Release Notes#

Version 2.0.14#

Major Changes

  • Documentation cleanup across release-facing docs.

Version 2.0.12#

Major Changes

  • Added support for Python 3.14.

  • Removed the pycapnp dependency. Scaler now directly relies on the C++ Capnp library.

  • Standardized multiprocessing process creation to consistently use spawn context.

Version 2.0.9#

Major Changes

  • Added AWS HPC Batch support with related bug fixes and docs updates.

  • Improved scaling behavior by reporting pending (launching) workers to prevent over-requesting.

  • Added and expanded examples/docs, including ORB AWS EC2 quickstart, SwapCVA, XVA, and order-flow docs.

Version 2.0.7#

Major Changes

  • Added CI support for Intel macOS.

  • Fixed macOS ARM release CI issues.

Version 2.0.5#

Major Changes

  • Fixed Linux ARM build issues.

Version 2.0.4#

Major Changes

  • Fixed Python 3.14 wheel build issues.

Version 2.0.3#

Major Changes

  • Added CI build, test, and wheel release support for ARM macOS.

Version 2.0.2#

Major Changes

  • Added a YMQ magic-string check to detect non-YMQ clients early.

Compatibility Notes

  • Network handshake protocol has been updated. Older clients, worker and scheduler will not be able to connect to newer versions.

Version 2.0.1#

Major Changes

  • Object storage address advertisement behavior is corrected and more predictable.

Version 2.0.0#

Scaler 2.0.0 focuses on policy-driven scaling, worker manager consolidation, and the YMQ-libuv networking/runtime transition.

Major Changes

  • Scaling policy system reworked: scaling controllers are stateless, REST-driven scaling control is removed, waterfall_v1 is introduced, capability-aware behavior is improved, and the default scheduler scaling policy is now vanilla (with fixed_elastic removed).

  • Worker manager architecture reorganized: worker manager lifecycle and control paths are unified across scheduler, configuration, and entry points, including stricter worker_manager_id handling.

  • New worker managers added:

    • AWS Batch worker manager

    • ORB AWS EC2 worker manager

  • Worker manager implementations simplified: fixed-native behavior is merged into native management paths, reducing overlap and simplifying deployment options.

  • YMQ-libuv becomes the standard network backend: Python wrappers and connector sets are added, object storage and scheduler paths are aligned, and YMQ now uses libuv as the supported I/O backend.

  • Reliability and correctness improvements: targeted fixes across asyncio, connector lifecycle, object storage handling, and YMQ error/overflow paths.

  • Web UI modernization: the Web UI is rewritten and refined with follow-up fixes and usability improvements.

Compatibility Notes

  • Client.map() and Client.starmap() align with Python semantics introduced in recent releases; deprecated tuple-style map usage was removed from tests and examples.

  • Ray compatibility remains available and aligned with current scheduler policy configuration patterns.