[RELEASE] ScyllaDB 2026.1.3

Release Notes
, ,
1

The ScyllaDB team is pleased to announce the release of ScyllaDB 2026.1.3, 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.

Compaction

  • A race condition in the compaction manager that caused a use-after-free crash during the re-evaluation of postponed compactions is now prevented. This enhances overall compaction stability and prevents potential node failures.
    scylladb#29527

CQL

  • Prepared statement metadata IDs are now correctly generated and returned to clients. This ensures drivers receive accurate metadata, significantly improving compatibility and reliability during schema changes.
    scylladb#29477

Materialized Views (MV)

  • The Materialized View update mechanism now correctly tracks range tombstones in the update stream. This ensures range deletions on the base table are correctly applied, maintaining data consistency.
    scylladb#29569

  • Range tombstones are reliably applied after an update stream finishes processing during Materialized View updates. This guarantees MV integrity, particularly after row deletions or range updates on the base table.
    scylladb#29649

Monitoring

  • The system.size_estimates table now aggregates SSTable size reports across all shards on a node. This provides cluster administrators with more accurate and complete table size estimates per node.
    scylladb#29381

  • Comprehensive diagnostic logging has been added for internal Barrier and drain operations. This improves visibility and debugging capability when tracking node state and resource management.
    scylladb#29770

Raft

  • An internal issue in the Raft consensus mechanism where wait_for_state_change did not properly await a future is addressed. This improves the internal correctness and reliability of Raft state transitions.
    scylladb#28753

  • The Raft server consistently throws a stopped_error exception upon abortion. This allows higher-level logic to gracefully handle server aborts and ensures proper cleanup routines are executed.
    scylladb#28903

  • An assertion failure during unsuccessful group0 Raft startup sequences is now prevented, improving node resilience during bootstrapping and recovery from initial configuration errors.
    scylladb#29754

Repair

  • Synchronization of data promotion during repair operations has been corrected to prevent data from being prematurely marked as repaired. This guarantees data consistency after repair.
    scylladb#29411

Security

  • The internal system.raft table is now consistently covered by the system info encryption logic. This strengthens security in encryption-at-rest configurations by protecting all necessary internal data structures.
    scylladb#29526

Stability

  • Handling of “ghost rows” in the internal role_permissions table is now more robust, improving stability during role and permission management operations in clusters with frequent authorization updates.
    scylladb#29771

  • A potential use-after-free error in the SSTable loader is prevented when a table is dropped concurrently with a streaming operation. This significantly increases cluster stability during schema changes alongside data streaming.
    scylladb#29558

  • Internal file creation logic now correctly tracks the I/O position after reading from a seekable source. This ensures accurate position tracking during disk I/O operations, maintaining data integrity.
    scylladb#29630

  • Race conditions in the storage_service that affected the REST API during node shutdown and cross-shard forwarding are now resolved.This improves the stability of the REST API and ensures graceful node shutdown operations.
    scylladb#29611

  • A segmentation fault (segfault) that could occur when an SSTable assertion message was a null pointer is now avoided. This prevents crashes when handling unexpected assertion failures related to SSTable metadata.
    scylladb#29597

  • An internal code improvement adds an insert() method to the loading_cache utility that performs a no-operation when caching is disabled. This is primarily a maintenance improvement for utility management.
    scylladb#29782

Vector Search

  • It is now possible to create local vector search indexes using only a partial segment of the partition key, improving flexibility for vector index creation.
    scylladb#29543