decide on node cluster options and behavior

[gregbanks]

node cluster support is enabled when passing the --cluster flag. the number of worker processes started defaults to the number of CPUs present and is configurable with the --num-cluster-workers option.

what other options do we need? do we want the master to automatically restart workers? do we want to have a command line flag to control this?

[willshulman]

If the master can auto-restart workers I guess that is cool. What is the default for Node, and what does it look like when one worker crashes?

If we do do it I think we should control with flag.

[willshulman]

For now lets just logError when child dies or fail hard (the whole cluster). Make this configurable via 'exitOnUnexpectedWorkerExit'

Then make P2 and later let's do more and have configuration around restarting and how often and which errors can cause restarts and which cannot.

[gregbanks]

sounds good, on this tomorrow

[gregbanks]

let me know how this looks

c8a015b

[willshulman]

Wow. Seems very non-trivial to implement exitOnUnexpectedWorkerExit although it also looks like you found some ways in which code I added since the first cluster impl did not properly account for cluster-ness.