RabbitMQ 3.x queues are not synchronized

[Dmitry Borodaenko]

Even though OpenStack creates all queues synchronized across all controller nodes, after RabbitMQ is restarted on a node, queues are no longer synchronized to that node.

See comments #40-#45 in https://bugs.launchpad.net/fuel/+bug/1285449 for more details.

Update: the problem is only present with RabbitMQ 3.x.

[Dmitry Borodaenko (angdraug)]

Correction: the above only occurs with RabbitMQ 3.2.4, downgrading back to RabbitMQ 2.8 causes the queues to synchronize immediately after failback.

[Dmitry Borodaenko (angdraug)]

HA for queues is configured differently with RabbitMQ 3.x:
https://www.rabbitmq.com/blog/2012/11/19/breaking-things-with-rabbitmq-3-0/
http://docs.openstack.org/high-availability-guide/content/_configure_rabbitmq.html

[Dmitry Borodaenko (angdraug)]

Running the following command on one controller is enough to ensure all queues are mirrored:
rabbitmqctl set_policy HA . '{"ha-mode": "all"}'

[Dmitry Borodaenko (angdraug)]

This problem is addressed as part of this change:
https://review.openstack.org/#/q/Iee79f14929e9f1ac5e5e7dd07aee37f641d3b2a8,n,z

The fix for 5.0 was merged and backport of the fix for 4.1.1 is currently under review.

[Bogdan Dobrelya (bogdando)]

Currently we have this ha-policy mode in the Rabbit's (v3) init.d script. Is it enough? What the status of this issue now?

[Dmitry Borodaenko (angdraug)]

As you can see from the link in my previous comment, the backport of that change to stable/4.1 hasn't been merged yet.