From http://permalink.gmane.org/gmane.org.wikimedia.mediawiki/41444
I have a strange problem on my (clustered and load-balanced) Mediawiki instance: the 'l10n_cache' table is getting written to *very* excessively (and constantly.) Mediawiki seems to write the same localization information to the database every 3-5 seconds. This is causing MySQL to constantly write binlog files (for replication to the slaves) which fills the partition after a number of hours. Removing all my extensions does not stop the problem. I have APC installed for opt-code caching, and Memcached installed for caching other things (sessions, parser-stuff, etc.) I assume that this is all due to the constant requests made by the load-balancer to the wiki which in turn cause MW to recalculate the cache.
Why is the l10n_cache table getting completely rewritten every few seconds? The cache is not functioning as a cache.
I found one thing that will stop this behavior: $wgLocalisationCacheConf['manualRecache'] = true;
But this feels wrong because it makes me manually recomputed the l10n cache whenever I update/release. (Yet another thing to tack onto the release process.) There are also some configuration variables which will slow the process ($wgUpdateRowsPerQuery) or cache it to disk ($wgLocalisationCacheConf), but this merely masks the problem. Does anyone have an insight into why the l10n_cache table is getting written to so much? Or why Mediawiki isn't actually *caching* the localization messages in the database?
Version: 1.20.x
Severity: normal
Whiteboard: aklapper-moreinfo