| kata-irc-bot | <samuel.ortiz> @james.o.hunt Good morning! | 06:27 |
|---|---|---|
| sameo | jodh: Hey there | 08:40 |
| kata-irc-bot | <jakob.naucke> @sandeep.gupta re your email "Re: [kata-dev] , "kata-containers-2.2.0-alpha1" with confidential_guest = true (SEV Guest)" :thread: | 08:45 |
| jodh | sameo: morning! | 08:46 |
| kata-irc-bot | <jakob.naucke> this is a shot in the dark and it's probably something with the SEV setup most of us cannot substantially help you with, but are you sure this is correct? > `kernel_params = " root=/dev/vda1...` because with standard Kata, this also prohibits QEMU from booting | 08:46 |
| sameo | jodh: I'm trying to enable tracing with kata+CLH. Is that a setup you're familiar with? | 08:47 |
| jodh | sameo: no - we've only tested with QEMU at this point I'm afraid. | 08:47 |
| jodh | it may well work and of course all the hypervisors need to be tested, but so far, we've had very limited offers for help on testing, etc ;( | 08:48 |
| sameo | jodh: I understand. I'm seeing really weird behaviour with CLH, I'll see if I can fix it. | 08:53 |
| jodh | sameo: thanks! | 08:53 |
| sameo | jodh: Does the runtime decide to upload traces to Jaeger, or it's out of its control? | 08:54 |
| jodh | if runtime tracing is enabled, the traces should be generated automatically for the runtime and sent to the Jaeger collector's UDP socket. | 08:55 |
| jodh | are you trying to get agent traces too? | 08:55 |
| sameo | jodh: No, I disabled agent traces. | 08:55 |
| sameo | To run the collector, I just need to start the all-in jaeger container, right? | 08:56 |
| jodh | yes | 08:57 |
| sameo | jodh: Every one second, I'm getting a "failed to upload traces", with a 400 http error from what seems to be the CLH API server. | 08:57 |
| sameo | that runs on a UNIX socket :-/ | 08:57 |
| jodh | wtf? ;) yeah, to reiterate, we haven't tested that yet ;) | 08:58 |
| sameo | wtf indeed... | 08:59 |
| sameo | jodh: in go.opentelemetry.io/otel/exporters/trace/jaeger/uploader.go, do you know how the collectorUploader http client is initialized? | 09:22 |
| fidencio | jodh: may I bother you to rebase https://github.com/kata-containers/kata-containers/pull/2023 atop of main? | 09:45 |
| sameo | jodh: Now I'm forcing jaeger to go through UDP, and the shim now tries to use UDP to talk to CLH over a unix socket. | 09:52 |
| sameo | jodh: It looks like the http socket is shared... | 09:52 |
| jodh | sameo: I'm afraid I can't dig into this atm, but will chat with cmaf about it later... | 10:00 |
| fidencio | jodh: it blows my mind why on that PR, only on that PR, the static checks are not being triggered | 10:02 |
| fidencio | jodh: oh, I get it! | 10:03 |
| fidencio | force-skip-ci label forces the static checks to **not** run, then being on that state forever | 10:04 |
| fidencio | what we should do, however, is force the static checks to short-circuit and return success on those cases | 10:04 |
| fidencio | as those are (and should be) required tests | 10:04 |
| sameo | jodh: No worries. | 10:21 |
| sameo | jodh: We're hitting a problem with the go runtime re-using sockets. | 10:21 |
| sameo | jodh: One generic tracing question though: Do I have to disable new_netns from the config file for spans to be pushed to jaeger? | 10:28 |
| jodh | sameo: this would be for agent tracing only presumably? no, that isn't required, atleast for qemu. But looking at the comments suggests it might be. we need to review all config options imho. | 10:38 |
Generated by irclog2html.py 2.17.2 by Marius Gedminas - find it at https://mg.pov.lt/irclog2html/!