Not tested (yet) patch (it still needs integration tests). Let me know what you think.

Added integration test and it works on EC2.

+1 Looks great.

A few comments, which we might address in follow-on issues.

• Have you thought about using Avro/JSON as the serialization format? It would then be easier to evolve.
• Have a parser object that has public methods that are easier to test than the ones on ClusterStateStore. They don't really belong there as the ClusterStateStore could use any representation it chooses.
• Rather than use NoopClusterStateStore, the default might be in-memory (which you implemented elsewhere)?

> Have you thought about using Avro/JSON as the serialization format? It would then be easier to evolve.

I believe the current format is good enough for now. I suggest we should change it later when the state data model evolves.

> Have a parser object that has public methods that are easier to test than the ones on ClusterStateStore. They don't really belong there as the ClusterStateStore could use any representation it chooses.
> Rather than use NoopClusterStateStore, the default might be in-memory (which you implemented elsewhere)?

I will address these two issues in one step later while improving code testability.

I want to commit this but unfortunately I'm still seeing some errors (even if the integration tests are passing):

endpoint.getHost() is null for https://whirr_test.s3.amazonaws.com/whirr-zookeeper

Any ideas? I will investigate more to get it fixed.

Updated patch with some fixes. Unfortunately it still doesn't work. Adrian, any ideas?

I was using an invalid container name. I've added code for strict format checking (thanks Adrian for help) and I've been able to start a cluster using blob store state persistence. I'm going to commit this.

+1

I've just committed this. Guys thanks for helping and reviewing.