[06:10:32] m2-master failed over [07:44:26] How does a varchar -> binary schema change work [07:45:36] For Miraheze when doing mediawiki upgrades, i categorise a schema change as can be done any time (new table/column, add default), must be done asap (index rename) or must be done after (column/table drop) [07:45:48] I'm not sure where varchar->binary would fit [07:46:54] Something like https://github.com/wikimedia/mediawiki-extensions-CentralAuth/compare/REL1_37...master#diff-eefb775240533cd9792690fdc4504fbbb5b1f40126a431aa1573c1ac9c66fd7fR218 [09:46:44] I need to run some slow database queries on the unredacted production centralauth databases (https://phabricator.wikimedia.org/T303590), what's the best place for that? just a random codfw replica? or should I be asking for dbstore access? [09:57:43] taavi: codfw should be fine yes [09:58:11] taavi: logging will always help for awareness [09:58:17] RhinosF1: it is hard to know your internal priorities, it really depends on you [10:28:16] marostegui: it's more of a question of when that can be ran [10:28:32] Like is it safe to run 1.37 with it applied or 1.38 with it not [10:29:06] Obviously it's impossible to run a schema change in flight without any users noticing an error during the upgrade so that's aimed to be avoided [10:31:30] RhinosF1: from a DB point of view, anytime. with regards of mw version, that I don't know [10:33:25] marostegui: do you know how varchar -> binary works? Like in terms of what's the difference between the data that's inserted. [10:34:02] it should make no difference if you don't change the data upstream of course [10:34:17] but that's a data type change so it will lock the table [10:34:35] so you cannot run the change on the master, or if you do, writes will get stuck until it is done [11:42:51] * Emperor waits for PCC to crush all their dreams [12:04:43] PCC says yes \o/ [13:17:20] m1-master is back with the original proxy after upgrading the kernel [14:11:04] taavi: I personally use analytics replicas [14:11:42] https://wikitech.wikimedia.org/wiki/Analytics/Systems/MariaDB [14:13:08] I am leaving db1099 depooled for the weekend and on MOnday I will start repooling [14:13:16] That was the host that caused issues yesterday [15:16:24] Amir1: i don't currently have access to those hosts, just trying to avoid an additional access request process if I can [21:23:28] marostegui: (for Monday), one of infra guys just tried to get an optimiser trace and it performs fine now