Starting multiple tasks where one fails to launch results in tasks running with no manager

[ollieparsley]

Starting multiple tasks where one fails to launch results in zombies. If there are tasks A, B and C. Task C has incorrect permissions. When starting the wp manager tasks A and B are started correctly, task C fails to start. The wp manager then exists straight away and leaves all instances of A and B running.

As a consequence when restarting the wp manager there are messages about receiving keep-alives for workers that are not registered.

So I think the options are these:

• Workers die when they don't have a manager (not sure this is feasible)
• Don't start workers until we are more sure all tasks will start
• Kill old tasks if they are running, on wp manager restart
• Adopt old tasks and allow the manager to re-balance to get the correct cardinality