Comment 4 for bug 1658913

@lihong, could you please check my fix https://review.openstack.org/#/c/429639 with your deployment. Does it help?

As I understood the root cause is that eventlet zmq forces zmq.NOBLOCK flag (btw you can see this constant in the stack trace) and in such case we need to handle zmq.Again exception. I respectfully disagree to rely on time.sleep in production because sleep doesn't guarantee that connection will be established in that specific time and we have probability of failure anyway.

Thanks for finding this bug, hope to hear that fix helped. Feel free to ping me if it doesn't.