Recently we have seen problems with HTTP timeouts when interacting with a GitLab instance. These timeouts seem to sometimes cause lingering lock files in the Git repository in the server, which then blocks all write operations on the repository until an admin can remediate. When this happens for the mlbridge bot repo where we store mbox files tracking what emails have been sent for what PRs, the bot stops being able to send more emails.

An admin for this server said that increasing the HTTP timeout might alleviate this issue and make it less likely to trigger. We currently have a timeout of 30s for all HTTP requests, and they recommended increasing this to 60s. I don't think we want to do that across the board, but perhaps for a certain subset of requests, which would at least cover PUT and POST operations on files for GitLab hosts.