Anyone running Harbor successfully using Podman?

[ColumbusAI]

I'm curious if this is possible as I would like to support open-source and move from Docker to Podman if possible with Harbor.

[av]

@FrantaNautilus spent quite a bit of time on certain aspects of this, see for reference: #101

In short, it's almost possible right now. We need a tiny capability detector and/or customization for docker and docker compose binaries - all possible and within reach. The largest drawback is that it's highly likely not all features will work, depending on Podman's Compose feature compatibility with the OG Docker Compose

[FrantaNautilus]

Allow me to summarize what I tried with Podman previously, just in case it helps you with testing.

1. Installed podman and podman-compose.
2. I needed an alias for podman <-> docker. Created a script calling podman named docker in ~/.local/bin/. On Nix OS the alternative is to set virtualisation.podman.dockerCompat to true (although it does mostly the same as the first option).
3. Created a local branch of harbor, where I removed the flag --wait from the harbor.sh. This flag is currently unsupported in Podman, see Add support for podman-compose up --wait option containers/podman-compose#710.
4. Assuming that one is using Nvidia, here is the part where one would need to configure CDI in harbor. See https://github.com/av/harbor/wiki/3.-Harbor-CLI-Reference#capabilities-detection.
5. Here I was able to start the Ollama with Open Webui and run and LLM on CPU (at that point in time I was not sure how to make CDI work).

I did not test this setup any further and I currently lack time to experiment with it (and everything works fine on Docker). One thing that was a slight problem was that Podman kept outputting warning messages, that podman-compose is being called with external command, although this should be possible to disable this warning in the config file.

[av]

Thank you so much for such a detailed write-up! 🙌

[ColumbusAI]

Ditto, amazing write up! Appreciate your efforts.