| kata-irc-bot | <eric.ernst> Hey ya'll -- I'm tempted to use ```@here``` | 00:05 |
|---|---|---|
| kata-irc-bot | <eric.ernst> FRIENDS OF KATA -- I'm planning on cutting RC0 for 1.4 release *very* soon. If you have a feature that you're hoping to be part of 1.4 that isn't merged yet, please let me know so we can track, and I can help you get it in. If your changes aren't merged by ~EOD Friday, you probably won't make it into release (planned 2/28). | 00:06 |
| kata-irc-bot | <eric.ernst> Similarly, if you landed a fix and have been lazy on backport, please finish this before end of week! | 00:08 |
| kata-irc-bot | <eric.ernst> its happening | 00:09 |
| kata-irc-bot | <eric.ernst> @feng.wang @james.o.hunt @ddebroy @eric.ernst - let's see if we can get consensus and move https://github.com/kata-containers/kata-containers/pull/3406 forward | 00:14 |
| kata-irc-bot | <eric.ernst> @samuel.ortiz @fidencio @james.o.hunt @bergwolf - the filesystem abstraction refactoring PR would be nice to get in, but it'll need a rebase and some love still. What do you think? https://github.com/kata-containers/kata-containers/pull/3623 | 00:15 |
| kata-irc-bot | <dgibson> welp, I was hoping for working sr-iov for dpdk workloads | 03:38 |
| kata-irc-bot | <dgibson> that's in, including some recent bugfixes | 03:38 |
| kata-irc-bot | <dgibson> unfortunately, I don't think it's gonna be usable, because dpdk also requires hugepages | 03:38 |
| kata-irc-bot | <dgibson> which are entirely busted: https://github.com/kata-containers/kata-containers/issues/3752 | 03:39 |
| kata-irc-bot | <dgibson> I doubt I'll have this debugged by friday | 03:39 |
| kata-irc-bot | <dgibson> how on earth does one usably deploy built-from-source kata on a k8s cluster | 03:40 |
| kata-irc-bot | <dgibson> I tried expanding a `make kata-tarball` artifact manually on a cluster that previously had kata-deply of an alpha release used | 03:40 |
| kata-irc-bot | <dgibson> but it looks like kata-deploy is still around, and keeps overwriting my manually installed things with old versions | 03:40 |
| kata-irc-bot | <eric.ernst> I"m guessing what you want to do is edit the existing daemonset | 03:41 |
| kata-irc-bot | <eric.ernst> to utilize the new container image, instead of the old one | 03:41 |
| kata-irc-bot | <eric.ernst> hehehe. | 03:42 |
| kata-irc-bot | <eric.ernst> Yeah, otherwise you have conflicting daemonsets, which certainly would explain what you're seeing (overwriting) | 03:43 |
| kata-irc-bot | <eric.ernst> Once you edit the daemonset, you'll see the new one take place. | 03:43 |
| kata-irc-bot | <dgibson> yeah, this is all greek to me | 03:43 |
| kata-irc-bot | <dgibson> I *vaguely* know what a daemonset is | 03:43 |
| kata-irc-bot | <dgibson> I have no idea how I'd edit one in any sense, let alone to do this specifically | 03:44 |
| kata-irc-bot | <dgibson> I'm guessing I don't want to just kill off the kata-deploy, since that will probably remove the runtimeClass as well | 03:45 |
| kata-irc-bot | <dgibson> and people wonder why I think Kata is an undebuggable nightmare | 04:07 |
| kata-irc-bot | <eric.ernst> kubectl edit works. | 04:11 |
| kata-irc-bot | <eric.ernst> or kubectl delete it | 04:11 |
| kata-irc-bot | <dgibson> edit doesn't help on its own | 04:11 |
| kata-irc-bot | <eric.ernst> Oh? | 04:11 |
| kata-irc-bot | <dgibson> the tarball is built into the container image | 04:11 |
| kata-irc-bot | <eric.ernst> Right; you edit the image that's utilized in the daemonset. | 04:12 |
