[09:18:56] Emperor: o/ for the -be hosts in https://phabricator.wikimedia.org/T401966 we can probably do the following - when you have time during the next days, just ping me and I'll run provisioning after the depool etc.. [09:19:10] so we can keep a minimal downtime for those hosts [09:19:13] does it make sense? [09:35:35] elukey: you can't depool backend hosts [09:36:19] If you're reasonably confident it'll be brief downtime and non-destructive, we can just do one, check it's OK, move on. [09:36:34] Otherwise, it has to be drain from rings, work, restore to rings. [09:36:38] But that takes weeks [09:36:59] Emperor: yeah that it what I meant with depool, even if I didn't remember it was so long [09:37:22] so brief may be some hours, worst case it can happen what we had to do with ms-fe1013 [09:37:37] best case scenario it is some minutes [09:38:04] I think that's OK to do live, but we should check each host is OK when done before moving to the next. The cluster is designed to withstand host failure, after all [09:38:13] and not on a Friday :) [09:45:13] anyhow, ms-fe1013 passes the kitten test (TM), so repooled [10:08:38] yep let's sync on monday [10:09:15] 👍