Read create_infra from containers.conf when kube play .yaml

[PiotrBzdrega]

When creating pod using podman pod create --infra=false we have possibility to avoid creating infra container.
As far i know there is no counterpart command in .yaml file.
In such situation i would suggest to create in containers.conf:

#Used to overwrite "CreateInfra" property during podman kube play .yaml command   
create_infra = true

[Luap99]

AFAIK k8s always uses a infra container and the kube command tires to match k8s behavior wherever possible as long as it makes sense for podman. And I think there is important functionality that requires a infra container to be created for the pod, i.e. we need a container running to keep namespaces open that can be shared for all containers in a pod.

What exactly is your use case?

[PiotrBzdrega]

@Luap99 Thank you Paul for reaction,
I thought that infra is not so inevitable if there is cli --infra=false 😀
I read in "Podman in Action" ~Daniel Walsh that

The infra container (pause container) is similar to the rootless pause container; its only purpose is to hold open the namespaces and cgroups, while containers come and go. However, each pod will have a different infra container

In my pod i won't change number of running containers ( don't need to add or remove during runtime).
I assumed that this is main task for infra, to let new container join namespace/cgroup.

But the main (dumb) reason ... is that i would like to filter out infra container during checking state of containers and have it short to not process it a lot 😅.

Additional side effect to consider what will happen if we would use this configuration variable durnig podman kube play:
Working with the current version (5.3.0), if the .yaml file is broken, pod will be created with only one infra container with pod status = created.
How podman will behave if there won't be dedicated infra 🤔, Will pod be created with 0 containers or maybe nothing will happen like with broken compose file in docker - compose

[Luap99]

Well without the infra the containers will not share any namespaces which is not what most people want. In that case they are just a group of containers where podman pod start/stop starts and stops the containers at the some time.

But the main (dumb) reason ... is that i would like to filter out infra container during checking state of containers and have it short to not process it a lot 😅.

I don't know what you are doing but there is a IsInfra field in the inspect and list container output that can be used to filter.

$ bin/podman ps -a
CONTAINER ID  IMAGE                                        COMMAND     CREATED        STATUS                    PORTS       NAMES
2b6ad4da01c0  localhost/podman-pause:5.4.0-dev-1731674600              6 minutes ago  Up 3 minutes                          a7f302810edb-infra
a940f402b017  docker.io/library/debian:bookworm            bash        3 minutes ago  Exited (0) 3 minutes ago              magical_hamilton
$ bin/podman ps -a --format '{{if not .IsInfra}}{{.State }}{{end}}'

exited

Additional side effect to consider what will happen if we would use this configuration variable durnig podman kube play:
Working with the current version (5.3.0), if the .yaml file is broken, pod will be created with only one infra container with pod status = created.

Yeah I think this is pretty much undefined behavior. I guess on errors we should clean the pod up again which the code seem to say as well

podman/cmd/podman/kube/play.go

Lines 329 to 343 in e195177

	if playErr := kubeplay(reader); playErr != nil {
	// FIXME: The cleanup logic below must be fixed to only remove
	// resources that were created before a failure. Otherwise,
	// rerunning the same YAML file will cause an error and remove
	// the previously created workload.
	//
	// teardown any containers, pods, and volumes that might have been created before we hit the error
	// reader, err := readerFromArg(args[0])
	// if err != nil {
	// return err
	// }
	// if err := teardown(reader, entities.PlayKubeDownOptions{Force: true}, true); err != nil && !errorhandling.Contains(err, define.ErrNoSuchPod) {
	// return fmt.Errorf("error tearing down workloads %q after kube play error %q", err, playErr)
	// }
	return playErr

So I am not sure why this isn't done today or what the problems with that are.

[github-actions]

A friendly reminder that this issue had no activity for 30 days.

[Luap99]

I close this as won't fix, the k8s behavior requires a infra container