Commit 452e0f8
authored
![renovate[bot]](https://avatars.githubusercontent.com/in/2740?v=4&size=40){kind=avatar}
Update dependency ray to v2.56.0 [SECURITY] (#8732)
> ℹ️ **Note**
>
> This PR body was truncated due to platform limits.
This PR contains the following updates:
| Package | Change |
[Age](https://docs.renovatebot.com/merge-confidence/) |
[Confidence](https://docs.renovatebot.com/merge-confidence/) |
|---|---|---|---|
| [ray](https://redirect.github.com/ray-project/ray) | `2.55.1` →
`2.56.0` |
|
|
---
> [!WARNING]
> Some dependencies could not be looked up. Check the [Dependency
Dashboard](../issues/357) for more information.
---
### [CVE-2026-57516](https://nvd.nist.gov/vuln/detail/CVE-2026-57516) /
[GHSA-hhrp-gw25-jr43](https://redirect.github.com/advisories/GHSA-hhrp-gw25-jr43)
/ PYSEC-2026-2273
<details>
<summary>More information</summary>
#### Details
Ray prior to 2.56.0 contains an unsafe deserialization vulnerability in
the WebDataset reader that allows attackers to achieve remote code
execution by supplying a malicious tar archive to the read_webdataset()
function. The _default_decoder() function in webdataset_datasource.py
unconditionally calls pickle.loads() on tar entries with .pkl/.pickle
extensions and torch.load() with weights_only=False on .pt/.pth entries,
executing arbitrary code inside Ray remote workers on every worker that
processes the malicious archive.
#### Severity
- CVSS Score: 8.6 / 10 (High)
- Vector String:
`CVSS:4.0/AV:N/AC:L/AT:N/PR:N/UI:A/VC:H/VI:H/VA:H/SC:N/SI:N/SA:N/E:X/CR:X/IR:X/AR:X/MAV:X/MAC:X/MAT:X/MPR:X/MUI:X/MVC:X/MVI:X/MVA:X/MSC:X/MSI:X/MSA:X/S:X/AU:X/R:X/V:X/RE:X/U:X`
#### References
-
[https://github.com/ray-project/ray/releases/tag/ray-2.56.0](https://redirect.github.com/ray-project/ray/releases/tag/ray-2.56.0)
-
[https://www.vulncheck.com/advisories/ray-unsafe-deserialization-rce-via-webdataset-reader](https://www.vulncheck.com/advisories/ray-unsafe-deserialization-rce-via-webdataset-reader)
-
[https://github.com/ray-project/ray/pull/63469](https://redirect.github.com/ray-project/ray/pull/63469)
-
[https://github.com/ray-project/ray/pull/63470](https://redirect.github.com/ray-project/ray/pull/63470)
-
[https://github.com/ray-project/ray/security/advisories/GHSA-hhrp-gw25-jr43](https://redirect.github.com/ray-project/ray/security/advisories/GHSA-hhrp-gw25-jr43)
This data is provided by
[OSV](https://osv.dev/vulnerability/PYSEC-2026-2273) and the [PyPI
Advisory Database](https://redirect.github.com/pypa/advisory-database)
([CC-BY
4.0](https://redirect.github.com/pypa/advisory-database/blob/main/LICENSE)).
</details>
---
### Release Notes
<details>
<summary>ray-project/ray (ray)</summary>
###
[`v2.56.0`](https://redirect.github.com/ray-project/ray/releases/tag/ray-2.56.0)
[Compare
Source](https://redirect.github.com/ray-project/ray/compare/ray-2.55.1...ray-2.56.0)
### Highlights
- **Ray Data Stability:** In this Ray release, we've added a variety of
stability improvements, including running multiple datasets in a
cluster, adding automatic batch size selection to CPU-based map-batches,
and default logical memory configuration to prevent OOMs. We've also
tightened `iter_batches` stability by reducing hidden buffering and
shutting down the executor when consumers exit early
([#​63660](https://redirect.github.com/ray-project/ray/issues/63660),
[#​63682](https://redirect.github.com/ray-project/ray/issues/63682),
[#​62949](https://redirect.github.com/ray-project/ray/issues/62949)).
