●Change8

Migrating to Ray ray-2.53.0

Version ray-2.53.0 introduces 2 breaking changes. This guide details how to update your code.

Released: 12/20/2025

2
Breaking Changes
6
Migration Steps
16
Affected Symbols

⚠️ Check Your Code

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

Dataset.summaryread_parquet_bulkshould_continue_on_errorDefaultCollateFnarrow_batch_to_tensorsHashShuffleAggregatorApproximateTopKread_lanceBlockOutputBufferDeploymentHandleDeploymentResponseDeploymentResponseGeneratorAggregationFunctionmake_fastapi_class_based_viewScalingConfig.label_selectorRAY_DATA_CLUSTER_AUTOSCALER

Breaking Changes

●Issue #1

Support for Pydantic V1 is dropped starting with Ray 2.56.0; users must upgrade to Pydantic V2 or adjust code accordingly.

●Issue #2

Removal of the deprecated `read_parquet_bulk` API in Ray Data may break code that still calls it; replace with `read_parquet`.

Migration Steps

  1. 1
    Upgrade code to use Pydantic V2 before upgrading to Ray 2.56.0.
  2. 2
    Replace any calls to `read_parquet_bulk` with `read_parquet`.
  3. 3
    Set environment variable `RAY_DATA_CLUSTER_AUTOSCALER=V2` to enable the new autoscaler.
  4. 4
    If using custom autoscaling enums in Serve YAML, ensure they are serializable or upgrade to the fixed version.
  5. 5
    Update any custom batch size logic to use the new custom batch size function signature if applicable.
  6. 6
    For Ray Train, use `label_selector` in `ScalingConfig` instead of previous placement mechanisms.

Release Summary

This release adds major new features such as a utilization‑based autoscaler, Kafka datasource, and deployment topology visibility, while dropping Pydantic V1 support and removing deprecated APIs.

Need More Details?

View the full release notes and all changes for Ray ray-2.53.0.

View Full Changelog