Page MenuHomePhabricator

Investigation: Make sure client change handling scales
Closed, ResolvedPublic1 Estimated Story Points

Description

Given that for example Q159 (Russia) is being used more than 100000 times on ruwiki (mostly with aspect L.ru which makes use run Title::invalidateCache), we should make sure our client change handling can keep up with that.

Right now ChangeHandler iterates over all pages using a given Entity and then hits of the purge/… actions synchronously. I seriously doubt that will work for more than a few thousand pages, at best (given that jobs can only run for a limited time and memory is limited as well).

We need to investigate this and fix change handling (if necessary).

Event Timeline

hoo raised the priority of this task from to High.
hoo updated the task description. (Show Details)
hoo added subscribers: hoo, aude, daniel and 2 others.
Tobi_WMDE_SW renamed this task from Make sure client change handling scales to Investigation: Make sure client change handling scales.Jun 30 2015, 1:29 PM
Tobi_WMDE_SW set Security to None.
hoo closed this task as Resolved.EditedJul 9 2015, 11:59 PM
hoo claimed this task.

Resolving, as this is only about investigating… This is definitely a problem:

Example log entry:

Jul  9 08:28:46 mw1014:  #012Fatal error: request has exceeded memory limit in /srv/mediawiki/php-1.26wmf12/extensions/Wikidata/extensions/Wikibase/client/includes/Usage/Sql/EntityUsageTable.php on line 441

Happened several times today:

hoo@fluorine:/a/mw-log$ grep -c 'Fatal error: request has exceeded memory limit in /srv/mediawiki/php-1.26wmf13/extensions/Wikidata/extensions/Wikibase/client/includes/Usage/Sql/EntityUsageTable.php' hhvm.log 
37

Maybe I shouldn't have closed this given we have no concrete ways forward.

@daniel: It would be nice if you could file tickets for improving this as needed (we probably want to implement batching on some level… AFAIR you already looked into this).

As Daniel didn't create any tasks, I went ahead and created T107722.