This reduces object-store spilling for common training workloads
- **Ray Serve:** We re-architected Ray Serve LLM by decoupling request
handling from token streaming response path
([#​62667](https://redirect.github.com/ray-project/ray/issues/62667),
[#​62680](https://redirect.github.com/ray-project/ray/issues/62680),
[#​62668](https://redirect.github.com/ray-project/ray/issues/62668),
[#​62669](https://redirect.github.com/ray-project/ray/issues/62669),
[#​63167](https://redirect.github.com/ray-project/ray/issues/63167)),
resulting in significant LLM serving performance improvements. We've
also introduced new routing policies such as session-sticky routing via
consistent hashing with `ConsistentHashRouter`
([#​62905](https://redirect.github.com/ray-project/ray/issues/62905),
[#​63096](https://redirect.github.com/ray-project/ray/issues/63096),
[#​62906](https://redirect.github.com/ray-project/ray/issues/62906))
and `CapacityQueueRouter`
([#​62323](https://redirect.github.com/ray-project/ray/issues/62323))
which is beneficial for supply-constrained workloads.
- **Ray Core:** We've added GPU-domain-aware placement groups using
label locality
([#​61442](https://redirect.github.com/ray-project/ray/issues/61442),
[#​61614](https://redirect.github.com/ray-project/ray/issues/61614),
[#​62487](https://redirect.github.com/ray-project/ray/issues/62487),
[#​62533](https://redirect.github.com/ray-project/ray/issues/62533)).
This enables placement groups to pack bundles onto nodes that share a
`ray.io/gpu-domain` label instead of only packing at the single-node
level. We've also added initial Kubernetes in-place pod resizing support
for Autoscaler v2
([#​55961](https://redirect.github.com/ray-project/ray/issues/55961),
[#​62369](https://redirect.github.com/ray-project/ray/issues/62369),
[#​62215](https://redirect.github.com/ray-project/ray/issues/62215)),
enabling Ray clusters to resize CPU and memory on existing worker pods
before scaling out new pods.
### Ray Data
##### 🎉 New Features
- Support multiple datasets per cluster via subcluster labels and
resource partitioning
([#​63331](https://redirect.github.com/ray-project/ray/issues/63331),
[#​63375](https://redirect.github.com/ray-project/ray/issues/63375),
[#​63982](https://redirect.github.com/ray-project/ray/issues/63982))
- Add `Dataset.mix()` public API and `MixOperator` for weighted dataset
mixing
([#​63168](https://redirect.github.com/ray-project/ray/issues/63168),
[#​62450](https://redirect.github.com/ray-project/ray/issues/62450))
- New DataSourceV2 framework: `ParquetDatasourceV2`, chunked reader,
predicate splitting, listing/scanner infra
([#​63113](https://redirect.github.com/ray-project/ray/issues/63113),
[#​63454](https://redirect.github.com/ray-project/ray/issues/63454),
[#​63163](https://redirect.github.com/ray-project/ray/issues/63163),
[#​62975](https://redirect.github.com/ray-project/ray/issues/62975),
[#​63027](https://redirect.github.com/ray-project/ray/issues/63027),
[#​62182](https://redirect.github.com/ray-project/ray/issues/62182))
- Add `batch_size='auto'` to `map_batches` to derive batch row count
from target row batch size
([#​62648](https://redirect.github.com/ray-project/ray/issues/62648))
- Implement distributed upsert for Iceberg using task-based merge
algorithm, preventing performance bottleneck on driver
([#​63482](https://redirect.github.com/ray-project/ray/issues/63482))
- Add `include_row_hash` to `read_parquet`
([#​61408](https://redirect.github.com/ray-project/ray/issues/61408))
- Add JAX data iterator
([#​61630](https://redirect.github.com/ray-project/ray/issues/61630))
- Expose flag to run read tasks on isolated worker processes via