| kata-irc-bot | <dgibson> so I'd also have to build a custom kata-deploy image, but it somewhere and then edit to point at that | 04:12 |
| kata-irc-bot | <eric.ernst> aah, yes indeed. | 04:12 |
| kata-irc-bot | <eric.ernst> It works as a container image -- that's where the "payload" lives | 04:12 |
| kata-irc-bot | <dgibson> sure | 04:12 |
| kata-irc-bot | <eric.ernst> You'd need equivalent image on a registry somewhere. | 04:12 |
| kata-irc-bot | <dgibson> https://github.com/kata-containers/kata-containers/issues/3755 | 04:13 |
| kata-irc-bot | <dgibson> fwiw | 04:13 |
| kata-irc-bot | <eric.ernst> i mean, this is less about kata, and more about managing assets / infrastructure. | 04:13 |
| kata-irc-bot | <dgibson> not really | 04:13 |
| kata-irc-bot | <dgibson> I mean, it's about both | 04:13 |
| kata-irc-bot | <eric.ernst> kata deploy is a decent pattern, but you need to decide how to install binaries on your nodes | 04:14 |
| kata-irc-bot | <eric.ernst> if you want to use a daemonset, you need a container image to deploy it. I think that's pretty slick/easy. | 04:14 |
| kata-irc-bot | <dgibson> if kata doesn't have a reasonable way to go from source to something I can test and debug, that's an abject failure in developer experience | 04:14 |
| kata-irc-bot | <eric.ernst> othrewise, you can make RPMs, and just do a yum install across the nodes | 04:14 |
| kata-irc-bot | <dgibson> sure, there's a million ways, and you have to discover them yourself | 04:14 |
| kata-irc-bot | <eric.ernst> if you have a single node, scp the binaries | 04:14 |
| kata-irc-bot | <eric.ernst> if you're doing a single machine, there are details on that. | 04:14 |
| kata-irc-bot | <dgibson> except if you've used kata-deploy before, it will screw you up | 04:14 |
| kata-irc-bot | <eric.ernst> :shrug: | 04:15 |
| kata-irc-bot | <dgibson> where are these details for a single machine? I haven't spotted them | 04:15 |
| kata-irc-bot | <eric.ernst> there's a pssh equivalent for scp IIRC too if that's easier. | 04:15 |
| kata-irc-bot | <dgibson> I have no idea what pssh is, and scp-ing stuff is not the problem | 04:16 |
| kata-irc-bot | <eric.ernst> pssh --> parallel ssh | 04:16 |
| kata-irc-bot | <eric.ernst> "do this thing on alllll those nodes" | 04:16 |
| kata-irc-bot | <eric.ernst> if you had RPMs, you'd pssh -C "yum install kata-runtime", etc. | 04:16 |
| kata-irc-bot | <dgibson> except that's not enough, because you still need something to install the runtimeclass | 04:16 |
| kata-irc-bot | <dgibson> which you have if you used kata-deploy before, but that will screw you over if you *then* scp things over | 04:17 |
| kata-irc-bot | <eric.ernst> runtimeclass object persists and is at cluster level. | 04:17 |
| kata-irc-bot | <eric.ernst> until you do a kubectl delete runtimeclass, it'll still be up there. | 04:17 |
| kata-irc-bot | <dgibson> huh, ok | 04:18 |
| kata-irc-bot | <dgibson> useful here, although not good usability for other cases of kata-deploy | 04:18 |
| kata-irc-bot | <eric.ernst> Yeah. Let me take a look and see if i can find _any_ docs | 04:18 |
| kata-irc-bot | <dgibson> so... I think the upshot of what you said is if I kill the daemonset, then I should be able to manually install my tarballs and test them | 04:19 |
| kata-irc-bot | <eric.ernst> perhaps it should be linked, this may have some more details to help with understanding: https://github.com/kata-containers/kata-containers/tree/main/tools/packaging/kata-deploy | 04:19 |
| kata-irc-bot | <dgibson> that describes it for a *released* version, not one you've built from source | 04:19 |
