I am looking into changing the num_tokens (vnodes) for a cluster from 16 to 4. The procedure I have been given says to decommission a node before making the num_tokens change. Drain seems to do the same thing. What are the pros and cons of using decommission versus drain? Thanks.
What is the difference between "nodetool decommission" and "nodetool drain"?
The nodetool decommission command removes a node from the cluster. As the name suggests, it decommissions the node so that it will no longer be part of the Cassandra ring of nodes.
The nodetool drain command does 2 things:
- it triggers a flush of memtables to disk (writes the data in memory to SSTable files)
- it shuts down the native transport server meaning the node will no longer accept client requests on the CQL client port
9042
You would typically drain a node before shutting down Cassandra to carry out maintenance work. When all the memtables are flushed to disk, the commitlogs are also cleared so there are none to replay on startup. This makes restarting Cassandra a lot faster.
By now you should work out that these two commands do not do the same thing. They provide two completely unrelated functions.
For more information on decommissioning nodes, see Removing a Cassandra node. Cheers!
5 People are following this question.