`isolate_read_workers`
([#​63490](https://redirect.github.com/ray-project/ray/issues/63490))
- Expose flag to set default logical memory for map operators via
`default_map_logical_memory_enabled`
([#​63814](https://redirect.github.com/ray-project/ray/issues/63814))
- Support predicate pushdown for Lance format
([#​61400](https://redirect.github.com/ray-project/ray/issues/61400))
- Support per-partition `start_offset` and `end_offset` for `read_kafka`
([#​61620](https://redirect.github.com/ray-project/ray/issues/61620))
- Add obstore async download backend for download operator
([#​61735](https://redirect.github.com/ray-project/ray/issues/61735))
- Support UDF retries on transient exceptions
([#​63023](https://redirect.github.com/ray-project/ray/issues/63023))
##### 💫 Enhancements
- Fix `iter_batches` spilling by replacing `make_async_gen` with
`iter_threaded` and reducing buffered batches
([#​63660](https://redirect.github.com/ray-project/ray/issues/63660),
[#​63682](https://redirect.github.com/ray-project/ray/issues/63682))
- Gate `restore_original_order` in `iter_batches` behind
`preserve_order`
([#​63792](https://redirect.github.com/ray-project/ray/issues/63792))
- Convert `drop_columns` to a `Project` logical operator when input
schema is known
([#​63813](https://redirect.github.com/ray-project/ray/issues/63813))
- Make `ConcatAggregation` and `TurbopufferDatasink` use `polars` for
sorting
([#​61904](https://redirect.github.com/ray-project/ray/issues/61904))
- Boost and vectorize `hash_partition` with `sort_indices`, zero-copy
slices, and pandas
([#​63498](https://redirect.github.com/ray-project/ray/issues/63498),
[#​62757](https://redirect.github.com/ray-project/ray/issues/62757),
[#​63152](https://redirect.github.com/ray-project/ray/issues/63152),
[#​62587](https://redirect.github.com/ray-project/ray/issues/62587))
- Enable `GPU_SHUFFLE` in `grouped_data.py`
([#​62410](https://redirect.github.com/ray-project/ray/issues/62410))
- Eager `StarExpr` expansion, schema inference for non-black-box UDFs,
and Expressions struct support
([#​63776](https://redirect.github.com/ray-project/ray/issues/63776),
[#​63387](https://redirect.github.com/ray-project/ray/issues/63387),
[#​62560](https://redirect.github.com/ray-project/ray/issues/62560))
- Make logging configurable via `RAY_DATA_LOG_LEVEL` and log `RAY_DATA`
env vars at execution start
([#​63487](https://redirect.github.com/ray-project/ray/issues/63487),
[#​63380](https://redirect.github.com/ray-project/ray/issues/63380))
- Display and track logical memory in progress bar
([#​63379](https://redirect.github.com/ray-project/ray/issues/63379))
- Honor `compute=` in `filter(expr=...)` and deprecate `concurrency=`
([#​63576](https://redirect.github.com/ray-project/ray/issues/63576))
- Enable filter pushdown through `StreamingRepartition` and read stage
column-rename removal
([#​62347](https://redirect.github.com/ray-project/ray/issues/62347),
[#​63384](https://redirect.github.com/ray-project/ray/issues/63384),
[#​63582](https://redirect.github.com/ray-project/ray/issues/63582))
- Cache deserialized Arrow schemas in `BlockMetadataWithSchema`
([#​63462](https://redirect.github.com/ray-project/ray/issues/63462))
- Track scheduling-loop step duration (p50/p90/max), peak
USS/object-store memory, and task block locality
([#​63586](https://redirect.github.com/ray-project/ray/issues/63586),
[#​63345](https://redirect.github.com/ray-project/ray/issues/63345),
[#​63489](https://redirect.github.com/ray-project/ray/issues/63489),
[#​63418](https://redirect.github.com/ray-project/ray/issues/63418),
[#​62249](https://redirect.github.com/ray-project/ray/issues/62249))
