Comment 6 for bug 1818260

The approach we need to take is the following:

During deploy before the {leader,client}_node_is_ready returns true, regardless of the juju config value write the config file with cluster-partition-handling = ignore, then after the cluster is complete ({leader,client}_node_is_ready returns true) re-write the config with the value set in juju config.

This will avoid deploy time collisions of the rabbitmq cluster management and the charm cluster build process.

Note: The problem on a very busy cloud still exists. If the rabbitmq cluster management happens to remove a node and a juju hook subsequently we will get hook execution errors. A simple juju resolved will fix the issue.

Note:
We may need to validate {leader,client}_node_is_ready is sufficient.