●Change8

Migrating to Ray ray-2.44.0

Version ray-2.44.0 introduces 3 breaking changes. This guide details how to update your code.

Released: 3/21/2025

3
Breaking Changes
6
Migration Steps
29
Affected Symbols

⚠️ Check Your Code

If you use any of these symbols, you need to read this guide:

ray.compiled_graphray.workflowsray.data.datasetray.data.operatorray.data.OpRuntimeMetricsray.data.map_operatorray.data.filter_exprray.data.override_num_blocksray.data.StandardScalerray.train.StateExportAPIray.train.XGBoostTrainerray.train.ScalingConfigray.train.TrainingFailedErrorray.train.FailurePolicyray.train.torch_ampray._private.storageray.serve.logging.additional_log_standard_attrsray.serve.llm.asyncacheray.serve.llm.cachetoolsray.serve.llm.backoffray.serve.llm.asyncio_timeoutray.serve.llm.jsonrefray.serve.llm.AutoscalingConfigray.serve.llm.DeploymentConfigray.serve.llm.pyarrow_fsray.serve.llm.usage_telemetryray.rllib.LearnerGroupray.rllib.Learnerray.rllib.EnvRunner

Breaking Changes

●Issue #1

The Ray storage dependency has been removed and the RAY_STORAGE environment variable configuration option has been deprecated. Code that imports from ray.storage or relies on RAY_STORAGE must be updated to use the new storage mechanisms or configuration APIs.

●Issue #2

Preprocessors now output a single column instead of one column per feature. Existing pipelines that expect multiple columns per feature need to be adjusted to handle the single-column output.

●Issue #3

Torch AMP wrapper utilities have been deprecated and may be removed in a future release, causing import errors for code that uses them.

Migration Steps

  1. 1
    If you rely on Ray Workflows, pin your Ray version to 2.44 until the library is removed.
  2. 2
    Replace any imports from ray.storage and stop using the RAY_STORAGE environment variable; use the new storage APIs provided by the individual libraries.
  3. 3
    Update preprocessing pipelines to handle the new single‑column output format.
  4. 4
    Remove imports of deprecated Torch AMP wrapper utilities or replace them with native PyTorch AMP usage.
  5. 5
    Adjust Serve LLM dependencies if you explicitly imported asyncache, cachetools, backoff, or asyncio_timeout.
  6. 6
    Review and update any code that assumes multiple columns per feature from vectorizers or hashers.

Release Summary

Ray 2.45 introduces the beta Ray Compiled Graph for ultra‑low‑overhead task graphs, adds Iceberg write support and LLM enhancements to Ray Data, a state export API to Ray Train, and numerous performance and usability improvements across Serve, RLlib, and other libraries, while deprecating Ray Workflows and removing the ray.storage dependency.

Need More Details?

View the full release notes and all changes for Ray ray-2.44.0.

View Full Changelog