Let rmi -af remove pods as well

[vrothberg]

podman rmi -af is able to remove containers but is NOT able to remove pods which I think is an inconsistency. If all containers would be removed, we can nuke the pods' infra containers as well.

@containers/podman-maintainers PTAL. If there's consensus I would like to change that behavior before 4.0.

~ $ podman rmi -af
Untagged: docker.io/library/alpine:latest
Deleted: c059bfaa849c4d8e4aecaeb3a10c2d9b3d85f5165c66ad3a4d937758128c4d18
Error: error removing image ed210e3e4a5bae1237f1bb44d72a05a2f1e5c6bfe7a7e73da179e2534269c459: container 8b32ea2a4c2fc8972d694b87c10ac4df7bd6e7a0ffec0aec8b32e10bdf299c05 using image could not be removed: container 8b32ea2a4c2fc8972d694b87c10ac4df7bd6e7a0ffec0aec8b32e10bdf299c05 is the infra container of pod 6364bc5b0b40d7b9cff4d4e6b337b29cff83d207d41dc1e966512b6aef53b8f9 and cannot be removed without removing the pod

[Luap99]

The same applies to podman rm -af

[mheon]

Is it? If I'm a consumer of the remote API via Python and my single call to delete a container removes 5 containers and the pod they're a part of, that seems like something the API should inform me of. The other APIs aren't specific to containers, so the potential for removing >1 container is always there (more than one container could be using a given image, or network), but container rm has always removed only a single container, without side effects.

[vrothberg]

I totally agree and think the remote API should remain untouched but we can handle it on the client side in pkg/domain/infra/{abi,tunnel}:

if ctr.IsInfra() { RemovePod(...) }

[mheon]

Oh, I see.

I was actually advocating for the remote API to be changes to return an array of removed containers, not just one, so we can actually remove the entire pod but report that we did so - but this also works.

[vrothberg]

Since the remote API returns an array now, I will take a stab 🔪

[vrothberg]

https://github.com/containers/podman/pull/12826/files

[Luap99]

Also podman network rm -f netname will remove the pod and does not error, see:

podman/pkg/domain/infra/abi/network.go

Lines 88 to 99 in b01a421

	if c.IsInfra() {
	// if we have a infra container we need to remove the pod
	pod, err := ic.Libpod.GetPod(c.PodID())
	if err != nil {
	return reports, err
	}
	if err := ic.Libpod.RemovePod(ctx, pod, true, true, options.Timeout); err != nil {
	return reports, err
	}
	} else if err := ic.Libpod.RemoveContainer(ctx, c, true, true, options.Timeout); err != nil && errors.Cause(err) != define.ErrNoSuchCtr {
	return reports, err
	}

[vrothberg]

Excellent sales pitch! Changing rmiand rmwould make the commands consistent

[TomSweeneyRedHat]

WDDD? (what does Docker do?)

[giuseppe]

I guess they don't have this problem since there is no concept of "pod"

[vrothberg]

I concur. Docker doesn't support pods, so they don't need to to worry about them.

[rhatdan]

I agree and I think this is a good idea. --force means --force.

[vrothberg]

Great. I'll take a stab at it.