HSE 3.0 is now available

[hse-root]

We are pleased to announce the availability of HSE 3.0. This release enhances how data is organized on storage to improve performance for several important workloads.

Summary of Changes

Cursor performance is now independent of the cursor filter length. This means HSE applications can iterate over keys using an arbitrary cursor filter and achieve the same throughput, which is important for many client workloads. Achieving this goal required enhancing how data is organized on storage, which means that HSE 3.0 storage is not backward compatible with prior versions.

Ingest performance is improved for workloads with monotonically increasing keys, such as time-series data. Throughput is increased and both write and read amplification are significantly reduced. Again, this is the result of the enhanced data organization.

Value compression can now be specified in the put API, allowing compressed and uncompressed values to be stored together. We also normalized several other APIs and configuration parameters, which means that a minor amount of application porting is required in migrating to HSE 3.0.

Java language bindings are available for the HSE 3.0 API, and the Python bindings were updated to accommodate the API changes noted above.

New KVDB open modes are provided to enable querying a KVDB on read-only storage.

We further streamlined the build infrastructure, extended test suites, and enhanced CI workflows in GitHub Actions which now leverage OCI images.

Call to Action

Read the documentation for further details on how HSE has changed.

Migrate your HSE 2.x applications to HSE 3.0. While HSE 2.x is not going away immediately, it is no longer actively developed or maintained.

Provide feedback on how HSE can better support your storage applications.

Cheers,
HSE Project Team