This message was deleted.

I've seen the same, 'resolved' by adding an exponential restart strategy. Gives the process more time for the down task manager to spin up again

If the job size is large, JM RPC busy or frequent full GC will cause heartbeat timeout.