Page MenuHomePhabricator

Index on oresc_probability, temporarily or permanently
Closed, ResolvedPublic


We're trying to improve performance for RecentChanges queries involving ORES (see parent task, as well as others). One of the thing we think might help is to add an index on (oresc_rev, oresc_probability).

However, we're not sure. Do you think it would be better to do it on a test replica, or do it in production and proceed from there?

We would like to test on a replica of large recentchanges tables (in particular enwiki and ruwiki), preferably on machines with performance similar to production.

Event Timeline

Not directly related, but for background, and could be relevant regarding recentchanges scaning, and why it has become a problem for many queries lately: T171027#3599821

@jcrespo asked on T171027 that one of us test "on beta or somewhere else with equivalent sizes/row distribution", so that is likely the next step.

I want to come back to this, given that ores_classification table is now only keeps data for 30 days meaning it won't grow too big anyway.

I'm officially picking this up but first, I need a maintenance script to clean the old scores lingering in the table.

This index is already in place.