Docker vs Podman why doesn't Rancher stay working in Podman?

[Greg_E]

As some of you might have seen, I'm learning (the hard way) Harvester, and to really have control over it you will want a Rancher node set up. I've fooled with a few different operating systems and not understanding what's happening, formatted and tried something different. Short of this is that I'm ending up on openSUSE LEAP Micro.

So I can get Rancher running in Podman through the terminal, after I wait about 10 minutes for it to load, great because even that took a long time to understand how much work it is doing in the background. But if I reboot the host, Rancher will no longer start.

The Podman test was after I had this running in Docker and figured out that it takes a long time and restarts several times during the process. Worked with it in Docker, decided to delete it and try and Podman. Anyway, when I run it in Docker, and restart the host, the docker container comes back up. It still takes about 6 minutes before everything is running, but at least it restarts.

For the time being this is running on a Pi4 8gb, when I have time and energy, it will be running on an HP T740 which should give it more resources. If anyone wants to give me a kidney to sell, I'll boost the ram in the T740 above the 8gb that's installed (not really enough). I ran out of kidneys getting more storage for the Harvester cluster.

 

[Ksager]

Ran into this exact same headache when I switched to Podman. The thing is, it's not that Rancher doesn't work in Podman - it's that Podman doesn't have that always-running daemon Docker has.

When you run podman run from the terminal, that's a one-shot deal. Container dies when you log out or reboot? Gone. Docker gets away with it because the dockerd daemon is sitting there in the background keeping track of everything.

What you need is a systemd service so your container comes back after a reboot. Try this:

# While your Rancher container is running, generate a systemd unit file
podman generate systemd --name rancher --new > ~/.config/systemd/user/rancher.service
# Reload and enable it
systemctl --user daemon-reload
systemctl --user enable --now rancher.service
# Make sure user services run at boot (this one's easy to forget)
loginctl enable-linger

After that, restart your machine and Rancher should come back automatically.

The key difference is Docker manages restart behavior internally, while Podman hands that off to your init system. Once you've got the systemd unit in place though, it works just as well - arguably better since you get all the normal systemd controls.

Also... yeah, you're going to want more than 8GB RAM for Rancher on the T740 if you're planning to actually use it for anything beyond playing around. That thing eats memory like candy.

 

[Greg_E]

Thanks. I've moved on to a cluster of cheapo single board n95 computers with 12gb of ram. Planning to go to k3s or rke2 and get Rancher up on that system. Also moving to openSUSE LEAP Micro for the host OS. Going to use the single command installer for Kubernetes and see where I fail to understand again, fix it and move on. This is really to push myself to learn more, and because Harvester really needs Rancher, and both run on top of Kubernetes. You can kind of see the circular thinking I'm going through, it's just taking a bunch of time.

Oddly, I see all kinds of people running these clusters on Pi4 with 8gb. I almost went that way, but the Pi's are getting so expensive. I did have Rancher sort of running on a Pi4, but I couldn't get the http gui and decided to move on a bit. Still lots of failure in my future, and hopefully learning from that failure.

 

[Ksager]

Nice, N95 boards with 12GB is actually a solid setup - way better price/performance than Pis these days. And yeah, the Pi pricing got ridiculous

A few thoughts on your path:

k3s vs RKE2 - If you're running Rancher anyway, I'd lean toward RKE2. It's basically Rancher's "blessed" Kubernetes distribution and plays nicer with the Rancher management layer out of the box. Less fighting with compatibility issues. k3s is great but sometimes you'll hit version mismatches between what Rancher expects and what k3s provides.

openSUSE LEAP Micro - Actually a smart choice for this. It's transactional like openSUSE MicroOS, so you get atomic updates and a read-only root by default. Great for immutable infrastructure, which is exactly what you want with a k3s/RKE2 cluster. The learning curve isn't too bad once you get used to transactional-update.

One thing to consider - since you're going down the Harvester → Rancher → Kubernetes rabbit hole anyway, you might look at Elemental SUSE's Kubernetes-based OS for edge/cluster stuff. It ties in nicely with Harvester if you end up going deeper down that path.

The circular thinking you're describing? That's just how this stuff works. You need Rancher to manage Harvester properly, but Rancher needs Kubernetes, and Harvester is Kubernetes... It's a lot. You're doing fine - breaking things and figuring them out is literally the job description.

Good luck with the N95 cluster!

 

[Greg_E]

I figure that LEAP Micro is about as close as I can get to SUSE Elemental as possible, while still staying in the open infrastructure. I know SUSE products get 90 days of updates, but the open side of things is probably "close enough".

 

[Ksager]

Yeah, y'all are just about spot on with that thinking. LEAP Micro is honestly a solid proxy for Elemental—same DNA, just positioned differently. The transactional nature, the read-only root, all of it translates over. You're not missing much from a learning standpoint.

One small thing worth noting, though: if you ever do decide to go full enterprise, SUSE does offer free dev/test licenses for their products. Not trying to push you toward spending money, just saying it's there if you want to poke around Rancher Premium features without the sticker shock. The free community version covers a ton, though, so no rush.

The 90-day update cycle thing, yeah, that's a consideration. LEAP Micro follows SUSE's general support lifecycle, which is actually longer than 90 days for point releases. But honestly, for homelab learning, by the time something goes EOL, you'll probably be onto the next thing anyway. And the openSUSE community tends to keep builds alive longer than the strict corporate timeline anyway.

Sounds like you've got a solid plan. N95 boards, LEAP Micro, RKE2, and Rancher on top—that's a legitimate setup. Not flashy, but it'll teach you the right stuff. Way more practical than throwing money at a T740 with 8 GB and hoping for the best.

Keep us posted on how the cluster comes together. Always cool to see folks actually learning the layers instead of just spinning up managed services and calling it infrastructure.