6:13 a.m.
Hi François,
Thanks for reaching out.
On Fri, Nov 4, 2022 at 10:34 AM POIROTTE Francois <
francois.poirotte(a)csgroup.eu> wrote:
Hi,
I'm working with a pod where 2 containers depend on one another. I'm
using podman v4.3.0 and the pod is started using systemd
(podman-kube@.service).
When one of the containers is in an improper state (crashes, stops
unexpectedly, or some other condition is not met), I would like to
restart the whole pod automatically (just restarting the one container
that stopped is not enough in my case).
Ideally, systemd would be in charge of restarting the pod (using the
Restart= directive to control precisely when it must be restarted).
So far, I've tried using the healthchecks mechanism, but while podman
correctly detects that the container is unhealthy, no further action is
taken.
Is there any way to achieve the desired result?
That is a very cool use case, thanks for elaborating on it. At the time of
writing, it is not possible with Kube. The restart policy in the
Kubernetes YAML only relates to the containers only but not to the entire
pod. Hence, when container A has a failure and gets restarted, other
containers won't be touched. It would be nice to integrate health checks
into the Kubernetes support in the future. Feel free to open an issue on
github.com/containers/podman for that.
But there *may* be a solution (I see space for improvement though) when
using Podman pods outside the context of `podman kube play` and
Kubernetes. It's a big complicated at first glance, so I'll go through
step by step:
- `*podman generate systemd --new*` can create systemd units for a pod
and its containers.
- `*podman pod create --name=foo --exit-policy=stop*` creates a pod
called "foo" that will "stop" once all containers have exited. We
can
exploit that in the systemd unit.
- `*podman create --pod=foo [...]*` creates a container inside the pod
"foo". This way, you can add your two containers to the pod.
- Starting with Podman v4.3, there is a cool new feature called
"on-failure actions" for health checks [1]. These allow, among other
things, to kill a container once it turns "unhealthy".
- So it's quite easy to deal with container A. How to deal with
container B really depends on the use case. Maybe there's a way to detect
inside container B if container A is still doing its job (e.g., via health
check) and do the same trick. `*podman create` *has a `--requires`
flag but that only works when *starting* containers but does not take
stopping or restarting into account yet. I am pulling in @Matt Heon
<mheon(a)redhat.com> to discuss this use case.
- If you want to run these things in systemd, you can use `podman
generate systemd --new foo`. In that case, you should use the "kill"
on-failure action. If you want to run it outside the context of systemd, I
recommend using the "restart" action.
I hope this gives you enough pointers for a working solution.
Kind regards,
Valentin
[1] https://github.com/containers/podman/commit/aad29e759c78
Regards,
François
_______________________________________________
Podman mailing list -- podman(a)lists.podman.io
To unsubscribe send an email to podman-leave(a)lists.podman.io