Migrating from docker to podman, encountering some issues

Linux@lemmy.ml – 42 points – 1 years ago

I've been messing around with podman in Arch and porting my self-hosted services over to it. However, it's been finicky and I am wondering if anybody here could help me out with a few things.

Some of my containers aren't getting properly started up by podman-restart.service on system reboot. I realized they were the ones that depended on my slow external BTRFS drive. Currently its mounted with x-systemd.automount,x-systemd.device-timeout=5 so that it doesn't hang up the boot if I disconnect it, but it seems like Podman doesn't like this. If I remove the systemd options the containers properly boot up automatically, but I risk boot hangs if the drive ever gets disconnected from my system. I have already tried x-systemd.before=podman-restart.service and x-systemd.required-by=podman-restart.service, and even tried increasing the device-timeout to no avail.

When it attempts to start the container, I see this in journalctl:

Aug 27 21:15:46 arch-nas systemd[1]: external.automount: Got automount request for /external, triggered by 3130 (3)
Aug 27 21:15:46 arch-nas systemd[1]: external.automount: Automount point already active?
Aug 27 21:15:46 arch-nas systemd[1]: external.automount: Got automount request for /external, triggered by 3130 (3)
Aug 27 21:15:46 arch-nas systemd[1]: external.automount: Automount point already active?
Aug 27 21:15:46 arch-nas systemd[1]: external.automount: Got automount request for /external, triggered by 3130 (3)
Aug 27 21:15:46 arch-nas systemd[1]: external.automount: Automount point already active?
Aug 27 21:15:46 arch-nas systemd[1]: external.automount: Got automount request for /external, triggered by 3130 (3)
Aug 27 21:15:46 arch-nas systemd[1]: external.automount: Automount point already active?
Aug 27 21:15:46 arch-nas systemd[1]: external.automount: Got automount request for /external, triggered by 3130 (3)
Aug 27 21:15:46 arch-nas systemd[1]: external.automount: Automount point already active?
Aug 27 21:15:46 arch-nas systemd[1]: external.automount: Got automount request for /external, triggered by 3130 (3)
Aug 27 21:15:46 arch-nas systemd[1]: external.automount: Automount point already active?
Aug 27 21:15:46 arch-nas systemd[1]: external.automount: Got automount request for /external, triggered by 3130 (3)
Aug 27 21:15:46 arch-nas systemd[1]: external.automount: Automount point already active?
Aug 27 21:15:46 arch-nas systemd[1]: external.automount: Got automount request for /external, triggered by 3130 (3)
Aug 27 21:15:46 arch-nas systemd[1]: external.automount: Automount point already active?
Aug 27 21:15:46 arch-nas systemd[1]: libpod-742b4595dbb1ce604440d8c867e72864d5d4ce1f2517ed111fa849e59a608869.scope: Deactivated successfully.
Aug 27 21:15:46 arch-nas conmon[3124]: conmon 742b4595dbb1ce604440 : runtime stderr: error stat'ing file `/external/share`: Too many levels of symbolic links
Aug 27 21:15:46 arch-nas conmon[3124]: conmon 742b4595dbb1ce604440 : Failed to create container: exit status 1

When I shutdown my system, it has to wait for 90 seconds for libcrun and libpod-conmon-.scope to timeout. Any idea what's causing this? This delay gets pretty annoying especially on an Arch system since I am constantly restarting due to updates.

All the containers are started using docker-compose with podman-docker if that's relevant.

Any help appreciated!

EDIT: So it seems like podman really doesn't like systemd automount. Switching to nofail, x-systemd.before=podman-restart.service seems like a decent workaround if anyone's interested.

Migrating from docker to podman... why?

Podman is fully open source, there is podman desktop

I just wish podman-compose wasn't so scuffed. I submitted a PR about some garbage months ago and it just seems dead.

What happens if you use regular docker-compose but with podman-docker? I've gotten better results with that.

It just bothers me that I have to use a tool outside of the ecosystem for that. Doesn't it also behave differently though? Like doesn't it assume everything is root when you use the socket required for docker-compose?

Yes, I think it does, but thats how docker-compose works with docker anyways so it's a closer experience. I think if you're using docker-compose files you sorta want that docker-like experience no?

I get where you're coming from though. It would be nice to run docker-compose files completely rootless with podman-compose.

Eh, I don't mind learning a new config if the tool requires it. I just want to define run commands in yaml and have it auto generate pods and startup scripts.

You need to mention muayyad-alsadi, because the only one who maintain it now seems only him. Just keep mentioning, and he will review the PR.

You're supposed to use podman kube play instead of that

I'm not a docker expect to be honest... but I'm pretty sure its because podman runs "rootless" while docker does not. Even if docker provides ways to mitigate that... its always great to have a "just werks" option without having to fiddle through system files, permissions and such.

Docker has a rootless capability now. I still use podman personally, but I think either one is fine nowadays. If anything, it's up to personal preference about which corporate boot tastes better.

3 years after dan walsh mentiong and opening PR and closed... docker could just merge and accept it that time.. and now they got spilt... **** docker...

Mostly for the better integration with Cockpit, but I do eventually want to play around with some of the rootless container stuff that podman has.

Mainly I looked at it to see if it would be viable for our non-techie contract developers who are forced to license Docker Desktop because they're on Windows.

Yes, $60/year/user isn't that much, however the procurement process for some of these large contracting firms will make your balls shrivel up and fall off.

Forget about podman, it will never be as polished as docker.

It's polished, just without crap of docker.

Did you even read OPs post? What about the problem and solution they described is polished?

I read it. And it's not podman wrong doing.

You need to be sure when enabling the auto start using systemd, so it's not podman fault.

Period.

1 more...