NewYour coding agent can read the release notes before it upgrades.Set up the MCP server →
PyPI, #978 most downloaded on PyPI
Ray provides a simple, universal API for building distributed applications.
Latest stable
2.60.0
Last release 12 days ago
04 Nov 2026
Ships fairly regularly
a new release about every 3 weeks
Nearly every release is documented
notes for 59 of the last 60 stable releases
1 version withdrawn
withdrawn after publishing
9 years old
137 releases · first in 2017
Drag across the ruler to compare, or tap a release to open it.
Remove deprecated Logger interface and logger_creator
DataSourceV2 API with scanner/reader framework, file listing, and file partitioning (#61220, #61615, #61997)rapidsmpf 26.2 (#61371, #62062)confluent-kafka, support datetime offsets (#60307, #61284, #60909)random(), uuid(), cast, and map namespace support (#59656, #60695, #59879)pathlib.Path support to read_* functions (#61126)cudf as a batch_format (#61329)ActorPoolStrategy for read_datasource() via compute parameter (#59633)ExecutionCache for streamlined caching (#60996)strict=False mode for StreamingRepartition (#60295)_map_task args, heap-based actor ranking, actor pool map improvements (#61996, #62114, #61591)DownstreamCapacityBackpressurePolicy threshold to 50% (#61890)get_parquet_dataset configurable in number of fragments to scan (#61670)SerializablePreprocessorBase (#61213, #61341)DataContext in JSON format at execution start for traceability (#61150, #61428)BlockList, locality_with_output, old callback API, PyArrow 9.0 checks (#60575, #61044, #62055, #61483)pyiceberg 0.11.0; cap pandas to <3 (#61062, #60406)local:// paths with a zero-resource head node (#60709)StreamingSplitDataIterator.schema() (#62057)ParquetDatasource handling of FileSystemFactory.inspect (#62065)read_parquet file-extension filtering for versioned object-store URIs (#61376)wide_schema_pipeline_tensors cloudpickle deserialization (#62149)OpBufferQueue race condition (#60828)OneHotEncoder max_categories to use global top-k instead of per-partition (#60790)ReservationOpResourceAllocator resource borrowing for ActorPoolMapOperator (#60882)DatabricksUCDatasource schema() shadowing by schema string attribute (#61282)AliasExpr structural equality to respect rename flag (#60711)_align_struct_fields failure with unaligned scalar fields (#58364)min_scheduling_resources fallback to incremental_resource_usage (#60997)ref_bundle + input_files (#61774)on_exit hook with __ray_shutdown__ to fix UDF cleanup race (#61700)Limit from getting pushed past map_groups (#60881)_shuffle_block to fix ColumnNotFound in chained left joins (#61507)TimeWindowAverageCalculator (#61580)end_offset in Kafka datasource (#61476)train_test_split (#60274)None when no outputs have been produced (#62029)raise with TypeError in string concatenation (#60795)exclude_resources docs for Train autoscaling changes (#61990)locality_with_output migration instructions (#61151)max_tasks_in_flight_per_actor vs max_concurrent_batches (#60477)MOD operation docs; improve ray.data.Datasource docs (#60803, #59654)polars usage instructions (#60029)policy_kwargs, so advanced users can package reusable autoscaling logic without custom forks. (#60964)AsyncInferenceAutoscalingPolicy documentation and clarified Serve performance guidance for HAProxy and inter-deployment gRPC use cases. (#61086, #61386)status attribute to ReportedCheckpoint (#61684)torchft environment (#61156)AutoscalingCoordinator in FixedScalingPolicy (#61703)datasets field from TrainRunContext (#61953)checkpoint_upload_fn when slow (#61720)StateManagerCallback to accept datasets explicitly (#62042)before_controller_shutdown (#61816)RayActorError (#61375)sync_actor to use wait_with_logging (#61063)UserExceptionWithTraceback in WorkerGroupError.worker_failures (#61153)PlacementGroupCleaner zombie actor (#61756)Logger interface and logger_creator (#61181)NaN values are present (#57160)PDProxyServer with decode-as-orchestrator PD