Description
Nodetool Drain is documented at http://wiki.apache.org/cassandra/NodeTool and in the nodetool help to flush a node and stop accepting writes. This is the behavior I see with 2.1.2.
In 2.0.11 and 1.2.19, instead the Cassandra Process is killed. In the 1.2.19 logs, I see:
INFO [RMI TCP Connection(2)-192.168.1.5] 2014-12-05 10:32:44,234 StorageService.java (line 964) DRAINING: starting drain process INFO [RMI TCP Connection(2)-192.168.1.5] 2014-12-05 10:32:44,235 ThriftServer.java (line 116) S top listening to thrift clients INFO [RMI TCP Connection(2)-192.168.1.5] 2014-12-05 10:32:44,239 Server.java (line 159) Stop li stening for CQL clients INFO [RMI TCP Connection(2)-192.168.1.5] 2014-12-05 10:32:44,239 Gossiper.java (line 1203) Announcing shutdown INFO [RMI TCP Connection(2)-192.168.1.5] 2014-12-05 10:32:46,240 MessagingService.java (line 696) Waiting for messaging service to quiesce INFO [ACCEPT-/127.0.0.1] 2014-12-05 10:32:46,241 MessagingService.java (line 919) MessagingService shutting down server thread.
So it appears this in an intentional shut down, in which case the docs and help are wrong. I could not find a JIRA that described the change in behavior moving to 2.1.
Other users on IRC report that drain works as expected for them on 1.2.19. Attached are 2.0 logs.