The ScyllaDB team announces ScyllaDB 2025.2.5, a bug-fix production-ready patch release for ScyllaDB 2025.2 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
The following issues are fixed in this release.
Build & Tooling
- The tools/toolchain/prepare script failed on Fedora 43 due to reliance on the deprecated reg tool.
The toolchain preparation process now uses skopeo, restoring compatibility with Fedora 43 systems.
scylladb#27178, scylladb#27196
Cloud Images & Packaging
- ScyllaDB machine images are updated to include ScyllaDB version 2025.2.5, ensuring consistency between packaged images and the released binaries.
No configuration changes are required when upgrading existing images.
scylla-machine-image commit 5a574ce
Logging & Observability
-
Streaming operations now provide more detailed logging.
The load_new_sstables and download_new_sstables operations log all option values, improving troubleshooting and operational visibility.
scylladb#27299, scylladb#27338 -
A new fatal variant of utils::on_internal_error() is introduced.
This allows internal errors to reliably terminate the process when required, improving failure visibility and diagnostics.
scylladb#27187
Performance
- Updated Profile-Guided Optimization (PGO) profiles for both aarch64 and x86_64 architectures.
These updates improve runtime performance by aligning optimizations with current workload characteristics.
commit af570f7, commit 3e6ef72
Reliability
-
Address map replication is made more efficient and robust.
The replication mechanism now uses a barrier-based approach, preventing situations where updates could get stuck or fail to propagate across shards, and fixing a lambda coroutine issue in address_map::replicate().
scylladb#26865, scylladb#26835, scylladb#27187 -
Locator node handling is corrected to preserve the _excluded field.
The _excluded state is now retained during cloning and included in verbose formatting and missing-place diagnostics, improving correctness during topology changes.
scylladb#27289 -
Notifications related to an expiring Extended Repair Manager (ERM) being held for too long are fixed.
The corrected logic ensures notifications are reliably triggered, and new test coverage validates the behavior.
scylladb#27141, scylladb#27274
Stability
-
A crash in SSTable dictionary compression when preparing sample data is fixed.
The fix prevents invalid memory access during sample preparation, improving stability during SSTable processing.
scylladb#27417 -
An overflow when computing data distribution across shards is fixed.
This prevents incorrect shard-level calculations that could previously lead to crashes or undefined behavior.
scylladb#27418 -
Multiple seastar_memory - oversized allocation errors encountered when creating thousands of tables with Vnodes are resolved.
The internal mapping type of tables_metadata::_ks_cf_to_uuid is changed to significantly reduce memory pressure.
scylladb#26787, scylladb#27195 -
A premature loop break in tablet_sstable_streamer::stream that caused SSTables to be skipped during streaming is fixed.
The corrected loop logic ensures all eligible SSTables are properly streamed.
scylladb#26979, scylladb#27150 -
A failing test (test_double_node_failure_during_mv_insert_3_nodes) caused by incorrect commitlog reads is fixed.
commitlog::read_log_file now correctly checks the end-of-file position on all data reads, preventing assertion failures.
scylladb#24346, scylladb#27340 -
The Seastar submodule is updated to ensure seastar_assert terminates the process using SIGABRT.
This improves crash diagnostics and aligns assertion handling with expected debugging behavior.
scylladb#27127, scylladb#27404