- Replace `TaskDurationStats` and Timer with `DistributionTracker`
([#​63488](https://redirect.github.com/ray-project/ray/issues/63488),
[#​63530](https://redirect.github.com/ray-project/ray/issues/63530),
[#​63825](https://redirect.github.com/ray-project/ray/issues/63825))
- Introduce `BlockEntry` on `RefBundle` in place of `(ref, metadata)`
tuples
([#​63654](https://redirect.github.com/ray-project/ray/issues/63654))
- Pre-resolve filesystem in threaded download to avoid IMDS herd
([#​62898](https://redirect.github.com/ray-project/ray/issues/62898))
- Convert logical operators to frozen dataclasses and consolidate
operator base/repr
([#​62593](https://redirect.github.com/ray-project/ray/issues/62593),
[#​62568](https://redirect.github.com/ray-project/ray/issues/62568),
[#​62400](https://redirect.github.com/ray-project/ray/issues/62400),
[#​63137](https://redirect.github.com/ray-project/ray/issues/63137),
[#​63140](https://redirect.github.com/ray-project/ray/issues/63140),
[#​63108](https://redirect.github.com/ray-project/ray/issues/63108))
- Non-blocking default autoscaling coordinator and resource-aware
auto-downscaling
([#​62725](https://redirect.github.com/ray-project/ray/issues/62725),
[#​62574](https://redirect.github.com/ray-project/ray/issues/62574))
- Release pinned blocks after dataset execution and shut down executor
on early `DataIterator` exit
([#​62456](https://redirect.github.com/ray-project/ray/issues/62456),
[#​62949](https://redirect.github.com/ray-project/ray/issues/62949))
- Optimize local shuffle with incremental index and configurable
compaction threshold
([#​62539](https://redirect.github.com/ray-project/ray/issues/62539))
- Speed up checkpoint filter and reduce memory usage
([#​60294](https://redirect.github.com/ray-project/ray/issues/60294))
- Preserve Arrow types through pandas roundtrip and reorder block
columns by name before schema ops
([#​63017](https://redirect.github.com/ray-project/ray/issues/63017),
[#​63582](https://redirect.github.com/ray-project/ray/issues/63582))
- Block pickle object columns when reading untrusted Parquet and gate
unsafe WebDataset deserialization
([#​63470](https://redirect.github.com/ray-project/ray/issues/63470),
[#​63469](https://redirect.github.com/ray-project/ray/issues/63469))
- Move backpressure escape hatch across all policies
([#​63539](https://redirect.github.com/ray-project/ray/issues/63539))
- Update `pandas`, `modin`, and `pyarrow` minimum versions
([#​62899](https://redirect.github.com/ray-project/ray/issues/62899))
- Add utilization monitoring and correct logical resource usage for
`ActorPool`
([#​61987](https://redirect.github.com/ray-project/ray/issues/61987),
[#​61528](https://redirect.github.com/ray-project/ray/issues/61528))
- Deprecate `ConcurrencyCapBackpressurePolicy`, `DataIterator.to_torch`,
and pandas UDF batches
([#​63392](https://redirect.github.com/ray-project/ray/issues/63392),
[#​62540](https://redirect.github.com/ray-project/ray/issues/62540),
[#​61733](https://redirect.github.com/ray-project/ray/issues/61733))
- Rank actors per node in a heap and avoid re-exporting actor class via
`.options`
([#​62309](https://redirect.github.com/ray-project/ray/issues/62309),
[#​62722](https://redirect.github.com/ray-project/ray/issues/62722))
- `read_delta` reads from preconfigured `pyarrow` dataset
([#​61721](https://redirect.github.com/ray-project/ray/issues/61721))
- Include column name and target type in `ArrowConversionError`; reduce
arrow conversion warning verbosity
([#​62407](https://redirect.github.com/ray-project/ray/issues/62407),
[#​61486](https://redirect.github.com/ray-project/ray/issues/61486),