architecture (#62076)bundle_per_worker config for simpler placement group setup (#59903)PlacementGroup config schemes (#62241)num_cpus=0 to reduce contention on low-CPU machines (#61191)format_messages_to_prompt with _build_chat_messages (#61117, #61372)data: [DONE] in streaming SSE responses (#62246)enable_log_requests=False not forwarded to vLLM AsyncLLM (#60824)OpenAiIngress scale-to-zero when all models set min_replicas=0 (#60836)init_app_state (#60812)trust_remote_code download (#60344)TRANSFORMERS_CACHE; treat HuggingFace config load failure as non-fatal (#60854)TorchRLModule forward passes (#61985)AlgorithmConfig (#61233)eval_results, env_steps, or agent_steps (#61563)PrioritizedEpisodeReplayBuffer bug (#60065)LayerNorm in RLModuleSpec (#61025)MultiAgentEpisode.env_t_to_agent_t (#60319)torch_learner.py crash under parameter-freezing edge cases (#62158)ResizeRayletResourceInstances to GCS/Python client, schema/status models, KubeRay provider (#61654, #61666, #61803, #61814)PlatformEvent proto and placement group events in one-event framework (#61701, #60449)Percentile metric type backed by quadratic histogram (#61148)fallback_strategy in TaskInfoEntry and ActorTableData (#60659)ray.put() generic: put(value: R) -> ObjectRef[R] (#60995)cloudpickle to 3.1.2, gRPC to v1.58.0, protobuf to 3.20.3 (#60317, #61499, #60736)pg.ready() performance via async GCS RPC; fix deadlocks (#60657, #62086)ActorHandle.__hash__ and fix __eq__ correctness (#61638)find_gcs_addresses (#61065)get_all_node_info (#61232)getName/setDaemon) (#62153)@ray.remote/@ray.method with num_returns (#59286)StopIteration on non-generator functions to RuntimeError (#60521)AuthenticationValidator in sync server (#60778, #60779)local_mode (#60647)worker_process_setup_hook on re-entry (#61473)memory_info instead of calling memory_full_info (#60000)NodeManager and InternalKVManager (#61002)HandleUnregisterNode/HandleDrainNode (#62226, #62112)bool env var parsing for RAY_CGRAPH_overlap_gpu_communication (#61421)OnNodeDead to destroy all owned actors when owner node dies (#60669)TASK_PROFILE_EVENT aggregation for multiple phases (#61559)WorkerPool::WarnAboutSize() (#61246)TaskLifecycleEvent.node_id using emitting node instead of executor (#61478)publisher_id type mismatch in GCS pubsub (#61518)dataclass.asdict with None in dashboard list_jobs API (#61033)ReadOnlyProvider.terminate() signature mismatch (#62251)set/get env races in OtlpGrpcMetricExporterOptions and metrics exporter init (#61034, #61281)ray start (#61837)ray.init() (#61029)Node._node_labels initializes regardless of connect_only (#61618)WrongClusterID on head restart (#60860)**kwargs through JobSubmissionClient to cluster info resolvers (#61902)torch to 2.7.0+cu128 and torchvision (#61328)jackson-databind 2.16.1 -> 2.18.6 (GHSA-72hv-8253-57qq) (#61808)build-image.sh and CLI for local Docker image builder (#61042, #61338)setup-dev.py (#61357)RAY_BACKEND_LOG_JSON environment variable documentation (#59962)RayCluster name as ServiceAccount name for RBAC authentication (#61785)RayCluster (#61719)development.rst with image build, wheel paths, and cross-references (#61500, #61501, #61504, #61596)Many thanks to all those who contributed to this release!
@justinyeh1995, @marwan116, @jddqd, @MkDev11, @mjd3, @XuQianJin-Stars, @elliot-barn, @DeborahOlaboye, @aaronscalene, @rayhhome, @ayushk7102, @bj-son, @nadongjun, @Daraan, @xinyuangui2, @Sparks0219, @justinvyu, @suppagoddo, @akyang-anyscale, @ambicuity, @Aydin-ab, @mickeyyliu, @MatthewCWeston, @vaishdho1, @jinbum-kim, @eicherseiji, @kouroshHakha, @karticam, @JasonLi1909, @ArturNiederfahrenhorst, @moktamd, @nrghosh, @dragongu, @andrewsykim, @mgchoi239, @ruoliu2, @harshit-anyscale, @Chong-Li, @pseudo-rnd-thoughts, @lee1258561, @khluu, @daiping8, @SolitaryThinker, @jonalee99, @yancanmao, @SohamRajpure, @rueian, @VitaliyEroshin, @Future-Outlier, @nehiljain, @JiangJiaWei1103, @Yicheng-Lu-llll, @KaisennHu, @jeffreywang-anyscale, @aslonnie, @alanwguo, @machichima, @limarkdcunha, @codope, @sampan-s-nayak, @kyuds, @thjung123, @abrarsheikh, @wingkitlee0, @preneond, @7ckingBest, @slfan1989, @win5923, @kaori-seasons, @israbbani, @andrew-anyscale, @zestze, @owenowenisme, @edoakes, @laysfire, @pushpavanthar, @tohtana, @leewyang, @liulehui, @Hyunoh-Yeo, @eureka0928, @ryanaoleary, @947132885, @Kunchd, @simonsays1980, @dpj135, @bveeramani, @raulchen, @Partth101, @dubin555, @richabanker, @bittoby, @sai-miduthuri, @RedGrey1993, @kamil-kaczmarek, @TimothySeah, @myandpr, @rishic3, @justinrmiller, @HassamSheikh, @chiayi, @petern48, @carolynwang, @MrKWatkins, @400Ping, @summaryzb, @peterxcli, @RocMarshal, @coqian, @yuhuan130, @ryankert01, @dayshah, @Anarion-zuo, @ZacAttack, @weimingdiit, @iamjustinhsu, @matthewdeng, @goutamvenkat-anyscale, @KeeProMise, @Sanskarzz, @yuchen-ecnu, @praneethkaturi, @rajeshg007, @ankur-anyscale, @Art0white, @xyuzh, @dancingactor, @MengjinYan, @dengkliu92, @alexeykudinkin
Your coding agent can read these notes before it upgrades. Set up the MCP server →
DataSourceV2 API with scanner/reader framework, file listing, and file partitioning (#61220, #61615, #61997)rapidsmpf 26.2 (#61371, #62062)confluent-kafka, support datetime offsets (#60307, #61284, #60909)random(), uuid(), cast, and map namespace support (#59656, #60695, #59879)pathlib.Path support to read_* functions (#61126)cudf as a batch_format (#61329)ActorPoolStrategy for read_datasource() via compute parameter (#59633)ExecutionCache for streamlined caching (#60996)strict=False mode for StreamingRepartition (#60295)_map_task args, heap-based actor ranking, actor pool map improvements (#61996, #62114, #61591)DownstreamCapacityBackpressurePolicy threshold to 50% (#61890)get_parquet_dataset configurable in number of fragments to scan (#61670)SerializablePreprocessorBase (#61213, #61341)DataContext in JSON format at execution start for traceability (#61150, #61428)BlockList, locality_with_output, old callback API, PyArrow 9.0 checks (#60575, #61044, #62055, #61483)pyiceberg 0.11.0; cap pandas to <3 (#61062, #60406)local:// paths with a zero-resource head node (#60709)StreamingSplitDataIterator.schema() (#62057)ParquetDatasource handling of FileSystemFactory.inspect (#62065)read_parquet file-extension filtering for versioned object-store URIs (#61376)wide_schema_pipeline_tensors cloudpickle deserialization (#62149)OpBufferQueue race condition (#60828)OneHotEncoder max_categories to use global top-k instead of per-partition (#60790)ReservationOpResourceAllocator resource borrowing for ActorPoolMapOperator (#60882)DatabricksUCDatasource schema() shadowing by schema string attribute (#61282)AliasExpr structural equality to respect rename flag (#60711)_align_struct_fields failure with unaligned scalar fields (#58364)min_scheduling_resources fallback to incremental_resource_usage (#60997)ref_bundle + input_files (#61774)on_exit hook with __ray_shutdown__ to fix UDF cleanup race (#61700)Limit from getting pushed past map_groups (#60881)_shuffle_block to fix ColumnNotFound in chained left joins (#61507)TimeWindowAverageCalculator (#61580)end_offset in Kafka datasource (#61476)train_test_split (#60274)None when no outputs have been produced (#62029)raise with TypeError in string concatenation (#60795)exclude_resources docs for Train autoscaling changes (#61990)locality_with_output migration instructions (#61151)max_tasks_in_flight_per_actor vs max_concurrent_batches (#60477)MOD operation docs; improve ray.data.Datasource docs (#60803, #59654)polars usage instructions (#60029)policy_kwargs, so advanced users can package reusable autoscaling logic without custom forks. (#60964)AsyncInferenceAutoscalingPolicy documentation and clarified Serve performance guidance for HAProxy and inter-deployment gRPC use cases. (#61086, #61386)status attribute to ReportedCheckpoint (#61684)torchft environment (#61156)AutoscalingCoordinator in FixedScalingPolicy (#61703)datasets field from TrainRunContext (Note truncated.
