ramalama pull/run hangs when host uses IPv6

[puttup]

This could be a "me" issue in the sense networking is broader than AI, but I'll report in the interest of not needing to configure the host, and for others to find.

ramalama --debug pull deepseek-r1:1.5b wouldn't complete on my system, and didn't print any messages.

Setting sudo sysctl net.ipv6.conf.all.disable_ipv6=1 allowed ramalama to pull successfully.

Looks like Ollama registry doesn't support ip6: ollama/ollama#2216. Huggingface didn't work for me either tho I've read they are dual ip4/ip6.

Is this something ramalama can gracefully handle, or is there some general config I should set on my machine (Fedora 41 Workstation) that's outside of ramalama's responsibility (eg disable ip6)?

[ericcurtin]

I see this as more of an Ollama registry issue :) But if someone opens a PR to prefer IPv4, no complaints. I think a global to force IPv4 could be too much, because some internal networks may want to use IPv6, etc.

I think we could force IPv4 for https://registry.ollama.ai (and huggingface if we prove that is also IPv4-only) though. But we should not kill IPv6 support for RamaLama.

Hope this makes sense.

[puttup]

Oh totally! I was more thinking fallback to ip4 if 6 falls. Providing a timeout in urlopen worked in my little test script but I didn't try this in ramalama