acoustic-carpenter-78188
04/24/2026, 4:23 PMray from 2.54.0 to 2.55.0
Release notes
Sourced from ray's releases.
Ray-2.55.0
Ray Data
🎉 New Features
• AddAPI with scanner/reader framework, file listing, and file partitioning (#61220, #61615, #61997)DataSourceV2
• Support GPU shuffle with26.2 (#61371, #62062)rapidsmpf
• Add Kafka datasink, migrate to, supportconfluent-kafkaoffsets (#60307, #61284, #60909)datetime
• Add Turbopuffer datasink (#58910)
• Add 2-phase commit checkpointing with trie recovery and load method (#61821, #60951)
• Queue-based autoscaling policy integrated with task consumers (#59548, #60851)
• Enable autoscaling for GPU stages (#61130)
• Expressions: add,random(),uuid(), and map namespace support (#59656, #60695, #59879)cast
• Add support for Arrow native fixed-shape tensor type (#56284)
• Support writing tensors to tfrecords (#60859)
• Addsupport topathlib.Pathfunctions (#61126)read_*
• Addas acudf(#61329)batch_format
• AllowforActorPoolStrategyviaread_datasource()parameter (#59633)compute
• Introducefor streamlined caching (#60996)ExecutionCache
• Supportmode forstrict=False(#60295)StreamingRepartition
• Port changes from lance-ray into Ray Data (#60497)
• Enable PyArrow compute-to-expression conversion for predicate pushdown (#61617)
• Add vLLM metrics export and Data LLM Grafana dashboard (#60385)
• Include logical memory in resource manager scheduling decisions (#60774)
• Add monotonically increasing ID support (#59290)
💫 Enhancements
• Performance: cacheargs, heap-based actor ranking, actor pool map improvements (#61996, #62114, #61591)_map_task
• Optimize concat tables and PyArrow schema hashing (#61315, #62108)
• Reduce defaultthreshold to 50% (#61890)DownstreamCapacityBackpressurePolicy
• Improve reproducibility for random APIs (#59662)
• Clamp batch size to fall within C++ 32-bit int range (#62242)
• Account for external consumer object store usage in resource manager budget (#62117)
• Makeconfigurable in number of fragments to scan (#61670)get_parquet_dataset
• Consolidate schema inference and make all preprocessors implement(#61213, #61341)SerializablePreprocessorBase
• Disable hanging issue detection by default (#62405)
• Make execution callback dataflow explicit to prevent state leakage (#61405)
• Login JSON format at execution start for traceability (#61150, #61428)DataContext
• Autoscaler: configurable traceback, Prometheus gauges, relaxed constraints (#62210, #62209, #61917, #61385)
• Add metrics for task scheduling time, output backpressure, and logical memory (#61192, #61007, #61436)
• Prevent operators from dominating entire shared object store budget (#61605)
• Eliminate generators to avoid intermediate state pinning (#60598)
• Default log encoding to UTF-8 on Windows (#61143)
• Remove legacy,BlockList, old callback API, PyArrow 9.0 checks (#60575, #61044, #62055, #61483)locality_with_output
• Upgrade to0.11.0; cappyicebergto <3 (#61062, #60406)pandas
• Refactor logical operators to frozen dataclasses (#61059, #61308, #61348, #61349, #61351, #61364, #61481)
• Prevent aggregator head node scheduling (#61288)
• Add error forpaths with a zero-resource head node (#60709)local://
🔨 Fixes... (truncated) Commits •
58af3fc Cherry picking to 2.55 (<https://redirect.git…
flyteorg/flyte-sdk