Did you observe
-- any communication issues between engine and data tier?
-- Did any engine got shutdown ?
Usually incase where in replica finds engine is not communicating well and/or is un-responsive for long time, it unlocks the callstates (essentially sessions) associated with the engine so that some other engine can work on that call. During that time the callstate is marked stale and most of time you see that error when other engine picks up the call.
Your scenario might be different. What else you see in logs around this message .,