edited.
I've depooled the precise appservers, stopped the precise job runners, and I've started running the script.
The script is finished.
@todo FIXME: Needs unit tests
(In reply to comment #3) > @todo FIXME: Needs unit tests Unit tests don't address this, necessarily. The sorting order is presumably consistent in both lucid and precise, in that Alucid sorts before Blucid and Aprecise sorts before Bprecise, it's just that Blucid sorts before Aprecise which is causing problems with a mixed-distro cluster. If something is due to interaction between two machines running different software, it's generally hard to write a unit test for it.
pt.wikipedia.org should be fixed for now, but will break again if we try to keep going with the Ubuntu Precise upgrade. My understanding of the plan: * We will need to temporarily have php5-intl built against ICU 4.2 for Precise. Faidon will build this package. * We deploy Precise using the custom php5-intl build. (Peter Youngmeister) * We deploy stock php5-intl during a single deployment window, then rebuild the sort keys (TBD)
This is now built and deployed in apt. I'm leaving the ticket open, to remind us to rebuild indexes and revert that change, when all appservers are precise.
(In reply to comment #6) > ... when all appservers are precise. Is this done?
All appservers are precise, however we haven't reverted that change and rebuilt indexes yet.
(In reply to comment #8) > All appservers are precise, however we haven't reverted that change and > rebuilt indexes yet. Faidon: Is that planned? Also wondering if fixing bug 46036 has any influence on this (as per comment 5).
(In reply to comment #8) > All appservers are precise, however we haven't reverted that change and > rebuilt indexes yet. This was done on March 13.