[00:49:35] <TimStarling>	 didn't have IRC switched on because I assumed nobody would be around
[00:50:07] <TimStarling>	 forgot that Krinkle is a robot and doesn't sleep ;)
[23:20:46] <TimStarling>	 I want to deploy the SQL mainstash
[23:21:15] <TimStarling>	 so ideally I wouuld like a reply to my comment on https://gerrit.wikimedia.org/r/c/operations/mediawiki-config/+/752807/
[23:22:16] <TimStarling>	 probably from Krinkle or AaronSchulz
[23:23:43] * AaronSchulz got distracted reading about timeouts
[23:29:01] <AaronSchulz>	 TimStarling: https://gerrit.wikimedia.org/r/c/mediawiki/core/+/789725 should still be resolved first
[23:29:37] <TimStarling>	 ok
[23:33:18] <Krinkle>	 TimStarling: replied on wmf-config change
[23:37:38] <TimStarling>	 ok, so you would support removal of that feature and simplification of the config? maybe it is best to do that before deploying it
[23:38:23] * Krinkle checks what, if any, of it was there before the refactor
[23:38:30] <Krinkle>	 https://gerrit.wikimedia.org/g/mediawiki/core/+/REL1_34/includes/objectcache/SqlBagOStuff.php#127
[23:38:57] <Krinkle>	 I wasn't sure if maybe some form of localLB already existed
[23:39:38] <Krinkle>	 looks like not, no, so yeah it wouldn't be used I think.
[23:39:59] <Krinkle>	 If in the future we need to spread it over multiple servers, and hash it etc, there's already affordances for that as we do with PC.
[23:40:14] <TimStarling>	 I would just have a config key "cluster" with the cluster name, having closures in config seems inelegant
[23:40:26] <Krinkle>	 yeah, like we do with externalstore
[23:41:14] <Krinkle>	 It could take LB for DI purposes, but we can prettify that through wiring like we do with other services, there's a case for this in the ObjectCache helper method we use.
[23:42:09] <Krinkle>	 TimStarling: ok, the above diff Aaron links, I see your point now. that helper sets a missing localLB
[23:42:11] <Krinkle>	 that's a problem indeed
[23:42:19] <Krinkle>	 I missed that.
[23:43:19] <Krinkle>	 I think that's extra reason and evidence that it's non-trivial complexity. This split has already been reason for half a dozen additional patches to get it working.
[23:47:19] <AaronSchulz>	 removal of localKeyLB seems pretty simple
[23:50:10] <AaronSchulz>	 the only serious LB-based configuration is the default (local LB fallback) and  (some global LB and global cluster and domain)
[23:56:14] <Krinkle>	 ack, in REL1_14 there is a reference to local LB though it wasn't properly DI'ed yet, it was instead a fallback in getConnection()
[23:57:07] <Krinkle>	 so that default is something we'll want to keep for continuity, and then to support global keys and/or move to separate hardware, one can set it to a different external LB cluster 
[23:57:19] <Krinkle>	 sounds good?
[23:59:44] <TimStarling>	 sounds good to me
[23:59:57] <TimStarling>	 it's confusing not having any documentation of the config