Being mentioned in Wochennotiz / weeklyOSM 339 lead to quite some additional traffic (which is good), but unfortunately also revealed an issue that could only surface with enough parallelism. This lead to random rendering failures, and eventually to a total stop of the rendering daemon on Tuesday and parts of Wednesday.
This has hopefully been resolved now, the rendering daemon has finished all pending jobs in the queue, and I manually restarted those jobs that were listed as “canceled by user”, knowing that at least some of these had actually been killed by the system instead.
Everything seems to be humming along fine again now after moving the Django database, which manages user sessions and the rendering request queue, from SQLite3 to PostgreSQL, with one exception:
I can see one rendering job that failed as it requested the surveillance overlay in combination with multi page output. This is a bug (or actually a known, but unhandled, omission) that I’ll hopefully be able to fix later today.