[RELEASE] ScyllaDB 2025.4.6

Release Notes
, ,
1

The ScyllaDB team is pleased to announce the release of ScyllaDB 2025.4.6, a production-ready patch release for ScyllaDB 2025.4 Feature Release.

Note there is a new Long Term Support (LTS) release 2026.1. You are welcome to upgrade to it for the latest and greatest features.

Related Links

Bug Fixes

Consistency

  • An error existed in the sanity check for the everywhere_replication_strategy when the read_new flag was enabled, potentially resulting in incorrect validation logic. The check is corrected to properly validate read replicas under this specific condition. This ensures the integrity of replication strategy checks, providing more reliable cluster validation.
    scylladb#29269

Materialized Views

  • Materialized Views (MVs) were unnecessarily updated when write operations modified collection types (like lists or maps) but did not functionally change the final collection state. The fix enables skipping the Materialized View update process if the collection’s value remains unchanged. This reduces write overhead and improves performance on base tables that feed MVs containing collections.
    scylladb#29094

Performance

  • Read performance was impacted when querying partitions that contained densely populated index pages due to inefficient handling of the index structure. This has been resolved by optimizing the partition index page logic. This improvement reduces read latency and increases throughput, especially for queries against high-density partitions.
    scylladb#29140

  • Requests that spanned across an entire token range were not being properly subjected to rate limiting, which could lead to resource contention and uneven load distribution across the cluster. The fix ensures that rate limiting is applied to all tokens covered by a range request. This provides consistent resource management and prevents performance degradation during wide-ranging operations.
    scylladb#29194

  • Write operations involving collection types suffered from performance overhead due to unnecessary copying of serialized collection data during the mutation/collection_mutation path. The system has been optimized to eliminate this redundant data copy. This provides a performance boost for all write operations that modify collection types.
    scylladb#29037

Repair

  • Running nodetool cluster repair would fail if a table that was part of the keyspace being repaired had been previously dropped. The cluster repair logic is updated to gracefully handle and skip over non-existent tables. This ensures the repair operation completes successfully on the remaining tables in the keyspace without interruption.
    scylladb#29038

Schema

  • An assertion failure could occur in the schema commitlog when cluster-wide schema changes were applied via group0. The fix ensures that any table update processed through the Raft group0 leader correctly utilizes the schema commitlog mechanism. This is a correctness guarantee that prevents subtle schema inconsistencies and unexpected node termination.
    scylladb#28986