I have noticed that for the past couple days, when Graylog goes to rollover indexes, the connection to Elastic stops on my master node, but my secondary node keeps processing messages. During this time, messages are all sent to the journal, and Graylog is not able to run any index processing. Restarting the graylog-server service brings everything back to normal.
Graylog v2.4.3 (2 nodes)
Elastic cluster v5.6.6 (3 nodes, separate from Graylog)