ramalama fails to detect that I have a podman machine with libkrun

[benoitf]

It happens when manually setting the type of the podman machine

Let's start a podman machine using the command CONTAINERS_MACHINE_PROVIDER=libkrun podman machine init --now

The podman machine list command is not seing the machine

ramalama/ramalama/common.py

Lines 38 to 44 in 9d02f7d

	podman_machine_list = ["podman", "machine", "list"]
	conman_args = ["podman", "machine", "list", "--format", "{{ .VMType }}"]
	try:
	output = run_cmd(podman_machine_list).stdout.decode("utf-8").strip()
	if "running" not in output:
	return None

so it won't use podman to launch the model.

That said the machine is started

podman machine list --all-providers
NAME                     VM TYPE     CREATED      LAST UP            CPUS        MEMORY      DISK SIZE
podman-machine-default*  libkrun     3 hours ago  Currently running  6           2GiB        100GiB

and if I use RAMALAMA_CONTAINER_ENGINE=podman ramalama serve instructlab/merlinite-7b-lab it works

[rhatdan]

Does the podman machine list command always fail, or only if there are two machines, one with vfkit and one with libkrun?

[benoitf]

it fails it my machine is not using the default provider ( applehv) or if libkrun is default but my machine using applehv

In short: it fails to recognize if there is a non default provider type machine

[rhatdan]

So we should just specify it in RamaLama, although@ericcurtin reported the option was not available on his laptop.

This this option come in post 5.2?

@ericcurtin what version of podman do you have installed?

[ericcurtin]

5.2.3

I think we need a bigger coding fix in general closing for now, also the container_manager() function is called 4 times in a typical run, not good for performance that level of forking. It's too late it my timezone closing.

[ericcurtin]

Reopening, meant closing the PR

[ericcurtin]

#762

[ericcurtin]

#763 reopened here and simplified the solution