Disable hanging issue detection (#61895) — The hanging issue detector was making blocking calls to the Ray State API, which could cause the scheduling
🔨 Fixes
Remove deprecated TENSOR_COLUMN_NAME constant
🎉 New Features
sql_params support to read_sql (#60030)AsList aggregation (#59920)CountDistinct aggregate (#59030)UDFExpr (#56725)write_parquet() (#59102)💫 Enhancements
preserve_order (#60555)DefaultClusterAutoscalerV2 thresholds via env vars (#60133)DownstreamCapacityBackpressurePolicy (#59753)node_id, pid, attempt number for hanging tasks (#59793)OpRuntimeMetrics for progress (#60304)OpMetrics logging (#59907)TENSOR_COLUMN_NAME constant (#60573)meta_provider parameter (#60379)ray.data imports (#60292)StatefulShuffleAggregation.finalize allow incremental streaming (#59972)OutputSplitter semantics to avoid unnecessary buffer accumulation (#60237)BackpressurePolicy to streaming executor progress bar (#59637)StandardScaler preprocessor with Arrow format (#59906)🔨 Fixes
MapBatches even if they modify the row count (#60756)map_batches by default (#60448)ActorPoolMapOperator to guarantee dispatch of all given inputs (#60763)ArrowInvalid error when backfilling missing fields from map tasks (#60643)UnionOperator.clear_internal_output_queue (#60538)DefaultClusterAutoscalerV2 raising KeyError: 'CPU' (#60208)ReorderingBundleQueue handling of empty output sequences (#60470)DefaultAutoscalerV2 not scaling nodes from zero (#59896)use_ray_tqdm (#59996)StreamingRepartition hang with empty upstream results (#59848)AutoscalingCoordinator double-allocating resources for multiple datasets (#59740)DownstreamCapacityBackpressurePolicy issues (#59990)AutoscalingCoordinator crash when requesting 0 GPUs on CPU-only cluster (#59514)TensorArray to Arrow tensor conversion (#59449)max_actors is set (#59632)📖 Documentation
resource_limits refers to logical resources (#60109)read_lance doc (#59673)read_unity_catalog docstring (#59745)enable_true_multi_threading (#60515)🎉 New Features
QueueMonitor actor that queries message brokers (Redis, RabbitMQ) for queue length, enabling TaskConsumer scaling based on pending tasks rather than HTTP load. (#59430)apply_autoscaling_config decorator allows custom autoscaling policies to automatically benefit from Ray Serve's standard parameters (delays, scaling factors, bounds) without reimplementation. (#58857)label_selector and bundle_label_selector in Serve deployments. Deployments can now specify node label selectors for scheduling and bundle-level label selectors for placement groups, useful for targeting specific hardware (e.g., TPU topologies). (#57694)serve_autoscaling_snapshot log per autoscaling-enabled deployment each control-loop tick, with an event summarizer that reduces duplicate logs. (#56225)@serve.batch. (#59334)💫 Enhancements
lookback_period_s must now be greater than metrics_interval_s, preventing silent misconfigurations. (#59456)root_path support for uvicorn. root_path now works correctly across all uvicorn versions, including >=0.26.0 which changed how root_path is processed. (#57555)serve.shutdown(), eliminating cross-library import dependencies. (#60067)ray.init() when Pydantic v1 is detected, as support will be removed in Ray 2.56. (#59703)🔨 Fixes
_ray_trace_ctx when calling actors from a different process than the one that created them (e.g., serve start + dashboard interaction). (#59634)TaskProcessorAdapter shutdown during rolling updates. Removed shutdown() from __del__, which was broadcasting a kill signal to all Celery workers instead of just the local one, breaking rolling updates. (#59713)test_router_queue_len_metric, ensured proxy replica queue cache is populated before GCS failure tests, and added metrics server readiness checks. (#60333, #60466, #60468)📖 Documentation
🏗 Architecture refactoring
RAY_SERVE_DEFAULT_HTTP_HOST, RAY_SERVE_DEFAULT_HTTP_PORT, RAY_SERVE_DEFAULT_GRPC_PORT, RAY_SERVE_HTTP_KEEP_ALIVE_TIMEOUT_S, RAY_SERVE_REQUEST_PROCESSING_TIMEOUT_S, RAY_SERVE_ENABLE_JSON_LOGGING, RAY_SERVE_ALWAYS_RUN_PROXY_ON_HEAD_NODE), cleaned up legacy constant fallbacks, and added documentation for previously undocumented env vars (e.g., RAY_SERVE_CONTROLLER_MAX_CONCURRENCY, RAY_SERVE_ROOT_URL, proxy health check settings, and fault tolerance params). Users relying on removed env vars should migrate to the Serve config API (http_options, grpc_options, LoggingConfig). (#59470, #59619, #59647, #59963, #60093)🎉 New Features
