DEV Community


alias docker=podman

Job Cespedes
Originally published at ・4 min read

I have heard about podman (Pod Manager tool) more and more often now. Whether it is that I have come closer to its developing environment or that it has come to mine, I'm not sure. It's both, I guess. I use Ansible a lot for automating baremetal and virtual infrastructure: for its definition, deployment, configuration, operation, among other things. In the recent years, I have being using more and more containers, particularly in the developing stages. And I really like open source, most of its concepts, and many initiatives around it. So, I have being wondering if I should be using mainly podman for containers already. There is, however, this one recurrent question in my head:

How much work will adapting those docker projects take?

Why that unionfs container?

I use Ansible a lot for automating infrastructure. It works well for having multiple stages/environments over that infrastructure. However, managing those multiple stage/environment directories sometimes become kind of messy. Demo-multienv looks for a way to avoid it. It uses unionfs inside a container to stack directories/files, reducing data and file duplication among them. You only need docker and its python library installed and Ansible will do it for you. But, could it be run with podman instead?

It was just docker. I wanted to add compatibility with podman also. I was hoping it would be little extra work, something very simple, just like alias docker=podman for cli. But it required more than that.

Ansible and podman

Docker has docker_container and I was kind of hoping to just change it to podman_container and mission accomplished. But I found that there is no such module included with Ansible, yet. There is WIP for one, though. And, there is available a Tripleo module. That is the one I tried. It has some different parameter names but its usage is similar overall. It worked perfect in this particular case. The playbook ran successfully.

The unexpected

Then, it came something unexpected. Files supposed to be exposed from the container to the host, were not there and neither was the mount point. The container successfully started. However, there were no files in the union host's directory. In the container, the files were there and a mount point for the unionfs as well.

By this point, I was running podman manually (not using Ansible and demo-multienv). Debug option for Ansible and Tripleo module helped me here. It shows the podman command being used. I copy-pasted it and added a debug level, using: podman --log-level=debug run. There were no errors. It just had a warning:

WARN[0000] Failed to add conmon to cgroupfs sandbox cgroup: error creating cgroup for cpu: mkdir /sys/fs/cgroup/cpu/libpod_parent: permission denied

However, I could not related it with my problem. podman logs did not show anything wrong. I was clueless, honestly. I just knew that with docker it worked 😜

Community to the rescue

I turned to the community, opening #5322 --after checking previous related issues/discussions of course πŸ˜„--. That was how I got to the cause of the problem and learned about other important concept around podman. One particular answer sums it up:

"this is not a bug. Rootless containers cannot create mounts in the host.

The shared mount is created inside of the rootless mount namespace, you can reach it with podman unshare" -- @giuseppe . From Github

Everywhere in podman documentation, it says it is able to run rootless. And still, I forget that aspect πŸ˜…. In docker, it was running as root. For it to work in podman:

  • It should be run as root

However, podman still allows one option for running it rootless. It is related to user namespaces and podman unshare.

  • If rootless, it could be accessed using podman unshare

For example:

# rootless
## prepare dir layout
mkdir -p $dir1 && echo "dir1_var1" > ${dir1}/var
mkdir -p $dir2 && echo "dir2_var1" > ${dir2}/var
mkdir -p $unionfs
cd $parent

# bind-propagation of unionfs dir inside 'podman unshare' environment
podman unshare mount --bind --make-shared ${unionfs} ${unionfs}

# run container
podman run --name multienv_unionfs --privileged=true --rm=true --env TZ=America/Costa_Rica --env PGID=1000 --env PUID=1000 --env UNION_DIRS=/dir2=RW:/dir1=RO --env MOUNT_PATH=/unionfs/ --env COW=true --volume ${dir2}:/dir2 --volume ${dir1}:/dir1 --volume ${unionfs}:/unionfs:shared --detach=true jobcespedes/multienv

# check that files are there inside 'podman unshare' environment
podman unshare ls -al ${unionfs}
# check that mount point is there inside 'podman unshare' environment
podman unshare mount | grep ${unionfs}

More than alias docker=podman

In retrospective, replacing docker with podman may require more than alias docker=podman. It required me to:

To be fair, in many cases the alias could be all you need. But in this particular one, those steps helped replacing docker with podman and extending demo-multienv. Now I could run it in either one of three modes using Ansible πŸ˜„: docker, podman or unionfs on the host.

Discussion (0)