| *** auk has joined #kata-dev | 03:08 | |
| *** auk has quit IRC | 04:28 | |
| *** auk has joined #kata-dev | 04:41 | |
| *** auk has quit IRC | 05:11 | |
| *** snir_ has quit IRC | 06:22 | |
| *** sameo has joined #kata-dev | 06:55 | |
| *** amorenoz has joined #kata-dev | 07:45 | |
| *** fgiudici has joined #kata-dev | 07:54 | |
| *** sgarzare has joined #kata-dev | 07:55 | |
| *** sameo has quit IRC | 07:59 | |
| *** ailan has joined #kata-dev | 08:02 | |
| *** snir_ has joined #kata-dev | 08:11 | |
| *** amorenoz_ has joined #kata-dev | 08:13 | |
| *** amorenoz has quit IRC | 08:16 | |
| *** jodh has joined #kata-dev | 08:17 | |
| *** jodh has quit IRC | 08:30 | |
| *** jodh has joined #kata-dev | 08:33 | |
| *** auk has joined #kata-dev | 08:36 | |
| *** auk has quit IRC | 08:36 | |
| *** davidgiluk has joined #kata-dev | 09:00 | |
| *** sameo has joined #kata-dev | 09:06 | |
| *** jodh has quit IRC | 09:25 | |
| *** snir_ has quit IRC | 10:09 | |
| *** snir_ has joined #kata-dev | 10:10 | |
| *** fgiudici has quit IRC | 11:05 | |
| *** fgiudici has joined #kata-dev | 11:05 | |
| *** sameo has quit IRC | 11:55 | |
| kata-irc-bot | <fidencio> @jfreiman, thanks for the quick answer in the BZ. | 12:03 |
|---|---|---|
| *** sameo has joined #kata-dev | 12:26 | |
| *** devimc has joined #kata-dev | 13:09 | |
| *** crobinso has joined #kata-dev | 14:04 | |
| *** fuentess has joined #kata-dev | 14:14 | |
| *** jodh has joined #kata-dev | 14:18 | |
| *** devimc has quit IRC | 15:24 | |
| *** devimc has joined #kata-dev | 15:24 | |
| *** snir_ has quit IRC | 15:31 | |
| *** dklyle has joined #kata-dev | 15:56 | |
| *** sgarzare has quit IRC | 16:58 | |
| kata-irc-bot | <fidencio> @eric.ernst, may I borrow your brain for a while? So, for $reason, we *must* rely on `cAdvisor` for what we're doing downstream, at least till Peter Hunt's KEP gets merged. Knowing that, is there a document on what kind of limitation we may hit on purely relying on `cAdvisor`? We know, we can't get `container` level stats with that, but having the `pods` one would be enough to *not* screw up the scheduler about the resources? I'm | 17:35 |
| kata-irc-bot | considering here that pod level stats== (or similar enough) to the container level stats *plus* the podOverhead, but am I missing some edge case? | 17:35 |
| kata-irc-bot | <fidencio> cc @adel.zalok.89 | 17:36 |
| kata-irc-bot | <eric.ernst> Pod stats is best scenario, and what you should use. | 17:41 |
| kata-irc-bot | <eric.ernst> You just lose ability to provide user the usage info for the container itself. | 17:41 |
| kata-irc-bot | <eric.ernst> which they may want. | 17:42 |
| kata-irc-bot | <eric.ernst> You should verify/make sure that you are gathering the pod level stats when using cadvisor. | 17:42 |
| kata-irc-bot | <eric.ernst> can you see these? | 17:42 |
| kata-irc-bot | <fidencio> @eric.ernst, yep, we're getting those. | 17:45 |
| kata-irc-bot | <eric.ernst> Cool. Yeah, I'd use that. Again, it may be confusing if you expose those to the user, but for infra ppl, this is what I'd use. | 17:45 |
| kata-irc-bot | <eric.ernst> (ie, measures *actual* usage). | 17:45 |
| kata-irc-bot | <fidencio> Those are the only things we can get. :slightly_smiling_face: Pod Level memory & CPU. We can hack around to get networking stats through the VMM. | 17:46 |
| kata-irc-bot | <fidencio> We lose `kubectl top` & similar commands, but that's fine for now, at least till Peter's KEP is implemented. | 17:46 |
| *** fgiudici has quit IRC | 17:47 | |
| kata-irc-bot | <fidencio> Mind that my main concern is not exactly "infra ppl", but rather providing some wrong measurement to the scheduler | 17:48 |
| kata-irc-bot | <fidencio> And, because of that, start "unknown" issues on vanilla pods. | 17:49 |
| kata-irc-bot | <eric.ernst> Default config of scheduler just looks at req/limits. | 17:54 |
| kata-irc-bot | <eric.ernst> I *think* eviction uses pod as well. | 17:54 |
| kata-irc-bot | <fidencio> Okay, I'll start a document about Kata Containers + cAdvisor and have it as part of our "known limitations" document | 18:00 |
| kata-irc-bot | <fidencio> I may need your review there, then. | 18:00 |
| *** jodh has quit IRC | 18:02 | |
| kata-irc-bot | <eric.ernst> Sounds good. | 18:09 |
| kata-irc-bot | <fidencio> @eric.ernst, btw, I'd appreciate if you have an `ack` to spare: https://github.com/kata-containers/kata-containers/pull/1423 | 18:14 |
| *** ailan has quit IRC | 18:35 | |
| *** davidgiluk has quit IRC | 20:17 | |
| *** fuentess has quit IRC | 21:02 | |
| *** devimc has quit IRC | 22:01 | |
| kata-irc-bot | <eric.ernst> done. | 22:13 |
| kata-irc-bot | <eric.ernst> Seems like stable-2.0 CI is failing a few of the jobs that I would expect. Are we up to date with respect to main? @gabriela.cervantes.te you know? | 22:14 |
| *** sameo has quit IRC | 23:26 | |
Generated by irclog2html.py 2.17.2 by Marius Gedminas - find it at https://mg.pov.lt/irclog2html/!