Run own woodpecker agent?

[chrysn]

When Codeberg CI load is high, CI often runs into timeouts (especially around observing expected CoAP behavior w/rt piggy-backed responses).

One way around would be to have agents on own CI agents on own hardware. I'm not really set up for docker on my machines, so exploring what is needed for sensible operation:

• new user, with own group and sufficient subuids / subgids
• from Support "native" podman as backend woodpecker-ci/woodpecker#85 (comment), "Run systemctl --user enable --now podman.socket which creates /run/user/$UID/podman/podman.sock (in my case, UID=1000)."
  • note that doing anything systemctl-related is best not done through sudo su but through sudo machinectl shell woodpecker-agent@ /bin/bash
• ensure podman-docker is installed
• download woodpecker-agent from their release files
• create a token in the woodpecker settings of the organization
• run: WOODPECKER_BACKEND_DOCKER_HOST="unix:///run/user/$UID/podman/podman.sock" WOODPECKER_GRPC_SECURE=true WOODPECKER_SERVER=grpc.ci.codeberg.org:443 WOODPECKER_AGENT_SECRET='NotTellingYou==' ./woodpecker-agent --backend-engine=docker --backend-docker-ipv6

… which gets me to the point where it attempts to start containers, but then runs into ""Error response from daemon: runc: runc create failed: unable to start container process: can't get final child's PID from pipe: EOF: OCI runtime error"

[chrysn]

Ah, what it needed was installing crun and writing in ~/.config/containers/containers.conf:

[engine]
runtime = "crun"

(at least to the point where something happens, as in: the logs show at least cloning run up to git-lfs, at which point it should be complete, but no subsequent step is taken)

(and maybe it also helped to move the the host to --backend-docker-host=; there were some hints that the environment variable might not work, but I didn't drill down on that)