Closed MatmaRex closed 1 year ago
I cleared out some disk space using:
journalctl --vacuum-size=100M
(Unrelated to the databases, there was just 1.5 GB of logs in /var/log/journal.)
And:
root@patchdemo3:/var/lib/mysql# find * -name 'patchdemo_*' -exec mysql -e "truncate l10n_cache;" {} \;
(Gained about 4 GB, but this is temporary until someone visits the wikis. Maybe we should do this in a cron job though…)
I think long term we need a wiki expiration policy (aka auto deleting) if we are to reduce the maintenance burden. Today it's a wiki using 5x the space, but tomorrow it could be 10x teams using the service.
One blunt way to deal with this problem is to disable the use of the cache:
$wgLocalisationCacheConf['storeClass'] = 'LCStoreNull';
I just realized that I've been using this setting locally for who-knows-how-many years, and my local testing wiki feels fast. Maybe that's an option.
We're running out of disk space on
/
again. Apparently this is because Patch demo has been very helpful for testing changes to the interface of language variants ;)There are a few wikis using the
zh
language code with databases around 150 MB, compared to the average of 30-35 MB:They are huge because all of the data in
l10n_cache
is duplicated for each language variant:I'm not sure whether this is expected, or a bug in MediaWiki.