I am experiencing some trouble with the communication between HAProxy and Neo4j when I test the failover of a replica in my Neo4j HA cluster. In my testbed I have 5 machines. 2 machines run Neo4j (1 Master and 1 Slave). 1 machine runs the Neo4j arbiter, 1 machine runs the HAProxy and the last one is used to run the client.
My experiment starts with both Neo4j Master and Slave up and running and the load is even distributed by the HAProxy to the 2 Neo4j nodes. When (for testing purposes) I kill the Neo4j Slave node, the HAProxy correctly redirects all the traffic to the Master node. Now, when the Neo4j Slave node is restarted (after a while) it is able to join the cluster and the HAProxy detects it is available again (I can see this on the proxy monitoring page) but all the load is still sent only to the master node, that is the load is not redistributed on the new replica even if it is available again. This behavior happens especially when the client node generates high load. In this case the only way I have to redistribute the load on the 2 replicas is to kill and restart the HAProxy using the command: "killall haproxy && haproxy -f haproxy.confg" any idea on what is the of cause this problem? The versions I am using are Neo4j 2.0.1 and HAProxy 1.2.24. best regards -- You received this message because you are subscribed to the Google Groups "Neo4j" group. To unsubscribe from this group and stop receiving emails from it, send an email to [email protected]. For more options, visit https://groups.google.com/d/optout.
