[RELEASE] ScyllaDB 2026.1.6

Release Notes
, ,
1

The ScyllaDB team is pleased to announce the release of ScyllaDB 2026.1.6, 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.

Commitlog

  • The commitlog now consistently embeds column mapping information in entries that take the oversized-allocation write path. Previously, a schema-version check used in that path could omit the column mapping when the segment did not yet know the schema version, which could prevent replay from resolving the schema for those entries. The check now matches the logic used elsewhere, so commitlog replay reliably resolves the schema for all entries.
    scylladb#30286

CQL

  • DESCRIBE now uses a chunked container when building its output, avoiding large single allocations on clusters with a very large number of tables. This prevents oversized-allocation warnings and improves stability and memory behavior when describing large schemas.
    scylladb#30243

Materialized Views (MV)

  • A static row update on a base table that has a secondary index or materialized view defined with a collection index on a regular column is now handled correctly. The update path that compares old and new values for computed collection columns is corrected, improving stability for views and indexes built on collection columns.
    scylladb#29978

Networking

  • The bundled Seastar version is updated to improve receive-queue configuration for Mellanox ConnectX-6 (mlx5) network adapters. This improves network setup and tuning on hosts using these adapters.
    scylladb#29952

Performance

  • A preemption point is added when sampling data during dictionary training in the storage service, and a redundant copy is replaced with a move. This reduces a brief stall that could occur during this operation, improving latency consistency.
    scylladb#30075

Raft

  • The Raft message buffer now uses a chunked container, so it no longer requires a single large contiguous allocation as the number of buffered messages grows. This avoids oversized-allocation warnings and improves stability under load.
    scylladb#30249

  • A race during shutdown involving a Raft read barrier is now prevented. The relevant code now holds the group0 gate while performing the read barrier, so the operation can no longer continue after the underlying server has been torn down. This improves stability during node shutdown.
    scylladb#30037

Repair

  • Follower-side repair state is now reliably released when a repair session closes during topology operations such as removenode. If a repair coordinator stops after followers have created their repair state but before sending the stop request, that state is now aborted promptly when the session closes, so it no longer remains pinned. The session abort source is now obtained through the guard that keeps the session alive, ensuring the reference stays valid for the lifetime of the repair state.
    scylladb#30085, scylladb#30320

Security

  • ScyllaDB now bounds the recursion depth and the length of CQL WHERE clauses during expression evaluation. Deeply nested expressions and very large WHERE clauses are now rejected with a clear limit, preventing unbounded recursion and quadratic processing time. The WHERE clause limits are configurable. This addresses CVE-2026-31947 and CVE-2026-31948.
    scylladb#14472

Stability

  • A detached background operation that sets up an internal system table is now observed during shutdown. If group0 is torn down while this setup is still in flight, the resulting failure is now handled and logged instead of being left unobserved, reducing error noise during shutdown.
    scylladb#30295

  • The multi-writer mutation path now propagates a consumer exception raised at end-of-stream. This ensures errors are surfaced correctly instead of leading to a follow-on internal error, improving robustness when a write is aborted.
    scylladb#30360

  • A use-after-free in topology snapshot merging is now prevented. The code now collects and freezes split mutations with proper sequencing rather than discarding an asynchronous result, so processing always completes against valid data. This improves stability during topology changes.
    scylladb#30400

Storage

  • Closing SSTable files now waits for file removal to complete. Previously, deletion of temporary files could continue in the background after the close call returned, so callers could briefly observe stale temporary files on disk. The close operation now awaits removal, so temporary files are gone once it completes.
    scylladb#30302

Tablets

  • Tablet cleanup statistics are now self-healing. Disk-space accounting during tablet cleanup is corrected so that concurrent compaction and rebuild activity can no longer drive the live and total disk-space counters to inconsistent or negative values, improving the accuracy of reported storage statistics.
    scylladb#30266