Space Utilization Keeps Going Up During Normal Operation¶
This troubleshooting guide describes what to do when Scylla space usage keeps going up.
Over the lifetime of the cluster, old data is compacted together into new SSTables, removing the old.
Spikes in storage utilization are expected during compactions but if it doesn’t reduce after a compaction
finishes it can be indicative of a problem.
You can use the
lsof Linux utility to check if there are files that Scylla has deleted but whose
deletion are not reflected in the filesystem.
$ lsof -p `pidof scylla` scylla 5864 scylla 936r REG 9,127 106689876 17184087558 /var/lib/scylla/data/<keyspace>/<table>/la-21647-Data.db (deleted) scylla 5864 scylla 938r REG 9,127 199501989 17184087553 /var/lib/scylla/data/<keyspace>/<table>/la-21609-Data.db (deleted)
- If you are running repairs or large reads, those could keep references to old files. Monitor those operations to see if space utilization goes down once they finish.
- In the absence of repairs or large reads or if the problem persists when those are over this is could be an indication of a Scylla bug. Contact the Scylla team an provide
the following data:
journalctl -u scylla-server > scylla_logs.txt
ls -lhRS find /var/lib/scylla/data/ > file_list.txt
- In the mean time, restarting the Scylla nodes will release the references and free up the space.
CentOS, RHEL or Ubuntu 16.04
sudo systemctl restart scylla-server
Ubuntu 14.04 or Debian
sudo service scylla-server restart