Disk balance issue in Kudu Tablet Server

Condition

In Kudu Tablet Servers configured with multiple fs_data_dirs directories, the disk space consumption might become uneven over time. This imbalance can occur due to several reasons, including the following:

  • Uneven data load distribution across disks.
  • Addition of new disks after the cluster is already in use.

Cause

Kudu’s current architecture does not dynamically redistribute data blocks among disks once they are written. As a result, disk space utilization can diverge significantly between directories, especially in long-running clusters or those with hardware changes, for example, adding new storage volumes.

Soultion

  1. Back up the Kudu tables.

    For instructions, see Kudu Backup.

  2. Delete the Kudu tables.
  3. Restore Kudu tables from backup.

    For instructions, see Restore tables from backups.