ZeroMQ driver fails to send casts if processing during exit

[Erica Windisch]

The ZeroMQ driver only completes a cast() when a remote host's olso-zmq-receiver process consumes a message, or if the cast() times out (via rpc_cast_timeout or the per-cast Timeout kwarg).

Because cast() is spawned inside a greenthread, Python will happily exit with unsent messages.

At a minimum, impl_zmq's cleanup() method should wait for all cast() messages to send or timeout.

[OpenStack Infra (hudson-openstack)]

Fix proposed to branch: master
Review: https://review.openstack.org/19334

[Erica Windisch (ewindisch)]

Note that this doesn't have much impact on long running tasks such as services, but is apparent in short-running ones such as command line rpc clients. There aren't any of those in common use, I caught this using my own rpc client which I put together to test another bug.

[Mark McLoughlin (markmc)]

Patch was abandoned 2 months ago with "Better implementation to follow"

[Oleksii Zamiatin (ozamiatin)]

If you find a similar bug in new driver please report a new one
https://review.openstack.org/#/c/223877/