Set deleting exchange status

[dcorbacho]

• Avoids race condition between declare and delete

Solves rabbitmq/rabbitmq-federation#7

[michaelklishin]

Oops, my comments ended up being on a commit in a fork.

I suggest that we rename exchange-delete to exchange-delete-in-progress. This is a fairly unexpected use of runtime parameters but no different in practice over using a separate local table, so why not.

Also, have we considered what happens when a node fails right after storing the parameter, then coming up? Runtime parameters are stored in a durable table. Should we perhaps clean up all such parameters on boot?

[dcorbacho]

Just renamed the parameter.

Good point in the persistence, you're right, I will look at it. Maybe the runtime-parameters are not such a good idea? I used that because was the most similar structure already in place.

[michaelklishin]

It is indeed a similar structure and it is already scoped per vhost, just like what we want. So I'm OK with using it.

[michaelklishin]

There are test failures in the Java client (exchange.delete, exchange.declare tests). I'm investigating.