CallbackManager and guardrail some callback hooks (#60117)💫 Enhancements
Predictor API (#60305)PlacementGroup and SlicePlacementGroup interface in WorkerGroup (#60116)RayTrainWorker actors (#59872)pg.ready() with pg.wait() in worker group (#60568)DatasetsSetupCallback to DatasetsCallback (#59423)🔨 Fixes
try-except for pg.wait() (#60743)TrainController reraises AsyncioActorExit (#59461)📖 Documentation
JaxTrainer template (#59842)checkpoint_upload_fn backend and cuda:nccl backend support (#60541)checkpoint_upload_func to checkpoint_upload_fn in docs (#60390)🔨 Fixes
🎉 New Features
/tokenize and /detokenize endpoints (#59787)/collective_rpc endpoint for RLHF weight synchronization (#59529)pooling parameter (#59534)guided_decoding (#59421)should_continue_on_error support for ServeDeploymentStage (#59395)HttpRequestUDF resources (#60313)💫 Enhancements
world_size==1 (#60403)compute instead of concurrency to specify ActorPool size (#59645)DataContext overrides in Ray Data LLM Processor (#60142)torch.Tensor serialization overhead (#59919)PrefixCacheAwareRouter imbalance threshold less surprising (#59390)tokenized_prompt without prompt in vLLMEngineStage (#59801)fn_constructor_kwargs (#59806)CUDA_VISIBLE_DEVICES deletion workaround (#60502)🔨 Fixes
Namespace conversion in vLLM engine initialization (#60380)ndarray exception in http_request_stage (#60299)EngineDeadError to enable recovery (#60145)📖 Documentation
vLLMEngineProcessor (#59446)🎉 New Features
BC and MARWIL (#59067)💫 Enhancements
np.nanmean warnings in EMA stats (#60408)🔨 Fixes
RLModule forward methods to handle dict spaces (#60451)LearnerGroup.load_module_state() and mark as deprecated (#60354)AlgorithmConfig (#59438)flatten_observations.py for nested spaces for ignored multi-agent (#59928)🎉 New Features
cgroup_path in ray.init() (#59372, #60183, #60726)is_canceled() (#58914)--entrypoint-resource (#59735)--ip option in ray attach (#59931)ray kill-actor --name/--namespace for force/graceful shutdown (#60258)💫 Enhancements
RAY_AUTH_MODE=k8s with separate config for Kubernetes token auth (#59621)shared_ptr caching and avoid per-RPC construction (#59500)OpenTelemetry metric recording calls (#59337)working_dir uploads (#59566)DEBUG in RaySyncer (#59616)std::unordered_map to absl::flat_hash_map (#59921)NodeDefinitionEvent proto (#60314)repr_name to actor_lifecycle_event (#59925)ALL in exposable event config (#59878)SubprocessModuleHandle.destroy_module() resource cleanup (#60172)is_head in dashboard agent startup (#59378)get_session_name() to RuntimeContext (#59469)MAX_APPLICATION_ERROR_LEN configurable via env var (#59543)🔨 Fixes
idle_time_ms resetting for nodes not running tasks (#60581)RAY_EXPERIMENTAL_NOSET_* environment variable parsing in accelerator managers (#60577)ray start --no-redirect-output crash (#60394)PSUTIL_PROCESS_ATTRS returning empty list on Windows (#60173)available_node_types on on-prem clusters (#60184)MetricsAgentClient exporter initialization (#59611)internal_ip() within StandardAutoscaler (#57279)uv_runtime_env_hook.py to pin worker Python version (#59768)STRICT_PACK placement groups ignoring bundle label selectors (#60170)psutil internal API usage in dashboard disk usage reporting (#59659)RUNNING vs FINISHED metrics (#59893)symmetric_run using wrong condition to check GCS readiness (#59794)TypeError when using bundle_label_selectors (#59850)📖 Documentation
ray.shutdown() behavior for local vs remote clusters (#59845)RAY_RUNTIME_ENV_BEARER_TOKEN env var (#60136)💫 Enhancements
🔨 Fixes
ray-cpp wheels are now py3-none, without specific Python versions. (#59969)RayJob InTreeAutoscaling with Kueue docs after Kueue 0.16.0 release (#59648)RunLLM chat widget for Ray docs (#59126)Thank you to everyone who contributed to this