[00:01:38] Digging down, it winds up pulling a lot of data from WANObjectCache. [00:03:04] So this might be a case where Mediawiki has been optimized for Wikimedia's caching infrastructure, and that's why it's performing poorly in your setup. [00:04:30] we heavily use redis for caching [00:05:56] and ironically, on both mediawikiwiki and the 'affected wiki', about 20% of the time is spent in that hook - very similar [00:07:23] we're a wiki farm - our infrastructure is not 100% the same as Wikimedia's, but there are a lot of similarities, I guess. [00:08:13] at least that makes collaboration on these kind of issue hunting tasks easier [00:09:34] If the relative % are about the same, that's a good thing. Means any optimizations should help both. [00:22:00] https://phabricator.wikimedia.org/T259956 enjoy [00:25:20] I was not sure which projects to add, hope I've done it correctly [14:44:05] dpifke: empty profiles are the result of (wrongly) enabling both "XHGui" and "Inline profile". At which point it seems our current code leads to xhgui losing out. could be handled in the XWD UI, or in our profiling code by ignoring one or serving an error. [14:44:25] also upstream could handle it better, but it's mostly a matter of garbage in/out from their POV.