continue_node_clean() in ironic tree used only in IPA driver for async cleaning. In rpcapi version 1.27 this method has been converted to cast, due to slow performance of cleaning many nodes at the same time. The main problem is - we can not to do exceptions handling with rpc cast on the caller side. Proposed changes: Deprecate this method. Switch IPA driver to direct calling of this method. For example one of solution: on the start (or init) conductor manager assign own instance to "private" base interface class variable (we use one manager instance per process). Add continue_node_clean() method to deploy interface with simple default implementation. Remove continue_node_clean() from RPC API after deprecation period.
I kind of feel like I would have to see a proof of concept of the implementation to wrap my head around this right now, at least right now without coffee on my brain. I kind of think it would be viable, although I suspect we would need to pull a long deprecation period.