Reputation: 135
I have configured only three nodes, however nodetool sees one more on 10.0.0.200 IP.
So, from where nodetool gets the information about this node?
How can I safely remove this node and split the information between 3.
> /usr/bin/cassandra/nodetool status
Datacenter: datacenter1
=======================
Status=Up/Down
|/ State=Normal/Leaving/Joining/Moving
-- Address Load Owns Host ID Token Rack
DL 10.0.0.200 ? 17.3% 8f2a4aab-0b83-454a-96b9-52ceff768892 17641649882960857202604498991049536772 rack1
UN 10.0.0.1 85.73 KB 56.3% 159d6a4e-631b-4f20-8c9c-fe9bbba4aeac 113427455640312821154458202477256070485 rack1
UN 10.0.0.11 32.94 MB 3.4% 2f93c222-5885-4141-a794-9913c68eb135 119167364151511659191387118070504026424 rack1
UN 10.0.0.3 15.92 MB 23.0% d6db9f0c-0adf-435d-ad38-a2c03f3b45b5 158286408450244808806328613098739900318 rack1
Upvotes: 1
Views: 448
Reputation: 7305
1) Always specify a keyspace when looking at the load column in nodetool status
otherwise the data will not be accurate.
2) This depends mostly on your initial tokens (make sure they're equally spaced out) and on your data model (avoid hotspots / super wide rows).
nodetool decomission
will get rid of the node if you need to kill it. It will stream the data to the other nodes in your cluster and then remove it from the ring.
If decomission fails 'nodetool remove' will remove a node thats already been terminated.
Finally if all else fails the jmx 'assassinate' mbean will remove a node from the ring. Only use if remove didn't work.
Upvotes: 4