Support forking of workers (rake jobs:work NUM_PROCESSES=3)

[wkonkel]

Especially useful for running multiple workers on a single Heroku dyno worker.

[coveralls]

Coverage increased (+0.01%) when pulling 6419e25 on wkonkel:master into 812f10b on collectiveidea:master.

[albus522]

Unfortunately it is not quite that simple. What happens when 1 or all of the workers die? Are signals properly handled in the forked processes?
The master process probably needs to continue running or heroku will think it finished and kill the dyno.

[wkonkel]

The master process has a "sleep" in it (line 21), so Heroku doesn't think it finished. However, your other points are valid. If I have some free time, I'll improve this pull request with:

• Master process should periodically check that child processes are still around and if not, re-fork.
• Master process should gracefully respond to SIGTERM. Currently the TERM is passed down to children who gracefully exit but the master process exits abruptly with a stack trace.

That said, we're successfully using this patch in our production environment and it's been working very well for us. We have the same worker throughput with 1/4th the dynos.

[dgobaud]

This is really helpful. I added code to monitor and re-fork the children #615