The ScyllaDB team announces ScyllaDB 2025.3.6, a bug-fix production-ready patch release for ScyllaDB 2025.3 Feature Release.
Note that there is a new Short-Term Support (STS) Feature Release 2025.4. You are welcome to upgrade to it for the latest and greatest features.
More information on ScyllaDB’s Support Policy is available here.
Related Links
Bug Fixes
Address Map & Replication
-
Address map updates could become stuck and fail to replicate across shards, potentially leading to inconsistent cluster state and stalled operations.
-
Address map replication was redesigned to use barriers and a more reliable replication mechanism, improving correctness and fault tolerance under load.
scylladb#26865
Tablets
-
During node replacement, tablets could be incorrectly migrated across racks, violating rack-awareness guarantees.
-
Tablet migration logic was corrected to ensure rack-aware placement is preserved during node replacement and rebalancing.
scylladb#24485 -
Tablet scheduler plan generation could be slow in clusters with multiple racks, impacting scale-out and rebalancing operations.
-
Scheduler planning was optimized and enhanced with rack-aware balancing when rf_rack_valid_keyspaces is enabled.
scylladb#26016
Failure Detection & Gossip
-
The direct failure detector could behave inefficiently under load, with unclear timeout handling and scheduling contention.
-
Failure detector handling was optimized by improving verb handlers, passing explicit timeouts, and running it in the gossiper scheduling group.
scylladb#27482 -
Memory Management - reader creation could crash the node when the memory kill threshold was exceeded, resulting in abrupt termination.
-
Exception handling was improved by allowing proper propagation of memory-related errors, enabling more graceful handling under memory pressure.
scylladb#27475
Raft & Topology
-
The topology coordinator did not consistently handle aborted Raft requests, and node removal events could be reported multiple times.
-
Handling of aborted requests was improved, node removal notifications were deduplicated, and streaming session IDs are now set at the correct time.
scylladb#23589
Secondary Indexes & Views
-
Secondary index builds could fail when a node was removed during index creation, especially in tablet-based deployments.
-
Index build behavior during node removal was stabilized, and log verbosity was reduced for expected aborts during view creation.
scylladb#27255