Better handling of heterogenous server/task pairings

[josharian]

The following (simplified) setup goes awry:

• Start two gearmand instances A and B (say, on different ports).
• Start one worker for task T that connects to A.
• Have a client submit a job T to servers A and B.

The outcome is based on a race condition: Whichever server (A or B) replies first gets the task. If it is A, the task succeeds; if it is B, the task fails.

In general, unless all servers have the exact same set of worker tasks available, clients can't rely on requests succeeding.

This makes it difficult to rely on multiple gearmand servers for load balancing and redundancy, even if the general intent is to have all workers connect to all servers (config errors, connection flukes, etc. can get in the way -- exactly what the redundancy is there to help with).

[josharian]

I'd be happy to take a crack at fixing this, but before I start, I just wanted to make sure that I wasn't Doing It Wrong, and that there wasn't a different intended way to use multiple gearmand servers for redundancy.