[09:32:50] <mszabo>	 we'll likely be looking at our pod sizing tbh, our current sizing is not that different from the current mwonk8s sizes (6 workers, 768M APCu SHM)
[09:33:05] <mszabo>	 the main factor that went into determining this size at the time was to keep it low-ish to help the scheduler
[09:33:33] <mszabo>	 however in practice most workers can be found running at least 2-3 mediawiki pods anyways, so we might adjust the size to account for this reality and benefit from improved cache coherency
[11:13:11] <mszabo>	 and I suppose that will be the case for wmf even more since y'all won't be running a "general purpose" k8s cluster but an MW dedicated one
[12:46:43] <Krinkle>	 Yeah, bigger is better given a large amount of memory spent on gigabytes of duplicate opcache and apcu. But only upto a point given php-fpm tendency to lock (seems to be due to apcu, but I haven't personally got to the bottom of that). We're not able to max out our bare metal so hence the downsize to better use hardware
[12:47:43] <Krinkle>	 https://github.com/krakjoe/apcu/issues?q=is%3Aopen+is%3Aissue+label%3Aperformance