kata-irc-bot1 | <eric.ernst> That’s about it for the rootfs. There was a chrony service introduced as well for time sync w host (ie if you’re running as systemd entry point instead). The img also has a header on it for nvdimm. | 00:00 |
---|---|---|
kata-irc-bot1 | <eric.ernst> We just export a container rootfs image for distribution of choice, and the systemd service files and agent, some sysctl settings then create the img. | 00:01 |
kata-irc-bot1 | <eric.ernst> The image part we use the osbuilder script (that one isn’t as insane - just take rootfs path and makes the image). | 00:02 |
kata-irc-bot1 | <cole.mickens> Is it more common to use systemd in the image, rather than just the agent? I assume the agent doesn't do extra fancy things like time-sync to the host, etc? | 00:36 |
kata-irc-bot1 | <eric.ernst> it doesn't. | 00:47 |
kata-irc-bot1 | <eric.ernst> Its been debated. | 00:47 |
kata-irc-bot1 | <eric.ernst> It didn't make much delta as far as boot time / footprint IIRC, and we debating patching fixes from systemd, versus ourselves essentially managing things like time sync. | 00:47 |
*** fuentess1 has quit IRC | 02:07 | |
kata-irc-bot1 | <cole.mickens> weird, looks like my containerd's default snapshotter isn't working. | 02:16 |
kata-irc-bot1 | <cole.mickens> `sudo ctr container create` fails unless I add `--snapshotter native`. | 02:16 |
kata-irc-bot1 | <cole.mickens> preferably I could coerce containerd to use zfs, but this seems under-documented, maybe? | 02:16 |
kata-irc-bot1 | <cole.mickens> "configuration.toml: having an initrd defined in the configuration file is not supported" also surprising | 02:33 |
kata-irc-bot1 | <cole.mickens> hm, and the code near that error doesn't give an immediate hint as to *why* | 02:35 |
kata-irc-bot1 | <cole.mickens> ah, that's only for the CLH case, which I'd like to use, but maybe isn't as mainstream as qemu? | 02:36 |
kata-irc-bot1 | <cole.mickens> with qemu: ```May 17 19:45:58 raisin kata[1616134]: time="2021-05-17T19:45:58.074879629-07:00" level=error msg="Failed to connect to QEMU instance" error="dial unix /run/vc/vm/foo/qmp.sock: connect: no such file or directory" name=containerd-shim-v2 pid=1616134 sandbox=foo source=virtcontainers subsystem=qemu``` | 02:51 |
kata-irc-bot1 | <cole.mickens> oh, looks like virtofsd is failing or something | 02:54 |
kata-irc-bot1 | <cole.mickens> okay, zfs snapshotter is working. I'm close, I can taste it. | 03:05 |
kata-irc-bot1 | <cole.mickens> It sure seems like qemu is starting up and then crashing, or it's starting up and then virtiofsd is crashing and then qemu comes down too? | 03:06 |
kata-irc-bot1 | <cole.mickens> ah, does Kata need a special qemu? | 03:07 |
kata-irc-bot1 | <cole.mickens> ```qemu-system-x86_64: -machine pc,accel=kvm,kernel_irqchip: warning: short-form boolean option 'kernel_irqchip' deprecated Please use kernel_irqchip=on instead qemu-system-x86_64: ../util/error.c:59: error_setv: Assertion `*errp == NULL' failed.``` well... interesting... | 03:09 |
kata-irc-bot1 | <cole.mickens> oh wait I even think I figured this out already, this kernel has virtiofs configured as module... gonna have to wait on a compile | 04:08 |
kata-irc-bot1 | <cole.mickens> same results. | 05:35 |
kata-irc-bot1 | <cole.mickens> Is disabling virtio-fs temporarily an optiion? | 05:35 |
*** sgarzare has joined #kata-general | 06:48 | |
kata-irc-bot1 | <fidencio> If you disable virtio-fs you're back at using 9p. Temporarily it may be an option, but for sure not recommended in the long term. | 07:21 |
kata-irc-bot1 | <cole.mickens> It looks like it's probably still just qemu exitting. | 07:22 |
kata-irc-bot1 | <cole.mickens> for some reason the init I'm putting into place in the initrd is causing an "error -8" when the kernel tries to start it. | 07:22 |
kata-irc-bot1 | <cole.mickens> still sorting it out. | 07:23 |
kata-irc-bot1 | <fidencio> On the Fedora side, what we ended up doing was using the same kernel as the guest and generating a initrd, rather than an image | 07:28 |
kata-irc-bot1 | <cole.mickens> That's what I'm doing, the kernel is 5_11 with FUSE_FS and VIRTIO_FS enabled. | 07:29 |
kata-irc-bot1 | <fidencio> And we ended up using https://src.fedoraproject.org/rpms/kata-containers/blob/rawhide/f/kata-osbuilder-generate.service to do so | 07:29 |
kata-irc-bot1 | <fidencio> Ooopsie, https://src.fedoraproject.org/rpms/kata-containers/blob/rawhide/f/kata-osbuilder.sh | 07:29 |
kata-irc-bot1 | <cole.mickens> The kernel boots, then complains about /init, but if I extract the initrd and look at /init, it symlinks to kata-agent built for x86_64-linux, so no idea what the `error -8` is about yet. | 07:30 |
*** sameo has joined #kata-general | 09:09 | |
*** yann-kaelig has joined #kata-general | 10:48 | |
*** devimc has joined #kata-general | 12:44 | |
*** fuentess has joined #kata-general | 12:49 | |
*** yann-kaelig has quit IRC | 13:51 | |
*** openstack has joined #kata-general | 14:11 | |
*** ChanServ sets mode: +o openstack | 14:11 | |
*** fuentess1 has joined #kata-general | 15:07 | |
*** fuentess has quit IRC | 15:09 | |
*** sgarzare has quit IRC | 16:44 | |
*** yann-kaelig has joined #kata-general | 20:10 | |
*** sameo has quit IRC | 20:14 | |
*** devimc_ has quit IRC | 21:18 | |
*** yann-kaelig has quit IRC | 21:44 | |
kata-irc-bot1 | <cole.mickens> how about a pointer to the kernel modules that the various initrd come with | 22:00 |
kata-irc-bot1 | <cole.mickens> I finally caught a bad typo and have gotten to kata-agent crashing at boot now, at least | 22:01 |
kata-irc-bot1 | <cole.mickens> "Address family not supported by protocol". | 22:13 |
kata-irc-bot1 | <cole.mickens> I'm guessing my initrd is _too_ slim still. | 22:13 |
kata-irc-bot1 | <cole.mickens> there we go, added "vhost_vsock" module and it looks like she's running | 22:36 |
*** fuentess1 has quit IRC | 22:40 | |
kata-irc-bot1 | <cole.mickens> kinda seems like I'm hitting this: https://patchwork.kernel.org/project/qemu-devel/patch/20210422145335.65814-2-mreitz@redhat.com/ | 22:52 |
kata-irc-bot1 | <cole.mickens> well, this is unfortunate, the `qemu` invocation that `kata-runtime` produces results in this when I invoke it manually: ```Please use kernel_irqchip=on instead qemu-system-x86_64: ../util/error.c:59: error_setv: Assertion `*errp == NULL' failed. zsh: abort sudo -name sandbox-foo-30051 -uuid dbf42433-2d1b-4628-b7bc-1636a1309ca4``` | 23:02 |
Generated by irclog2html.py 2.17.2 by Marius Gedminas - find it at https://mg.pov.lt/irclog2html/!