YugabyteDB can be configured to use multiple storage disks by setting the --fs_data_dirs flag. This introduces the possibility of disk failure and recovery issues.

Cluster replication recovery

The YB-TServer service automatically detects disk failures and attempts to spread the data from the failed disk to other healthy nodes in the cluster. In a single-zone setup with a replication factor (RF) of 3: if you started with four nodes or more, then there would be at least three nodes left after one failed. In this case, re-replication is automatically started if a YB-TServer or disk is down for 10 minutes.

In a multi-zone setup with a replication factor (RF) of 3: YugabyteDB tries to keep one copy of data per zone. In this case, for automatic re-replication of data, a zone needs to have at least two YB-TServers so that if one fails, its data can be re-replicated to the other. Thus, this would mean at least a six-node cluster.

Failed disk replacement

To replace a failed disk, perform the following steps:

  1. Stop the YB-TServer node.
  2. Replace the disks that have failed.
  3. Restart the yb-tserver service.

On restart, the YB-TServer will see the new empty disk and start replicating tablets from other nodes.