[RELEASE] ScyllaDB 2025.3.8

Release Notes
, ,
1

The ScyllaDB team announces ScyllaDB 2025.3.8, 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

Alternator

  • Cross-shard access in alternator::executor::delete_table() could occur, potentially leading to instability or data corruption in the Alternator service. The issue was resolved by correcting the code to use the storage_proxy from the correct shard, which stabilizes table deletion and maintains data integrity.
    scylladb#27223 scylladb#27259

Commitlog

  • A race condition could occur during commitlog shutdown when a reserve replenish completed before an allocation call was made. This was addressed by ensuring the replenish queue is always aborted when the shutdown loop exits, which safely resolves the race condition during commitlog shutdown.
    scylladb#28678 scylladb#28691

Materialized Views

  • A Materialized View (MV) could be incorrectly marked as built when the underlying reader produced no partitions on a partial range, leading to data inconsistency. This was fixed to ensure the view is only marked as built when partitions are successfully processed, thereby maintaining MV consistency.
    scylladb#26887

Networking

  • Network connection establishment could fail due to transient DNS resolution issues. The connection factory was improved to introduce a Time-To-Live (TTL) timer for DNS addresses and implement a retry mechanism on failure. This enhances connection resilience for external connections, such as the S3 client, against transient DNS resolution issues.
    scylladb#28403

Repair

  • Repair and streaming operations lacked proper session support in repair_service::rebuild_with_repair, and the topology guard was not correctly passed to the Read-Before-New-Owner (RBNO) service logic. This was fixed by implementing session variable support for streaming and ensuring the topology guard is correctly passed to RBNO, which improves the overall robustness and reliability of repair and streaming across the cluster.
    scylladb#27759 scylladb#28297

Storage

  • Errors in S3 part size calculation and a lack of controlled concurrency for multipart uploads could cause issues during large object operations. The S3 client was fixed to correctly calculate part sizes respecting all limits and to restrict multipart upload concurrency to avoid overwhelming S3 or exhausting local resources. This ensures stable and efficient S3 interactions for operations like backups and restores.
    scylladb#28695 scylladb#28664