No description
- Python 60.1%
- Nix 35.3%
- Shell 4.6%
01dbe2b18d
Some checks failed
T3 Code image verification / verify (push) Failing after 14s
T3 reads `git remote get-url` to decide which server a repository lives on and compares that host with the host on each pull request link. Links carry the public host, so rewriting Git to the in-cluster service left T3 looking for a host no link uses: it logged "Change requests cannot be browsed for this project's host yet." and skipped pull request sync even though its Forgejo provider lookup kept working. Resolve the public name to the pod instead and relay it to the Forgejo HTTP service from runtime/forward-forgejo.py on ports 80 and 443. Port 443 presents the certificate cert-manager issued from the existing cluster issuer, because `fj` builds https for a bare hostname and Forgejo advertises https clone and LFS URLs. Everything stays in the namespace: the relay, the bootstrap and the credential helper come from the runtime ConfigMap, so no image rebuild is needed, and cluster DNS and the gateway are untouched. Start the relay before the logins that depend on it, retry a failed login instead of crashing the container, and drop the rewrite, the HTTP/1.1 workaround and the second `fj` login that only papered over the mismatch. |
||
|---|---|---|
| .forgejo/workflows | ||
| cluster | ||
| docs/runbooks | ||
| hosts | ||
| images | ||
| lib | ||
| monitoring | ||
| .gitignore | ||
| .sops.yaml | ||
| AGENTS.md | ||
| AUDIT.md | ||
| deploy.sh | ||
| flake.lock | ||
| flake.nix | ||
| LICENSE | ||
| renovate.json | ||
| secrets.yaml | ||