not a support question
Casa
Member
I have a Cassandra Multi DC cluster (DC1:2,DC2:2). I want to replace one node in DC1 with a new node. i have to do this on live. Are these steps best way to do that ?
1. Install Cassandra and make default settings in the new node.
2. Set the initial token of the new node to initial_token of the node that need to replaced - 1.
3. Start the new node
4. Bring down the old node.
5. Run cleanup on all the nodes
Your steps are correct. We have documentation you can follow if you'd like: http://www.datastax.com/docs/1.0/operations/cluster_management#replacing-a-dead-node
Casa
Member
I went through the above link. But, bringing down the old node is before starting up new node. which is right for dead node replacement.
My steps had reverse. The reason is, my node is not dead still and i do not want any request fail when i do node replacement. am i right here ?
Yeah, apologies. Steps 1-6 still apply. You can avoid step 7 (repair) as long as auto_boostrap=true.
Casa
Member
Something like doing a decommission of a node live. but i want the data to be copied only to the new node. Can i assign the token of the new node very near to the token of old node so that when old node is decommissioned, all data goes to the new node.
Is this right way ?
Yes. You should not bring down the "old node" until the new node is up and running. Step 2 in your original post is the "very near to the token of the old node" part you're mentioning. Just be sure to set auto_bootstrap to true on the new node, and it will automatically stream the data it needs when it joins the ring.
One extra thing to note is to be sure you have RF > 1 on all of your keyspaces. That will ensure that if something does go wrong during the swap, you won't actually lose any data.
Casa
Member
Thanks, summing up the steps again.
1. Install Cassandra in new node
2. Set the initial token of the new node to initial_token of the node that need to replaced - 1.
3. Start the new node. wait for bootstrap
4. Bring down the node which needs to be replaced.
please confirm.
Confirmed. If it's feasible, it would be a good idea to run repair on all of your nodes afterwards, just to make sure all data is where it's supposed to be.
How did you bring the node down? If you use decommission, it should be removed from the ring (and that decommission is what would stream the old node's data to the new node)
You must log in to post.
