Hystrix Command should be cancellable

[panterose]

At the moment, if you call .queue() on a command and get a future, you cannot call the cancel() method and abort its execution. This is true for the current 1.3.XX version (it seems to work if you don't use the cache for some reason) but also in the upcoming 1.4 version.

I can supply a reduced code sample if needed, but it's very easy to reproduced. It's an issue for me as we have some long running command.

We also have the case where we encapsulate child command inside main command, ideally the cancel() should be cascading.

[benjchristensen]

I think I know what is causing this, it's the decoupling done by the Subject used because caching necessitates multicasting. I'll need to think about supporting multicasting and unsubscribe at the same time.

[benjchristensen]

Putting into 1.4.x to consider, but this remains a tricky thing when caching is supported. It would have to behave like Subject.refCount() where it only cancels the original HystrixCommand if all consumers cancel/unsubscribe.

[mattrjacobs]

Talked to @benjchristensen about this. Since the semantics are slightly different than Subject.refCount(), it will take some time/care to do this properly. As a result, targeting this for 1.4 series, but not gating RC7 on it.