What happens when running 'nodetool rebuild' on a node in same datacenter?

Question

This might seem silly, but I deleted everything in the data directory (/var/lib/scylla/data/*) on one of the nodes. Now, to bring back data I can run a nodetool repair or nodetool rebuild.

I have confusion regarding rebuild. I read that it is primarily used when creating new datacenter. So,in my case, If I run nodetool rebuild on my node in the existing cluster, what would happen? Will it simply copy data from the replicas for all the token ranges that it was responsible for? or Will it somehow disturb the token range and cause data loss?

I understand that nodetool repair will create consistent data copy. But assume that I have no strict requirement for consistency and just want to bring the data back on my node as quickly as possible. Which option is more safe repair or rebuild.

Answer 1

It depends on the state of the node.

If you deleted everything including the system tables, the node will try to auto-bootstrap but fail because it can't just re-join without the replace_address flag. For this reason, you won't be able to run a repair since the node won't be up.

You can only run rebuild if the node is added back into the cluster with auto_bootstrap: false. That's problematic because in this state the node will accept read requests even though it doesn't have the data. This means that if reads are done with a ONE or LOCAL_ONE consistency, there won't be any data returned. Cheers!