| kata-irc-bot | <eric.ernst> Yeah, you should be good if you delete it. | 04:20 |
| kata-irc-bot | <eric.ernst> Right, my point was more that it seemed you had questions about what it does, figured this was background udnerstanding w rt daemonset, runtimeclass, etc. | 04:20 |
| kata-irc-bot | <dgibson> well, sure | 04:20 |
| kata-irc-bot | <dgibson> but if you have to be an expert on all those things before you can even make a simple change and deploy it, we're already failed at developer experience | 04:21 |
| kata-irc-bot | <eric.ernst> but yeah agreed, would be nice if it was easier. | 04:22 |
| kata-irc-bot | <dgibson> in this specific case I am all of the above | 04:22 |
| kata-irc-bot | <dgibson> as any developer would likely be when working with pre-push patches | 04:23 |
| kata-irc-bot | <eric.ernst> Congrats, same here :) | 04:23 |
| kata-irc-bot | <eric.ernst> Anyway, based on what you're describing, I'd definitely just delete the daemonset and go ahead and scp and test your node | 04:23 |
| kata-irc-bot | <dgibson> well, sure | 04:24 |
| kata-irc-bot | <eric.ernst> unless you want to start testing across a large set of nodes, i think that makes more sense then updating the deploy image | 04:24 |
| kata-irc-bot | <dgibson> be nice if it didn't come after half a day's frustration going "why, oh why are my changes not doing anything?" | 04:24 |
| kata-irc-bot | <eric.ernst> yeah. daemonset's gonna keep doing its thing. | 04:24 |
| kata-irc-bot | <dgibson> Nope | 04:48 |
| kata-irc-bot | <dgibson> it appears to have removed the `kubernetes.io/kata-runtime=true` label from my nodes | 04:49 |
| kata-irc-bot | <dgibson> so, I can't run kata containers | 04:49 |
| kata-irc-bot | <eric.ernst> the example runtimeclass you used has a NodeSelect on it. | 04:54 |
| kata-irc-bot | <dgibson> well sure | 04:54 |
| kata-irc-bot | <dgibson> it's the runtimeclass that's suggested with kata-deploy | 04:54 |
| kata-irc-bot | <dgibson> also one of my nodes has gone SchedulingDisabled for no obvious reason :( | 04:55 |
| kata-irc-bot | <dgibson> oh, you've got to be kidding me | 04:59 |
| kata-irc-bot | <dgibson> my kata-static-tar.xz doesn't actually have a guest image | 04:59 |
| kata-irc-bot | <dgibson> things only seemed to work before because I guess the kata-deploy one did for whatever reason | 05:00 |
| kata-irc-bot | <fidencio> David, that's a bug fix rather than a new feature. | 06:38 |
| kata-irc-bot | <fidencio> IMHO, that could be merged *after* -rc0 is cut (as in, backported there) | 06:38 |
| kata-irc-bot | <fidencio> @eric.ernst, @samuel.ortiz, @bergwolf, @james.o.hunt, last night I just got the cloud hypervisor driver working with td-shim | 06:39 |
| kata-irc-bot | <fidencio> I need that in, but I need to cleanup the hell out of what I have. | 06:40 |
| kata-irc-bot | <fidencio> I think I can get that out later Today | 06:40 |
| kata-irc-bot | <fidencio> About the other PRs, I don't think I have cycles to review https://github.com/kata-containers/kata-containers/pull/3623, but I am not opposed to have it in. | 08:27 |
| kata-irc-bot | <samuel.ortiz> @eric.ernst I’ll look at https://github.com/kata-containers/kata-containers/pull/3623 today. | 09:03 |
| kata-irc-bot | <jakob.naucke> Thanks for the heads-up @eric.ernst -- https://github.com/kata-containers/kata-containers/pull/3764 is a fix of a recent regression, but we can backport if it absolutely doesn't work out. | 11:19 |
Generated by irclog2html.py 2.17.3 by Marius Gedminas - find it at https://mg.pov.lt/irclog2html/!