release! @KaisennHu, @MiXaiLL76, @slfan1989, @krisselberg, @JasonLi1909, @Priya-753, @pseudo-rnd-thoughts, @zzchun, @ZacAttack, @pushpavanthar, @jjyao, @ryanaoleary, @pcmoritz, @akshay-anyscale, @HassamSheikh, @yurekami, @Hyunoh-Yeo, @ruoliu2, @nrghosh, @wxwmd, @myandpr, @J-Meyers, @trilamsr, @kouroshHakha, @limarkdcunha, @manhld0206, @jreiml, @preneond, @yuchen-ecnu, @Yicheng-Lu-llll, @AchimGaedkeLynker, @vaishdho1, @israbbani, @OneSizeFitsQuorum, @Sathyanarayanaa-T, @nadongjun, @xinyuangui2, @Rob12312368, @as-jding, @lee1258561, @popojk, @coqian, @rajeshg007, @jeffreywang-anyscale, @kamil-kaczmarek, @alexeykudinkin, @Aydin-ab, @mgchoi239, @dragongu, @edoakes, @smortime, @tk42, @abrarsheikh, @jakubzimny, @Future-Outlier, @axreldable, @owenowenisme, @g199209, @cem-anyscale, @dayshah, @akelloway, @daiping8, @dlwh, @robertnishihara, @400Ping, @matthewdeng, @antoine-galataud, @cristianjd, @Partth101, @goutamvenkat-anyscale, @codope, @seanlaii, @andrew-anyscale, @andrewsykim, @liulehui, @simonsays1980, @Sparks0219, @yifanmai, @landscapepainter, @win5923, @kangwangamd, @srinarayan-srikanthan, @KeeProMise, @srinathk10, @my-vegetable-has-exploded, @MengjinYan, @yancanmao, @yuhuan130, @ArturNiederfahrenhorst, @akyang-anyscale, @rushikeshadhav, @kongjy, @harshit-anyscale, @justinvyu, @dancingactor, @Vito-Yang, @cr7258, @marwan116, @muyihao, @DeborahOlaboye, @bveeramani, @kriyanshii, @khluu, @machichima, @Kunchd, @jonded94, @iamjustinhsu, @sampan-s-nayak, @wingkitlee0, @sunsetxh, @dkhachyan, @can-anyscale, @TimothySeah, @raulchen, @elliot-barn, @ryankert01, @xyuzh, @stephanie-wang, @hao-aaron, @simeetnayan81, @cszhu, @richardliaw, @yuanjiewei, @kyuds, @eicherseiji, @RedGrey1993, @rueian, @jeffreyjeffreywang, @crypdick, @ankur-anyscale, @aslonnie
Deprecated API removal: Remove deprecated read_parquet_bulk API
🎉 New Features
Dataset.summary() API for quick dataset inspection (#58862)should_continue_on_error for graceful error handling in batch inference (#59212)with_column expressions: Enable expressions for grouped with_column in Ray Data (#58231)DefaultCollateFn, arrow_batch_to_tensors (#58821)💫 Enhancements
__init__ (#59105)iter_batches (#58657)HashShuffleAggregator breaks down blocks on finalize (#58603)ApproximateTopK aggregator (#58659)read_lance() (#58895)time_to_first_batch and get_ref_bundles metrics to data dashboard (#58912)iter_prefetched_bytes statistics tracking (#58900)iter_batches: Add configurable batching for resolve_block_refs to speed up iter_batches (#58467)read_parquet_bulk API (#58970)🔨 Fixes
obj_store_mem_max_pending_output_per_task reporting (#58864)get_parquet_dataset (#57047)📖 Documentation
vision_preprocess and vision_postprocess in VLM docs (#59012)huggingface_hub instruction (#59109)🎉 New Features
external_scaler_enabled flag to application config, enabling third-party autoscalers to control replica counts. (#57727, #57698)@serve.batch, useful when batch items have varying weights (e.g., token counts in LLM inference). (#59059)ray_serve_deployment_target_replicas, ray_serve_autoscaling_decision_replicas), batching statistics, and router queue latency for improved observability. (#59220, #59232, #59233)💫 Enhancements
GET, PUT) corresponding to a route. (#58927)@ingress now preserve original class metadata (__qualname__, __module__, __doc__, __annotations__). (#58478)DeploymentHandle, DeploymentResponse, and DeploymentResponseGenerator for better IDE support and type inference. Adds .result() stub to DeploymentResponseGenerator to fix static typing errors. (#59363, #58522)🔨 Fixes
RepresenterError when using serve build with AggregationFunction enum values in autoscaling config. (#58509)last_scale_up_time and last_scale_down_time on autoscaling context. (#59057)DeploymentResponse objects in a chain of deployment calls from different event loops. (#59385)make_fastapi_class_based_view to properly handle inherited methods. (#59410)📖 Documentation
🎉 New Features
label_selector to ScalingConfig. This allows users to control worker placement by targeting specific labeled nodes in the cluster. (#58845, #59414)JaxTrainer running on GPU machines. (#58322)CheckpointConsistencyMode to get_all_reported_checkpoints, providing options for handling checkpoint retrieval consistency. (#58271)DataConfig now supports setting execution_options on a per-dataset basis for finer-grained control over data loading. (#58717)💫 Enhancements
Result.get_best_checkpoint now supports nested metrics, allowing for more flexible metric tracking and checkpoint selection. (#58537)get_all_reported_checkpoints no longer blocks when only metrics are reported. (#58870)🔨 Fixes
setup_mlflow API to ensure full compatibility with Ray Train V2. (#58705)ValueError is now raised if checkpoint_upload_fn fails to return a valid checkpoint. (#58863)📖 Documentation
ray.train.get_all_reported_checkpoints method. (#58946)💫 Enhancements:
Result.get_best_checkpoint now supports nested metrics, allowing for more flexible metric tracking and checkpoint selection. (#58537)💫 Enhancements
