Wednesday, 2021-01-20

*** fuentess has quit IRC00:01
kata-irc-bot1<bergwolf> yeah, we can create a main branch first, make sure it works with CI and then make it default, wdyt?02:07
*** dklyle has quit IRC02:52
*** david-lyle has joined #kata-dev02:52
*** david-lyle has quit IRC03:01
*** pcaruana has joined #kata-dev06:35
*** ailan has joined #kata-dev07:08
*** sameo has joined #kata-dev07:09
*** Yarboa has joined #kata-dev07:46
*** sameo has quit IRC08:03
*** fgiudici has joined #kata-dev08:03
*** Yarboa has quit IRC08:10
*** Yarboa has joined #kata-dev08:10
*** sameo has joined #kata-dev08:17
kata-irc-bot1<fidencio> That would be perfect.08:19
*** Yarboa has quit IRC08:30
*** Yarboa has joined #kata-dev08:32
*** fgiudici_ has joined #kata-dev08:38
*** jodh has joined #kata-dev08:39
*** sameo_ has joined #kata-dev08:40
*** fgiudici has quit IRC08:41
*** sameo has quit IRC08:41
*** sgarzare has joined #kata-dev09:00
*** davidgiluk has joined #kata-dev09:01
*** sameo_ has quit IRC09:02
*** sameo has joined #kata-dev09:02
*** Yarboa has quit IRC09:06
*** Yarboa has joined #kata-dev09:20
*** snir_ has quit IRC09:26
*** pcaruana has quit IRC12:43
*** pcaruana has joined #kata-dev12:55
*** devimc has joined #kata-dev13:09
*** ailan_ has joined #kata-dev13:35
*** ailan has quit IRC13:38
kata-irc-bot1<aadam> I see you are already on the invite :(13:52
kata-irc-bot1<aadam> Can you check if you have an invite for next week?13:53
*** fuentess has joined #kata-dev14:10
*** crobinso has joined #kata-dev14:14
*** irclogbot_2 has quit IRC14:27
*** irclogbot_1 has joined #kata-dev14:30
*** snir_ has joined #kata-dev15:06
*** sameo has quit IRC15:19
*** sameo has joined #kata-dev15:19
*** devimc has quit IRC15:27
*** devimc has joined #kata-dev15:27
*** dklyle has joined #kata-dev15:31
*** devimc has quit IRC15:36
kata-irc-bot1<christophe> When I run `podman run -it --rm --runtime kata --pod some_pod fedora bash`, I get an error ```Error: open /run/vc/sbs/distracted_khorana: no such file or directory: OCI not found```16:10
kata-irc-bot1<christophe> I have not tried that in a long time, but I think it used to work.16:10
kata-irc-bot1<christophe> That's with `kata-runtime` 1.12.16:10
kata-irc-bot1<christophe> The pod was started, and there is a `pause` in it16:11
kata-irc-bot1<christophe> Also, if I run without the `--runtime kata` option, then it starts a container and reports that it's in the pod, but it does not look like it is running in a VM based on a `cat /proc/cmdline` (showing host kernel)16:14
kata-irc-bot1<jakob.naucke> Maybe I'm thinking way too simple here, but isn't it expected behaviour that a container runs with runc or similar and thus unhypervised when you don't tell your engine to use Kata?16:15
kata-irc-bot1<christophe> Well, probably so, but I was checking if there were some smarts deducing the runtime from the pod. It's weird that `podman` reports two containers in the same pod, when one is running in a VM and the other is not.16:16
kata-irc-bot1<afrosi057> @christophe Do you want to understand if you are running in a standard or in a kata container within the container? I think if you check with `mounts`  you should get different output if you are running in a standard container or kata. There was this trick mentioned somewhere in the Kata docs. Otherwise a nice tool is `systemd-detect-virt`  this should returns the virtualization. You need however to install it in the container16:46
kata-irc-bot1<jakob.naucke> Since I couldn't find instructions on this in Kata's docs or issues: Is it possible to get a guest console in case of an immediate agent failure (due to segfaults in my case)? By default, the sandbox shuts down instantly; it would be helpful to be able to keep it alive and re-inspect the agent execution.16:56
*** devimc has joined #kata-dev17:12
*** sgarzare has quit IRC17:14
*** Yarboa has quit IRC17:20
kata-irc-bot1<eric.ernst> There isn't a config for this.  I think you'd have to add some changes (including in the caller itself, who will likely timeout and kill the container/pod)17:25
*** Yarboa has joined #kata-dev17:31
*** fgiudici_ has quit IRC17:44
kata-irc-bot1<fidencio> @eric.ernst, last year you were playing with cgroups, do you remember getting no info about Blkio at all? This is what I'm getting here: `virtcontainers.BlkioStats={[] [] [] [] [] [] [] []}`18:02
*** jodh has quit IRC18:02
kata-irc-bot1<fidencio> And this is 2.x, latest commit.18:02
fidenciodevimc: ^^18:07
devimcyeah, blame Eric18:08
devimc:D18:08
*** Yarboa has quit IRC18:08
kata-irc-bot1<eric.ernst> Ha.18:13
kata-irc-bot1<eric.ernst> I haven't looked at these stats.18:13
devimcfidencio, afaik we don't support it yet18:14
kata-irc-bot1<eric.ernst> I mostly was interested in the right hierarchy/setup.18:14
devimcafaik, only cpu and devices are supported18:14
