[Release] ScyllaDB 2025.4.8

Release Notes
, ,
1

The ScyllaDB team is pleased to announce the release of ScyllaDB 2025.4.8, a production-ready patch release for ScyllaDB 2025.4 Feature Release.

Note there is a new Long Term Support (LTS) release 2026.1. You are welcome to upgrade to it for the latest and greatest features.

More information on ScyllaDB Version Support Policy is available here.

Related Links

Bug Fixes

Alternator

  • Alternator lacked reporting for per-table batch read and write operations because the necessary histograms were not updated. This issue is fixed by introducing new per-table batch latency metrics, which improves monitoring visibility and provides more detailed operational insights for Alternator workloads.
    scylladb#28721, scylladb#29365

Authentication

  • A stability issue in the LDAP client could cause unexpected failures during authentication results polling. This is resolved by hardening the internal memory handling, which improves the overall reliability of LDAP-based authentication.
    scylladb#29454

Compaction

  • A stability issue in the compaction manager’s scheduling logic was present during compaction reevaluation. This fix improves the internal memory management, enhancing stability during compaction scheduling.
    scylladb#29629

  • When a table’s main compaction group was deliberately disabled, the system would still create new split compaction groups. This is fixed by preventing the creation of new split compaction groups if the main group is inactive, granting better administrative control over compaction resources and behavior.
    scylladb#29572

CQL

  • A concurrency issue could occur when preparing CQL statements. This is fixed by improving the concurrency control around the prepared statement cache, which enhances the stability and reliability of prepared statements.
    scylladb#29195

  • Null values were incorrectly formatted during data value formatting for CQL. Correcting the null value handling ensures accurate and consistent data representation in CQL query results.
    scylladb#29434

Encryption

  • A deadlock could occur when retrieving encrypted data, causing read operations to stall. This issue is resolved, improving the availability and responsiveness of data access when using encryption at rest.
    scylladb#29359

  • The encryption information reporting failed to include the internal system.raft table. This is fixed by ensuring the table is covered in the encryption status report, providing complete and accurate security status reporting for all critical system tables.
    scylladb#29582

Materialized Views

  • When updating a Materialized View (MV), the update stream failed to track existing range tombstones, potentially leading to data consistency issues. This is fixed by ensuring range tombstones are correctly tracked during view update building, improving consistency in Materialized Views.
    scylladb#29643

  • Existing range tombstones were not applied after the update reader finished during a view update operation. The fix ensures all existing range tombstones are properly applied, which improves consistency within Materialized Views.
    scylladb#29708

Networking

  • A memory accounting problem existed in the transport layer when handling large responses over slow networks. Improvements to the transport layer logic provide better memory accounting, resulting in more accurate resource monitoring and allocation for network-heavy operations.
    scylladb#29455

  • Race conditions were possible in the storage service’s REST API during node shutdown and cross-shard forwarding. These races are fixed, enhancing the stability and reliability of the REST API and inter-node communication, particularly during administrative operations.
    scylladb#29668

Object Storage

  • Azure KMS key retrieval could fail due to DNS timeouts following a node soft reboot. This is resolved by stabilizing DNS resolution during the boot process, ensuring reliable Azure KMS key access following a reboot.
    scylladb#26519, scylladb#29354

  • When S3 credential refresh failed, the system erroneously reset the expiration timer, which could lead to unexpected service interruptions. The logic is corrected to prevent this incorrect timer reset upon failure, avoiding unexpected service interruptions.
    scylladb#29056, scylladb#29364

  • Logging for the object storage endpoint parameter was incorrect, hindering troubleshooting. The fix ensures the parameter is correctly formatted, improving logging and debugging visibility when configuring object storage.
    scylladb#27718, scylladb#27733

Performance improvements

  • Inefficient standard memory structures were used for storing promoted indexes in SSTables. This is optimized by introducing a more memory-efficient structure for promoted indexes, which improves memory efficiency and lookup performance for SSTable partition indexes.
    scylladb#29772

  • An unnecessary function call was made even when the caching feature was disabled, leading to redundant work. This function is now skipped when caching is disabled, eliminating redundant work and improving overall system efficiency.
    scylladb#29790

Raft

  • The Raft state change handling included incorrect control flow, potentially affecting cluster coordination reliability. This is fixed by improving Raft state change handling to ensure consistent behavior during cluster coordination operations.
    scylladb#29594

  • The Raft component failed to signal its state properly when a server was aborted. This is fixed by ensuring the system correctly signals the stopped state, enabling graceful shutdown and correct error propagation for Raft-based services.
    scylladb#29715

Reliability

  • A resource leak was present in the read concurrency mechanism. This leak is resolved, ensuring system resources are correctly managed and preventing eventual resource exhaustion, thus improving long-term system stability.
    scylladb#29196

  • A memory management issue could occur in the SSTables loader if a table was dropped during an active streaming operation. This is fixed by safeguarding memory management during these operations, which prevents unexpected termination during streaming and schema changes.
    scylladb#29600

  • SSTable assertion parsing could mishandle null assertion messages. This is fixed by safely handling null messages, which prevents unexpected termination and improves the robustness of SSTable handling.
    scylladb#29720

  • Tablet merge exceptions were difficult to troubleshoot due to a missing formatter for the group_id argument. The fix adds the necessary formatter for the group_id argument, resulting in clearer and more actionable error messages for tablet merge exceptions.
    scylladb#29453

  • The task service could incorrectly fail the wait request if an internal RPC call failed. The fix modifies the task service to prevent the wait request from failing on RPC errors, which enhances the resilience of asynchronous operations.
    scylladb#29193

  • Task service requests for waiting operations were failing if the associated table had been dropped. The task service is updated to return a successful status if the table was dropped, ensuring task completion consistency despite ongoing schema changes.
    scylladb#29361

Vector Search

  • A race condition was present during the Vector Search connection timeout process. The race condition is fixed, which improves the stability and predictability of Vector Search operations, especially under variable network conditions.
    scylladb#29360