timeouts?

[lloyd]

shall the client be able to cancel enqueued work? or specify a timeout - the max time it should take?

[lmorchard]

This seems pretty important. Looking into using compute-cluster for kumascript, and child processes have been known to die in production for initially mysterious reasons (ie. eventually discovered bugs, resource limits, etc).

The workaround I'm looking at right now is to start a timeout before enqueue, and to just bail out if the enqueue callback never gets called before the timeout fires. Workable, but is a bit awkward.