dbperformance.log on fluorine shows lots of DB writes on page views. This could be faster in multi-DC setup (and avoid SPOFs in general) if they used the local 'enqueue' queue to get the jobs into the master DC for execution. The code would have to be put into a job class of course.
|Open||aaron||T88445 MediaWiki active/active datacenter investigation and work (tracking)|
|Resolved||aaron||T91837 Consider moving various DB writes on page views to using local jobs|
The summary is pretty brief, but I (and the other archcom people) support the general idea. Theoretically this could be fleshed out by adding a list of writes involved and strategies for job-ifying them, but that doesn't change the fact that I think this is a good idea and should go ahead. You (@aaron) probably know what you're doing here better than anyone else anyway, so just go ahead and do it :)