If we treated timeouts as failures on the worker we would attempt to
clean up e.g. push actions while the master might still process the
event.