how to deal with data move

[hedap2p]

I check the code，find rebanlance only switch master with least_loaded_slave,no data is moved or deleted. If disk will full,we need move some data to a new server,how to deal with this situation?

[newpoo]

Good question!

Internally we have a command "expand" to expand a cluster. It uses existing APIs in rocksdb_admin.thrift to operate the hosts involved. At a high level, to move a shard from host A to B, we do the followings.

1. backup the shard from A to HDFS.
2. restore the shard from HDFS to B.
3. remove the shard from host A.

We didn't include the command in the public repo because it requires some refactoring to remove the internal dependencies. Also, we are now working on an automatic cluster management system, which will take precedence over the current script based solution. We will make new cluster management system in the public repo once it's in production at Pinterest.