In a nutshell every 5 minutes when jobs are ran there is a load spike up to often 7/8 cores, or sometimes even above 8, which needs to be moved to a dedicated job runner eventually to avoid it impacting the rest of the site.
Description
Description
Event Timeline
Comment Actions
In a nutshell we either need to optimise how jobs are ran or switch to dedicated job runners
Comment Actions
Given the recent outage caused by what I can only assume was fpm running out of resources, this is definitely a UBN issue now
Comment Actions
I'm guessing that's why both SkyWiki and WikiOasis had loading issues a few days ago.
Comment Actions
On a side note, it seems jobs handled by TimedMediaHandler don't blow up the servers. Strange I must say...
Comment Actions
I don't have any issues even hours or days after transcoding, and video works very finely. If anything, there could be other things in conjunction with TimedMediaHandler causing these blow ups. Sure, TMH is one of the most resource-intensive, but that alone hasn't caused any issues whatsoever. Other resource-intensive tasks alongside TMH might be the issue.
Comment Actions
It's more a case of when the jobs do run they are very intensive; although now mwtask11 is online job running is less intensive, but the cross continent latency for mwtask11 does still slow it down.
Comment Actions
Should now be resolved with the switch to running jobs on mwtask11 and running the web server on wikioasis11