[FLUME-150] restart of node in multi-master setup creates LOST node when reregisters with new master

• Type: Bug
• Status: Resolved
• Priority: Major
• Resolution: Won't Fix
• Affects Version/s: 0.9.0
• Fix Version/s: 0.9.5
• Component/s: Master

Environment

linux

• Assignee: Unassigned
• Reporter: Disabled imported user
• Created: 12/Aug/10 16:06
• Updated: 05/Nov/14 07:29
• Resolved: 05/Nov/14 07:29

Description

I have a 3 node flume-master setup that is backed by an external 3 node zookeeper cluster.
flume-node registers on flume-master node A. If the flume-node process is restarted/box rebooted/whatever, when it comes back up and it registers on flume-master node B, node A shows a LOST node even though the node is registered on node B.

Screenshots I'll attach:
1) beforeRestart.png shows 'cdh3collector3 registered on master (server id:1)
2) afterRestartS0.png shows 'cdh3collector3' registered on new master (server id:0)
3) deadNodeS1.png shows the old registration for 'cdh3collector3' as not seen in a long time on old master (server id:1). Eventually this changes to 'LOST'

It seems that the old registration should disappear if the node reappears on a different master node.

Issue Links

• is duplicated by: FLUME-108 Restarting a Node doesn't remove it from prev masters list