Currently, clone instances are created for every node in the cluster, regardless of whether it is possible for any instance to run there. In particular, bundle nodes will have stopped instances of non-bundled clones, and non-bundle node will have stopped instances of bundled clones.

Worse, this is affected by the order in which resources are unpacked. If the configuration has clone1 then bundle1 then clone2, clone2 will have stopped instances for bundle1's nodes, but clone1 will not. This makes it unpredictable, and higher-level tools can't intelligently hide stopped instances.

A likely solution would be to ensure bundle nodes are created before any clone instances. Ideally, we wouldn't create non-bundled clone instances for bundle nodes or vice versa.