transformers to 4.57.3 (#58980)vllm_engine.py to check for VLLM_USE_V1 attribute (#58820)VLLM_RAY_PER_WORKER_GPUS from fractional placement-group bundles automatically (#58949)🔨 Fixes
🎉 New Features
num_env_runners \> 0 (#58495)💫 Enhancements
MetricsLogger tweaks+ Stats rewrite (#56838)EnvRunner (#56750)AlgorithmConfig deprecated argument with incorrect behavior/semantics (#59138)examples/ (#58893)/component/tests (#58890)pytest_runtest_makereport) across tests (#59003)asv.conf.json (#58934)byod_rllib.sh (#59157)🔨 Fixes
📖 Documentation
🎉 New Features
RAY_ENABLE_ZERO_COPY_TORCH_TENSORS (#57639).rayignore file support for controlling cluster uploads (#58500)AuthenticationError exception (#58737)X-Ray-Authorization fallback header for auth token in dashboard (#58819)💫 Enhancements
RAY_core_worker_num_server_call_thread (#58771)OtlpGrpcMetricExporter wrapper to log export failures (#58929)get_if_exists=True for actor lookup (#58628)AuthenticationError from Python for token loading errors (#59031)secrets.token_hex(32) to generate auth tokens (#58818)AUTH_MODE=token check in get-auth-token CLI (#58848)🔨 Fixes
grpc_authentication_server_interceptors streaming response handling (#59104)IsProcessAlive on Windows (#59106)RAY_enable_open_telemetry (#59095)RAY_DISABLE_FAILURE_SIGNAL_HANDLER option (#58984)RayletClient causing driver crash (use-after-free) (#58660)shared_ptr for pins_in_flight_ to prevent use-after-free (#58744)add_command_alias (#58719)cluster_full_of_actors_detected_* fields (unused in autoscaler v2) (#59052)📖 Documentation
token-auth.md documentation page (#58829)💫 Enhancements
time_to_first_batch and get_ref_bundles metrics to data dashboard (#58912)rich, cupy-cuda12x, and memray (#58983)lxml to 6.0.2 (#58808)requests from 2.32.3 to 2.32.5 (#58724)openlineage-python in the dependency set (#58724)Thank you to everyone who contributed to this release! @xinyuangui2, @harshit-anyscale, @Sparks0219, @israbbani, @siyuanfoundation, @robertnishihara, @thomasdesr, @spencer-p, @aslonnie, @ZacAttack, @soodoshll, @marosset, @simeetnayan81, @soffer-anyscale, @abrarsheikh, @400Ping, @richo-anyscale, @as-jding, @rueian, @kshanmol, @yancanmao, @zzchun, @coqian, @matthewdeng, @Future-Outlier, @YoussefEssDS, @ykdojo, @pseudo-rnd-thoughts, @lowdy1, @ArturNiederfahrenhorst, @myandpr, @komikndr, @machichima, @RisinT96, @curiosity-hyf, @alanwguo, @CaiZhanqi, @Aydin-ab, @MengjinYan, @suzuri-lollipop, @jeffreyjeffreywang, @rushikeshadhav, @alexeykudinkin, @meAmitPatil, @zcin, @teddygood, @elliot-barn, @dayshah, @srinathk10, @XLC127, @simonsays1980, @kevin85421, @bveeramani, @kunling-anyscale, @khluu, @andrew-anyscale, @KaisennHu, @kouroshHakha, @ryankert01, @pavitrabhalla, @jjyao, @dragongu, @SolitaryThinker, @justinrmiller, @wxwmd, @Haustle-v, @TimothySeah, @goutamvenkat-anyscale, @liulehui, @raulchen, @HassamSheikh, @Priya-753, @vaishdho1, @dancingactor, @daiping8, @eloaf, @JasonLi1909, @rayci-bot, @richardliaw, @SheldonTsen, @Yicheng-Lu-llll, @ktyxx, @pschmutz, @iamjustinhsu, @ahao-anyscale, @cem-anyscale, @eicherseiji, @edoakes, @rajeshg007, @arki05, @andrewsykim, @nrghosh, @ryanaoleary, @kyuds, @Daraan, @can-anyscale, @sampan-s-nayak, @xyuzh, @owenowenisme
More robust handling for CVE-2025-62593: test for more browser-specific headers in dashboard browser rejection logic
Clean up checkpoint config and trainer param deprecations
Ray Core:
Ray Data:
🎉 New Features:
💫 Enhancements:
🔨 Fixes:
📖 Documentation:
🏗 Architecture:
🎉 New Features:
💫 Enhancements:
📖 Documentation:
/api/users/{user_id}) instead of just route prefixes, enabling granular endpoint monitoring without high cardinality issues. Performance impact is minimal (~1% RPS decrease). (#58180)list_outbound_deployments() method to discover downstream deployment dependencies, enabling programmatic analysis of service topology for both stored and dynamically-obtained handles. (#58345, #58350)ReplicaRank schema with global, node-level, and local ranks to support advanced coordination scenarios like tensor parallelism and model sharding across nodes. (#58471, #58473)serve.run() completes, improving deployment reliability. (#57723)RAY_SERVE_THROUGHPUT_OPTIMIZED without manually configuring all flags, improving flexibility for performance tuning. (#58057)ray_serve_*) and improved observability