ContextQMD
Back to Scylladb

Dropped (or truncated) Table (or keyspace) and Disk Space is not Reclaimed

docs/troubleshooting/drop-table-space-up.rst

latest1.5 KB
Original Source

Dropped (or truncated) Table (or keyspace) and Disk Space is not Reclaimed

This troubleshooting guide describes what to do when ScyllaDB keeps using disk space after a table or keyspaces are dropped or truncated.

Problem ^^^^^^^

When performing a DROP or TRUNCATE operation on a table or keyspace, disk usage is not seen to be reduced. Usually this is verified by using an external utility like the du Linux command. This is caused by the fact that by default, ScyllaDB creates a snapshot of every dropped table. Space won't be reclaimed until the snapshot is dropped.

Solution ^^^^^^^^

  1. Locate /var/lib/scylla/data/<your_keyspace>/<your_table>. Inside that directory you will see a snapshots subdirectory with your dropped data. Follow the procedure to use nodetool to :doc:remove the snapshot </operating-scylla/procedures/backup-restore/delete-snapshot>.
  2. If you are deleting an entire keyspace, repeat the procedure above for every table inside the keyspace.
  3. This behavior is controlled by the auto_snapshot flag in /etc/scylla/scylla.yaml, which set to true by default. To stop taking snapshots on deletion, set that flag to false and restart all your scylla nodes.

.. note:: Alternatively you can use the rm Linux utility to remove the files. If you do, keep in mind that the rm Linux utility is not aware if some snapshots are still associated with existing keyspaces, but nodetool is.