kata-irc-bot1<fidencio> CPU and Memory, you mean?18:14
devimcwe rely on container manager18:14
devimcfor memory18:15
kata-irc-bot1<eric.ernst> ?18:15
kata-irc-bot1<eric.ernst> We gather the usage stats per container, as well, for memory and cpu.18:15
kata-irc-bot1<eric.ernst> I'm not sure how useful blkio really is, but we can look at adding.18:15
devimcbut we don't apply any constraint18:16
kata-irc-bot1<eric.ernst> its probably more useful to know the actual blkio for the whole thing (VMM, etc)18:16
kata-irc-bot1<eric.ernst> in the guest we do18:16
kata-irc-bot1<eric.ernst> for sure.18:16
kata-irc-bot1<fidencio> Well, we need it as in OCP displays those metrics to the user18:16
devimcyeah, but I guess fidencio wants the stats in the host18:16
kata-irc-bot1<fidencio> I'm working on what will be the effort on implementing it18:16
devimcfidencio, right?18:16
kata-irc-bot1<eric.ernst> Let's clarify :slightly_smiling_face:18:16
kata-irc-bot1<eric.ernst> you'll want pod level stats for administration.18:17
kata-irc-bot1<eric.ernst> And container level stats as well for users.18:17
kata-irc-bot1<fidencio> devimc, that's a good question.  I need pod level stats.18:17
kata-irc-bot1<fidencio> let me dig into this a little bit more and open an issue about that later Today18:18
kata-irc-bot1<eric.ernst> then it should be via the pod cgroup; who collects blkio today in your setup? cadvisor?18:18
devimcfidencio, see this https://github.com/kata-containers/kata-containers/blob/4bb23ed9903c1a47157444088e9255313c9282d2/src/runtime/virtcontainers/sandbox.go#L543-L60218:19
devimconly cpu and devices constraints are applied18:19
devimcthe rest is up to k8s/containerd or whatever18:19
*** snir_ has quit IRC18:20
*** Yarboa has joined #kata-dev18:20
fidenciodevimc: ack, let me see what containerd does on their side18:21
fidencioI may need to get back here asking silly questions in a day or so :-)18:21
kata-irc-bot1<eric.ernst> yeah, but even if unconstratined, we should be able to measure usage.19:19
kata-irc-bot1<fidencio> @eric.ernst, this is interesting: https://github.com/kata-containers/kata-containers/blob/4bb23ed9903c1a47157444088e9255313c9282d2/src/agent/rustjail/src/cgroups/fs/mod.rs#L14[…]L15119:25
kata-irc-bot1<eric.ernst> yeah, i'm honestly not sure how 'useful' per contaienr (in guest) stats are for blk io.19:26
kata-irc-bot1<eric.ernst> versus what is observed on the host.19:26
kata-irc-bot1<eric.ernst> especially if you're trying to assess io in the guest for somethign that is proxied into the guest, but is actually a mount on the host......19:27
kata-irc-bot1<fidencio> I tend to agree, Eric.  And I guess the same would apply for the networking stats19:28
kata-irc-bot1<fidencio> Okay, I was chatting with Vivek and Dave and, yep, BlkIO stats is something that doesn't seem to make any sense on the kata-containers context as in our case "there is no block layer. filesystem messages (fuse) are directly sent to virtiofs daemon adn they don't go through block layer at all"19:34
*** davidgiluk has quit IRC20:12
fidencio@eric.ernsts, devimc, I guess it does make sense to get the stats from the VMM and pass those up to the container runtime20:38
devimcfidencio, but you should also consider the resources used by virtiofsd20:57
fidenciodevimc: I was thinking about checking the sandboxStats for that21:00
fidencioand then be sure we have both VMM and virtiofs accounted there21:00
devimcsounds good21:02
*** ailan_ has quit IRC21:04
*** sameo has quit IRC21:27
*** sameo has joined #kata-dev21:32
*** snir_ has joined #kata-dev21:33
kata-irc-bot1<eric.ernst> Can you just obtain these from the pod-cgroup?21:54
kata-irc-bot1<eric.ernst> It should account for both the VMM, virtiofsd, etc.21:54
*** devimc has quit IRC21:55
*** Yarboa has quit IRC21:59
*** Yarboa has joined #kata-dev22:11
*** Yarboa has quit IRC22:29
*** Yarboa has joined #kata-dev22:31
kata-irc-bot1<fidencio> @eric.ernst, I think we're talking about the same thing. :slightly_smiling_face:22:34
kata-irc-bot1<fidencio> when I mentioned sandbox Stats I was referring to the pod cgroups we could get when using sandbox_cgroup_only=true22:34
kata-irc-bot1<fidencio> are we talking about the same thing?22:34
kata-irc-bot1<eric.ernst> I see, I thought you meant gathering from the VMM directly (somehow)22:35
kata-irc-bot1<eric.ernst> looks good.22:35
kata-irc-bot1<fidencio> devimc, @eric.ernst, thanks for the brainstorm!22:37
*** sameo has quit IRC22:39
*** Yarboa has quit IRC22:49
*** Yarboa has joined #kata-dev23:01
*** Yarboa has quit IRC23:19
*** Yarboa has joined #kata-dev23:20
*** fuentess has quit IRC23:26
*** crobinso has quit IRC23:33
*** Yarboa has quit IRC23:49
*** Yarboa has joined #kata-dev23:54

Generated by irclog2html.py 2.17.2 by Marius Gedminas - find it at https://mg.pov.lt/irclog2html/!