[CAMEL-11457] camel-atomix - No new leader when all nodes are killed forcefully

• Type: Bug
• Status: Closed
• Priority: Minor
• Resolution: Won't Fix
• Fix Version/s: 2.20.0
• Component/s: camel-atomix
• Estimated Complexity: Unknown
• Assignee: Luca Burgazzoli
• Reporter: Nicola Ferraro
• Created: 26/Jun/17 17:31
• Updated: 12/Sep/17 06:59
• Resolved: 12/Sep/17 06:59

Description

I'm testing the leader service with the following scenario.

Client mode with an external bootstrap service on Openshift. Using a spring-boot application.
The service configuration is:

AtomixClusterClientService service = new AtomixClusterClientService();
service.setId(InetAddress.getLocalHost().getHostName());
service.setNodes(Collections.singletonList(new Address("atomix-boot-node", 8700)));

Steps:

• I start 3 pods of the application, one is the leader.
• Once started, I kill forcefully all three pods (calling "Runtime.getRuntime().halt(1)" from the JVM code) one after the other at short distance (few seconds)

When all three pods become available again, the "leadershipChanged" callback is not called in any of the pods (waited > 1 hour). If I restart one pod after some time, that one become the leader. The other two pods receive the notification that there's a new leader.

It seems that a timeout occurs, so that a new leader can be determined upon restart, but nobody participate in the election if pods are started before the timeout.

lb any idea?