infrastructure. (#56432)RunningReplicaInfo objects passed in long-poll updates, avoiding complex reference counting patterns. (#58174)IMPLICIT_RESOURCE_PREFIX from ReplicaConfig.ray_actor_options to prevent internal resource annotations from leaking into user-visible configurations. (#58275)from_proxy_manager argument to get_target_groups() for finer control over returned routing targets. (#57620)_TaskConsumerWrapper during async inference implementation. (#57664)serve run now respects proxy_location from config files instead of hardcoding EveryNode, and serve.start() no longer defaults to HeadOnly when http_options are provided without an explicit location. (#57622)stabilityai/stable-diffusion-2 was deprecated on Hugging Face. (#58609)RankManager class with type-safe ReplicaRank representation, creating a cleaner foundation for future multi-level rank support. (#58471, #58473)💫 Enhancements:
🎉 New Features:
💫 Enhancements:
🔨 Fixes:
📖 Documentation:
🎉 New Features:
💫 Enhancements:
🔨 Fixes:
📖 Documentation:
🏗 Architecture:
💫 Enhancements:
🔨 Fixes:
📖 Documentation:
🎉 New Features:
💫 Enhancements:
📖 Documentation:
Thank You to all the Contributors! @marosset, @curiosity-hyf, @bveeramani, @Future-Outlier, @saihaj, @ZacAttack, @ArthurBook, @crypdick, @Aydin-ab, @elliot-barn, @Kunchd, @justinvyu, @jjyao, @gangsf, @sunsetxh, @Daraan, @justinyeh1995, @MatthewCWeston, @kyuds, @daiping8, @sauravvenkat, @omatthew98, @CowKeyMan, @morotti, @israbbani, @goutamvenkat-anyscale, @fscnick, @Zakelly, @xyuzh, @kouroshHakha, @owenowenisme, @Qiaolin-Yu, @czgdp1807, @shen-shanshan, @wph95, @iamjustinhsu, @MengjinYan, @jugalshah291, @Yicheng-Lu-llll, @ryanaoleary, @nadongjun, @xinyuangui2, @ideal, @my-vegetable-has-exploded, @lucaschadwicklam97, @tianyi-ge, @ahao-anyscale, @abrarsheikh, @Blaze-DSP, @rueian, @thomasdesr, @CaiZhanqi, @harshit-anyscale, @jeffreyjeffreywang, @TimothySeah, @codope, @sampan-s-nayak, @andrewsykim, @xingsuo-zbz, @aslonnie, @OneSizeFitsQuorum, @ryankert01, @Sparks0219, @soffer-anyscale, @akyang-anyscale, @alanwguo, @chrisfellowes-anyscale, @richo-anyscale, @alexeykudinkin, @JasonLi1909, @ruisearch42, @EkinKarabulut, @MarcoGorelli, @SolitaryThinker, @srinathk10, @dayshah, @richardliaw, @pseudo-rnd-thoughts, @win5923, @axreldable, @matthewdeng, @ArturNiederfahrenhorst, @can-anyscale, @khluu, @landscapepainter, @kevin85421, @seanlaii, @edoakes, @nrghosh, @eicherseiji, @Artimislyy, @cem-anyscale, @coqian, @chiayi, @liulehui
Fix for CVE-2025-62593: reject Sec-Fetch-* other browser-specific headers in dashboard browser rejection logic
Reuse previous metadata if transferring the same tensor list with nixl
nixl (https://github.com/ray-project/ray/pull/58309)Improve deprecation handling when ray.train methods are called from ray.tune
Ray Train:
RAY_TRAIN_V2_ENABLED=0.Ray Serve:
AutoscalingContext with total_running_requests, total_queued_requests, and total_num_requests, plus adds support for min, max, and time-weighted average aggregation functions. These improvements give users fine-grained control to implement sophisticated custom autoscaling policies based on real-time workload metrics.🎉 New Features:
💫 Enhancements:
🔨 Fixes:
📖 Documentation:
🎉 New features
ray.train aliases for public APIs (#57758)💫 Enhancements
jax.distributed.shutdown() for JaxBackend (#57802)TrainingFailedError module (#57865)ray.train methods are called from ray.tune (#57810)🔨 Fixes
ControllerError triggered by after_worker_group_poll_status errors (#57869)iter_torch_batches use of ray.train.torch.get_device outside Train (#57816)ThreadRunner (#57249)📖 Documentation
🏗 Architecture / tests
test_util, torch_trainer) (#57939, #57873)💫 Enhancements:
🎉 New Features:
💫 Enhancements:
🔨 Fixes:
📖 Documentation:
🏗 Architecture refactoring:
🎉 New Features:
💫 Enhancements:
🔨 Fixes:
📖 Documentation:
🎉 New Features:
💫 Enhancements:
🔨 Fixes:
🎉 New Features:
💫 Enhancements:
🔨 Fixes:
📖 Documentation:
🏗 Architecture refactoring:
💫 Enhancements:
🔨 Fixes:
🎉 New Features:
💫 Enhancements:
🔨 Fixes:
Thank you to everyone who contributed to this release! Special thanks to all the contributors who helped make Ray 2.51.0 possible through bug fixes, features, documentation improvements, and testing efforts.
Ray Core: Fix deadlock when cancelling stale requests on in-order actors
Ray Core: Fix deadlock when cancelling stale requests on in-order actors (#57746)
One column per quarter.