Bug: neutron-server lock (when one compute node is down?)
Even if both the nova-compute and neutron-plugin-capnet-agent agents are marked as administratively down, a "down" compute node seems to cause neutron-server to lock. It seems that this is likely due an RPC (to the down node?) happening during a DB transaction, and hangs (this would require the RPC to be synchronous; we have some of those). There are also some internal locks we use, but not many; it is unlikely to be those. In any case, this is non-deterministic and the logfiles do not help.
I don't see any RPCs inside DB transactions started within the capnet server-side code; but if the session is begun outside the Capnet code, who knows. I don't recall that the rules for situations like this are really spelled out.