[16:54:08] Krinkle: I read through the arclamp code, but I didn't see an obvious way to add in the OS name. [16:54:31] or were you thinking we'd set up an additional pipeline (https://gerrit.wikimedia.org/g/operations/puppet/+/8a4a2272d5d9bb14365d9ed4af8d0dd8f67c6732/modules/arclamp/manifests/init.pp#142) for stretch/buster? [17:38:37] dpifke: thanks for taking a look. disclaimer is that I don't understand how this whole system fully works and what the constraints are, so please suggest better options if you have some :) [17:41:16] I think I'm mostly on board with having the buster hosts (temporarily?) duplicate their traces to an excimer-buster pipeline, in addition to the regular one. We can then compare the two during the transition, and not have any long-term impacts to historical data. [17:42:52] The other option I think is to treat buster as a separate entrypoint, and just generate another set of graphs from the same input logs. [17:43:30] I'm not 100% sure the best way to do this, but I think it might be what Krinkle originally had in mind. [17:43:58] We'll discuss at our team meeting later today and I can help you get this set up later this week. [17:43:58] we would like to be able to compare specific buster entrypoints to stretch ones if that's possible, the current theory is that the slowdown is only on API hosts [17:44:05] thank you :) [17:45:26] So long as buster is a small % of hosts, we don't really need a separate stretch pipeline. The data is being sampled anyways, so we can assume there are very few buster samples in the main logs. [17:47:32] * legoktm checks spreadsheet [17:48:57] ~10% of eqiad is buster right now [17:49:24] 75% of codfw is buster, but that should be minimal to non-existent traffic [17:51:12] * legoktm updates patches to drop stretch pipeline [23:56:35] gilles: so I can actually log in with matrix.org on chat.fosdem. It's a little weird, because I cannot join channels on chat.fosdem.org from the matrix.org client. So I have to be on chat.fosdem.org to join the channels. *but* when I log in to chat.fosdem.org I actually see all the other channesl I am in on matrix.org, so it's a full matrix-ified client that just exposes a few more join options. And once I join them, my matrix.org client [23:56:35] elsewhere sees all fosdem stuff I joined. [23:57:20] quite nice overall, just a little weird UI initially. It is authorising chat.fosdem as fully delegate client of all my matrix traffic, instead of merely authenticating me as me on their local server. [23:57:39] Might be that they forgot to open up some remaining part of the setup to allow joining directly from matrix.org clients [23:58:36] oh, I figured it out. [23:58:43] it's not at chat.fosdem.org but fosdem.org [23:58:50] when you interact with it from elsewhere [23:59:01] chat. is just where the element web client is