Long-running actions

This has come up in a number of discussions.

Here's a sketch of an idea for supporting long-running actions.

Step 1: Stand up a "compute farm" (ideally elastic) that can run docker containers, such as a Docker swarm, mesos, kubernetes, or even a vendor's container service.

WLOG, let's suppose the farm is a swarm cluster.

Step 2: Write a "swarm" package with actions that can start a container in the swarm, rm a container, and monitor the output or logs from a container running there, perhaps by polling.

Step 3: Dogfood our system to dispatch and monitor long-running whisk actions by delegating the heavy lifting the swarm package.

Step 4: Profit! (couldn't resist)