[14:25:03] <Krinkle>	 AaronSchulz: apparently job execution is for the first time in 10 years now moving from HTTP (back) to CLI at WMF. This seems like a big deal given the various differences in behaviour and configuration there, but it's not clear that any input is asked. In any case, you may want to have a look at https://phabricator.wikimedia.org/T369048. I did a review just now as well, and found 1 minor issue. I've not yet fully reviewed what the 
[14:25:03] <Krinkle>	 MaintenanceRunner overhead and (lack of) restInPeace might do, as well as CLI conditions more generally. I guess it helps that it still uses the same EventBus/JobExecutor wrapper which already took care of most things.
[14:25:39] <Krinkle>	 TLDR: There is a new runSingleJob maint script which is being invoked as one-off k8s jobs 
[14:26:30] <Krinkle>	 It's not clear how or when it is being rolled out. It might be unused right now, with intent to mainly use for videoscaling at first (to get them off bare metal).
[16:34:15] <AaronSchulz>	 Krinkle: another job runner script? heh :)
[17:00:36] <AaronSchulz>	 Krinkle: https://phabricator.wikimedia.org/T379809 reminds me of some of the main things that might change. The few jobs and deferred updates using the "no explicit trx round" flags would have some subtle changes. This could effect DomainEvent listener execution in annoying ways.
[17:01:48] <AaronSchulz>	 I was already working on quashing those cli/web differences, though I was assuming jobs were staying on web mode. Good to know what's going on.
[17:52:15] <Krinkle>	 AaronSchulz: yeah.. I think we're in a much better position now than 10y ago, but still somewhat unexpected/surprising.
[18:11:45] <Krinkle>	 cscott: I'm improving the 'test coverage' CI comment, I think you ran into this last week that "decreased" can sometimes be falsey reported when the build simply fails. Patch - https://gerrit.wikimedia.org/r/1105412