[#​62521](https://redirect.github.com/ray-project/ray/issues/62521))
- Show external consumer bytes in verbose operator progress log
([#​63728](https://redirect.github.com/ray-project/ray/issues/63728))
- Disable `DataSourceV2` by default after earlier enabling
([#​63674](https://redirect.github.com/ray-project/ray/issues/63674),
[#​63326](https://redirect.github.com/ray-project/ray/issues/63326))
##### 🔨 Fixes
- Rename subcluster label key from `__subcluster__` to `ray-subcluster`
([#​63982](https://redirect.github.com/ray-project/ray/issues/63982))
- Fix `get_or_create_stats_actor` crash in Ray Client mode
([#​63402](https://redirect.github.com/ray-project/ray/issues/63402))
- Fix datasource pushdown crashes for generic `UDFExpr` filter
predicates
([#​63781](https://redirect.github.com/ray-project/ray/issues/63781))
- Fix hash-shuffle aggregator memory estimation: metadata propagation,
node-size clamp, column pruning
([#​63809](https://redirect.github.com/ray-project/ray/issues/63809))
- Fix `CheckpointConfig` `FileNotFoundError` on Azure Blob Storage
([#​63606](https://redirect.github.com/ray-project/ray/issues/63606))
- Fix silent credential drop for fsspec-S3 in download expression
([#​62897](https://redirect.github.com/ray-project/ray/issues/62897))
- Fix missing f-string prefix in `_concatenate_extension_column`
([#​62939](https://redirect.github.com/ray-project/ray/issues/62939))
- Fix `HashAggregate` duplicate group rows for `AggregateFnV2`
([#​63066](https://redirect.github.com/ray-project/ray/issues/63066))
- Fix JSONL read retry with advanced file cursor
([#​63233](https://redirect.github.com/ray-project/ray/issues/63233))
- Fix `read_parquet` `ArrowNotImplementedError` for nested column types
exceeding \~2GB row group
([#​61824](https://redirect.github.com/ray-project/ray/issues/61824))
- Fix `read_parquet` nested-type fallback and parquet scanner memory
accumulation
([#​63175](https://redirect.github.com/ray-project/ray/issues/63175),
[#​62745](https://redirect.github.com/ray-project/ray/issues/62745))
- Fix memory leak in `DataIterator.to_torch()` by switching to `PyArrow`
([#​60966](https://redirect.github.com/ray-project/ray/issues/60966))
- Fix `ZipOperator` freeing shared blocks via `_split_at_indices`
([#​62665](https://redirect.github.com/ray-project/ray/issues/62665))
- Fix concurrent writes race condition in `write_parquet`
([#​62377](https://redirect.github.com/ray-project/ray/issues/62377))
- Fix GPU shuffle output ordering when using
`ShuffleStrategy.GPU_SHUFFLE`
([#​62351](https://redirect.github.com/ray-project/ray/issues/62351))
- Fix incorrect `DatasetStat` uuid propagation
([#​62255](https://redirect.github.com/ray-project/ray/issues/62255))
- Fix none issue when `DATA_ENABLE_OP_RESOURCE_RESERVATION=False`
([#​61718](https://redirect.github.com/ray-project/ray/issues/61718))
- Fix filesystem compatibility check for fsspec-wrapped `PyFileSystem`
([#​61850](https://redirect.github.com/ray-project/ray/issues/61850))
- Forward `try_create_dir` to `pyarrow.dataset.write_dataset`
([#​58302](https://redirect.github.com/ray-project/ray/issues/58302))
- Fix autoscaler bug blocking timely release of leased resources
([#​62592](https://redirect.github.com/ray-project/ray/issues/62592))
- Ensure consistent `nan_is_null`/nans-as-nulls semantics in encoder
([#​62623](https://redirect.github.com/ray-project/ray/issues/62623),
[#​62618](https://redirect.github.com/ray-project/ray/issues/62618))
- Skip unconditional null strip in `find_partition_index`
([#​62594](https://redirect.github.com/ray-project/ray/issues/62594))
