I had a cluster with three nodes. The keyspaces have a replication factor of three. I am using vnodes with 8 tokens. Each node owns 100% of data. So far so good. I added a fourth node. I ran a nodetool repair and nodetool cleanup on all the nodes. Now the new node owns 30% of the data another node owns 70% and the other two owns 100% of the data. How can I balance this cluster?
Virtual nodes with a small number of tokens tend to have a bit of a skew compared to vnodes with num_tokens: 256. When the size of the data is small (say less than 50GB) and there is a large range between the smallest and largest partitions, expect significant discrepancies between nodes. As the node density increases, you should see the delta between nodes to decrease. Cheers!
Datastax recommendation is 8 tokens, right? Is there still value in using 256 tokens?
You can continue using 256 tokens but be aware of the slight performance hit from the management overhead. Cheers!
Also, I realize that although I have 4 nodes, they are in 3 racks. So Cassandra, divided the data in three racks.
@gmldba_107428 Just a friendly note that you need to comment instead of "writing an answer" so other users don't confuse this as an answer in the future. Please edit and re-post. Cheers!
5 People are following this question.
