ScyllaDB Docs ScyllaDB Manual ScyllaDB for Administrators Procedures ScyllaDB Maintenance Procedures Automatic Scrub

For AI agents: a documentation index is available at https://docs.scylladb.com/manual/master/llms.txt. A Markdown version of this page is at https://docs.scylladb.com/manual/master/operating-scylla/procedures/maintenance/automatic-scrub.md.

Caution

You're viewing documentation for an unstable version of ScyllaDB Manual. Switch to the latest stable version.

Automatic Scrub¶

Disks can silently corrupt data and corruption can hide in unused sstables. ScyllaDB offers a way to periodically scrub sstables and detect corruption. Automatic scrub checks sstables for corruption one at a time for each shard. This operation is not read only and may rewrite sstables. This feature only covers local sstables, object-storage sstables are skipped.

Which sstables are scrubbed¶

An sstable is considered validated if, within the scrub period, it was:

written (this includes being created by compaction) or streamed, as file streaming validates component digests, or
selected for automatic scrub and subsequently validated with scrub in either VALIDATE or ABORT mode.

The timestamp of the last validation is stored in the scylla-metadata component as scrub_time. It is only present in sstables written or rewritten by ScyllaDB versions which include automatic scrub. Manual nodetool scrub operations in VALIDATE mode do not update the scrub time and remain read-only. When corruption is detected, invalid sstables are placed into quarantine. Once quarantined, an sstable will not be eligible for automatic scrub.

For sstables with the scylla-metadata component and component digests present, the VALIDATE mode for scrub will be used and scrub time will be updated using component rewrite. For other sstables, scrub in ABORT mode will be used. Consequently, automatic scrub will rewrite eligible sstables so that both the scylla-metadata component and component digests are present. Additionally, if an sstable was not validated for at least half the scrub period, regular compactions will verify component digests. Regardless of the scrub time, if corruption is found in the sstable during a regular compaction, the malformed input sstable will be quarantined.

Caution

Sstables without a scrub_time value are considered not validated. Additionally, if they lack component digests, they will be rewritten by scrub in ABORT mode. Consequently, enabling automatic scrub for the first time may cause a large number of sstable rewrites.

Setting the scrub period¶

Automatic scrub is controlled by the auto_scrub_period_seconds cluster configuration option. To enable it for the whole cluster:

ALTER CLUSTER WITH auto_scrub_period_seconds = 86400;

The period is given in seconds and must be at least 3600 (one hour). To disable it, set it to 0 (the default). The option can also be set for a single keyspace or table: a value set on a TABLE overrides the one set on its KEYSPACE, which overrides the CLUSTER value. See Cluster configuration for how scopes are resolved.

ALTER KEYSPACE ks WITH auto_scrub_period_seconds = 43200;
ALTER TABLE ks.tbl WITH auto_scrub_period_seconds = 21600;

Use auto_scrub_period_seconds = 0 to disable automatic scrub even if a broader scope enabled it. Use auto_scrub_period_seconds = NULL to remove a value and fall back to the broader scope.

Was this page helpful?