Login
You're viewing the techdon.dev public feed.
  • May 8, 2026, 5:07 PM

    It has been in the works for months, but the recent increase of downtime has accelerated work. We've moved from Caddy to HAProxy for codeberg.org traffic.

    Starting over with a new reverse proxy has enabled us to make choices that enable us to set up HAProxy to scale with our increasing traffic. For example, traffic is split (when possible, decided via heuristics) over several instances to load balance expensive traffic.

    We hope to share more details at a later time about this change.

    💬 1🔄 22⭐ 81

Replies

  • May 8, 2026, 7:37 PM

    You might've encountered a 504 error, which means Forgejo did not respond within a specified timeout (15s in most cases).

    When we used Caddy, we did not configure timeouts, and it would indefinitely wait for a request to finish. This is not ideal when Forgejo is under high load and gets increasingly more traffic. For HAProxy we have set a max amount of requests and timeouts to prevent (and possibly recover from) Forgejo from a state where it's overloaded and stops responding.

    💬 1🔄 1⭐ 4
  • May 8, 2026, 7:41 PM

    In the most recent case, it was unfortunately a deadlock issue causing Forgejo to not respond within that timeout. It takes more than 15 seconds to be returned as a deadlock error. Hence, you get to see the timeout error instead of a 500 error (codeberg.org/Codeberg/Communit).

    The timeout was not the issue here, although we have bumped it for POST method requests to give Forgejo extra margin to work with.

    💬 0🔄 0⭐ 3