Clients stop when receiving unmatched RPC response

[tilsche]

If a client receive an RPC response to its former self, it will throw an RPCError and log "error in rpc: no function but also no response callback". That seems sensible, but apparantly this can happen and cause a restart loop.

[phijor]

Shouldn't we simply discard such a reply as invalid? Anyone^{me, stupidly} can issue invalid RPCs and clients really shouldn't crash because of that. Log a warning and proceed as normal.

[bmario]

Probably, the problem is that the manager has a backlog of unprocessed RPC messages. In the meantime, the client disconnects for some reason and reconnects under the same token later. Therefore, a new RPC queue with the same name is created. Once the manager processed the old RPC, the recipient is already gone, but there is a new one under the same name.

Maybe we should think about adding UUIDs to RPC queues or tokens like a session id.

[bmario]

One more data point for adding UUID's to RPC queues: metricq/metricq-python#25