| This can happen if a configuration reload happens while threads are trying to put quota size update diffs on an already full diff queue:
- The threads get blocked on the full queue
- The reload triggers a restart of the disk quota subsystem, with the creation of a new queue
- The old queue never gets cleaned and the old threads never get freed (there is no timeout on the queue offering)
Eventually the server will run out of threads that can be used to serve requests (to get the effect quicker, install control-flow to limit the number of requests that can actually run). |
