[RELEASE] ScyllaDB 2026.1.4

Release Notes
, ,
1

The ScyllaDB team is pleased to announce the release of ScyllaDB 2026.1.4, a production-ready patch release for ScyllaDB 2026.1 Long Term Support (LTS) Release.

More information on ScyllaDB Version Support Policy is available here.

Related Links

Bug Fixes

The following issues are fixed in this release.

Authentication

  • Certificate authorization setup was incomplete when the node started up. The system now performs full connection setup for CertificateAuthorization during node startup, guaranteeing consistent application of authorization rules from the beginning of the connection lifecycle.
    scylladb#29409

CQL

  • Running DESCRIBE on a table with compaction disabled incorrectly reported NullCompactionStrategy. DESCRIBE now accurately reflects the schema state when compaction is not active, improving schema introspection via CQL.
    scylladb#29886

Monitoring

  • Tombstone garbage collection (GC) metrics included unexpired tombstones in their count. The mutation_compactor logic is updated to count only expired tombstones, which provides users with accurate metrics for monitoring and optimizing tombstone management.
    scylladb#28744

Performance

  • A long-running operation in the trie index writer did not include a preemption point, which could block shard progress during certain background tasks. A preemption point was introduced into trie_writer::complete_until_depth(), reducing scheduler stalls and improving responsiveness under heavy I/O workloads.
    scylladb#29816

Repair

  • ScyllaDB previously accepted repair requests where the start and end tokens were identical, which is an invalid token range. The system now explicitly rejects repair requests where the start and end tokens are the same, preventing invalid repair sessions from starting, streamlining repair operations and resource usage.
    scylladb#29863

Schema Management

  • Local system table creation could experience synchronization issues during startup. This is addressed by ensuring proper asynchronous behavior and preventing contention, improving stability during the node boot process.
    scylladb#29833

  • The internal group0_schema_version was not consistently set during node boot. This is now enforced during startup, guaranteeing that internal schema management operations begin with the correct reference point and improving schema consistency across the cluster after a restart.
    scylladb#29858

Stability

  • A memory safety problem causing a use-after-free condition could occur during prepared statement invalidation within table_helper. This has been addressed, reducing the risk of rare crashes during workloads that frequently invalidate prepared statements.
    scylladb#29802

Tablet Management

  • The topology_coordinator did not properly join the tablet load stats refresh fiber during shutdown. The refresh process is now cleanly joined and stopped when the coordinator stops, leading to more reliable shutdown behavior.
    scylladb#29811

  • The tablet allocator in the load_balancer encountered issues when handling tables that had been dropped. This is fixed to ensure allocation logic remains sound, avoiding unexpected behavior during topology changes. scylladb#29883

Vector Search

  • Vector Search requests could abort unexpectedly with an error under rare conditions. The underlying problem was fixed, improving stability and reliability for applications using Vector Search functionality.
    scylladb#29773