[RELEASE] ScyllaDB 2026.1.1

Release Notes
, ,
1

The ScyllaDB team is pleased to announce the release of ScyllaDB 2026.1.1, 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

  • Cache invalidation ranges during compaction were handled in a way that could use more memory than optimal. utils::chunked_vector is now used for managing cache invalidation ranges in the database and compaction layers. This reduces memory pressure and improves compaction and cache management performance.
    scylladb#28975

CQL

  • An issue in CQL prepared statement handling, related to an invalid weak handle, could occasionally lead to race conditions and was associated with test failures involving Lightweight Transactions (LWT) and tablet operations. The fix pins the prepared cache entry during the prepare phase to prevent the invalid weak handle race condition. This enhances the reliability of LWT and prepared statements.
    scylladb#27657, scylladb#29001

Materialized Views

  • Previously, Materialized View updates were unnecessarily generated even if a collection field in the base table was unmodified. The logic was updated to allow skipping view updates when a collection remains unchanged. This improves performance by reducing unnecessary writes to materialized views.
    scylladb#28839, scylladb#28977

  • A memory corruption issue (AddressSanitizer: use-after-poison) was observed during Materialized View updates, linked to holding keyspace references across preemption points. The fix prevents holding the keyspace reference across preemption boundaries during the mutate_MV operation, resolving the memory safety issue.
    scylladb#28925, scylladb#28968

Monitoring

  • The load statistics filtering mechanism has been improved for tablets. This change provides more accurate and relevant load statistics for monitoring and optimization.
    scylladb#29034

Performance

  • Collection mutations unnecessarily copied the serialized collection data during processing, leading to increased memory usage and overhead. The fix avoids copying the serialized collection during mutation. This reduces memory overhead and improves write performance for tables utilizing collections.
    scylladb#29024

  • Read performance was impacted when accessing densely populated partition index pages. The issue is fixed by improving the efficiency of operations on these index pages. This significantly boosts read performance when dealing with partitions containing a high number of columns or rows.
    scylladb#29136

  • Token processing was not properly rate-limited across a range, which could allow large range-based operations to overwhelm cluster resources. A rate limit mechanism has been introduced to limit all tokens originating from a given range. This helps prevent cluster resource saturation, improving overall cluster stability under heavy load.
    scylladb#29107

  • A temporary workaround for a reader concurrency semaphore leak could affect the consistency of query concurrency limits. The fix addresses and corrects the workaround for the reader concurrency semaphore leak. This prevents long-term resource exhaustion and ensures consistent query concurrency limits.
    scylladb#28988

Reliability and Stability

  • A tablet virtual task wait could fail during concurrent tablet resize because not all tablets were scanned. The fix ensures all tablets are included, allowing the wait operation to complete reliably even during resizing.
    scylladb#28202, scylladb#28991

  • Task wait requests would fail if the underlying RPC call failed during execution, which could occur in environments with unstable networking. The fix modifies the task execution logic to prevent the wait request from failing solely due to an RPC failure. This enhances the resilience of task execution.
    scylladb#29126

  • A deadlock could occur in the background storage group merge fiber during tablet operations. The fix resolves the deadlock condition within the tablet background storage group merge fiber. This prevents system stalls and improves the stability of tablet operations.
    scylladb#29144

Repair

  • Running nodetool cluster repair previously failed if one or more tables had been dropped. This fix ensures that nodetool cluster repair gracefully handles dropped tables by not failing the operation. This improves the robustness of routine cluster maintenance tasks.
    scylladb#29006

Schema Management

  • A problem existed where tables updated through group0 operations might bypass the schema commitlog. The fix adds an assertion to guarantee that tables updated via group0 properly utilize the schema commitlog. This ensures data integrity and consistency for schema-related operations.
    scylladb#28916

Streaming

  • File streaming operations previously released disk space less incrementally, which could result in higher temporary disk usage during large streaming operations. Space is now released incrementally during file streaming. This reduces peak disk space usage.
    scylladb#28769

Vector Search

  • TLS connections for Vector Search failed when the server name was provided as an IP address instead of a hostname. The fix corrects how the TLS server name is handled when an IP is used. This enables secure, encrypted connections for Vector Search clients communicating using IP addresses.
    scylladb#28964