There is a default master_timeout of 30s, we are running over that at time preventing index creation. This has happened both when creating the indexes for a new wiki and again today creating indexes for the titlesuggest feature. We can override the timeout by setting master_timeout in the query string, this is not supported by the Elastica client library we use in all instances.
Ideally we could fix this by making the cluster respond in < 30s, but i fear we might need to make the timeout adjustable until then.