Due to an abnormally slow switch between the primary and replica of our database clusters, the worker machines that the host actor runs got restarted. The accumulated workload caused some runs to be delayed by about 3-5 minutes. Runs continued normally with no data loss but runs with a low timeout may have timed out.