- Need a way to run separate workers, so that job A does not block on job B.
- Encourage daemonizing the workers, e.g. with supervisord, so they are respawned in case of failure.
- Requires both client and workers to use Net_Gearman. That defeats one of the advantages of gearman: the ability to run clients in a different language than the workers.