When running tests serially, order of nodes returned by Clusters should be deterministic

[ewencp]

I think this used to be the case and am not sure when it changed, but it's a pain when developing/debugging tests if the same node isn't used for the same thing between test runs. It would be best if, at least when running serially, the Clusters handed out nodes in a deterministic order and the test runner guaranteed they'd be used in the same order. This allows you to, e.g., stay logged into the same worker and monitor/debug a specific process without having to check which node it is running on every time.

I'm not sure why this isn't the case anymore, but it seems like it should be as simple as using an ordered/sorted collection in the Cluster classes.