- V1 `_split_predicate_by_columns` correctness fix
([#​63176](https://redirect.github.com/ray-project/ray/issues/63176))
- Avoid importing cudf in `_is_cudf_dataframe` when cudf not loaded
([#​62302](https://redirect.github.com/ray-project/ray/issues/62302))
- Revert raw-modulo hash partition fast path
([#​63097](https://redirect.github.com/ray-project/ray/issues/63097))
- Remove `tfx-bsl` support from `read_tfrecords`
([#​63245](https://redirect.github.com/ray-project/ray/issues/63245))
##### 📖 Documentation
- Document `isolate_read_workers` for `read_parquet`
([#​63816](https://redirect.github.com/ray-project/ray/issues/63816))
- Remove docs recommending increased object store memory proportion
([#​63389](https://redirect.github.com/ray-project/ray/issues/63389))
- Update docs minimum version for `build_processor` and `"auto"` batch
size
([#​61757](https://redirect.github.com/ray-project/ray/issues/61757),
[#​62790](https://redirect.github.com/ray-project/ray/issues/62790))
- Remove outdated limitation of `DefaultClusterAutoscalerV2` and stale
object-store-memory warnings
([#​62385](https://redirect.github.com/ray-project/ray/issues/62385),
[#​62387](https://redirect.github.com/ray-project/ray/issues/62387))
### Ray Serve
##### 🎉 New Features:
- Add custom ingress request router app interfaces and HAProxy ingress
dispatch path
([#​62680](https://redirect.github.com/ray-project/ray/issues/62680),
[#​62668](https://redirect.github.com/ray-project/ray/issues/62668),
[#​62669](https://redirect.github.com/ray-project/ray/issues/62669),
[#​62667](https://redirect.github.com/ray-project/ray/issues/62667))
- Expose `choose_replica`/`dispatch` on deployment handles and
`AsyncioRouter` with replica-side slot reservation
([#​63255](https://redirect.github.com/ray-project/ray/issues/63255),
[#​63254](https://redirect.github.com/ray-project/ray/issues/63254),
[#​63252](https://redirect.github.com/ray-project/ray/issues/63252))
- Introduce experimental round robin router and `ConsistentHashRouter`
for session-sticky routing
([#​63238](https://redirect.github.com/ray-project/ray/issues/63238),
[#​62906](https://redirect.github.com/ray-project/ray/issues/62906),
[#​63096](https://redirect.github.com/ray-project/ray/issues/63096),
[#​62905](https://redirect.github.com/ray-project/ray/issues/62905))
- Central capacity queue for token-based request routing via
`CapacityQueueRouter`
([#​62323](https://redirect.github.com/ray-project/ray/issues/62323))
- Add experimental `ray-haproxy` support behind
`RAY_SERVE_EXPERIMENTAL_PIP_HAPROXY`
([#​62589](https://redirect.github.com/ray-project/ray/issues/62589))
- Add deployment actor context API and broadcast API for deployment
handles
([#​62532](https://redirect.github.com/ray-project/ray/issues/62532),
[#​61472](https://redirect.github.com/ray-project/ray/issues/61472))
- Add `ControllerOptions` for configurable controller `runtime_env`
([#​63352](https://redirect.github.com/ray-project/ray/issues/63352))
- Make rolling update percentage configurable
([#​62160](https://redirect.github.com/ray-project/ray/issues/62160))
- Support per-request timeout and disconnect in HTTP proxy path
([#​62867](https://redirect.github.com/ray-project/ray/issues/62867))
##### 💫 Enhancements:
- HAProxy stability improvements: wait for old workers before drain,
redirect stdout/stderr, redispatch+retry-on, coalesce broadcasts,
quarantine released ports
([#​63620](https://redirect.github.com/ray-project/ray/issues/63620),
[#​63621](https://redirect.github.com/ray-project/ray/issues/63621),
[#​63622](https://redirect.github.com/ray-project/ray/issues/63622),
[#​63623](https://redirect.github.com/ray-project/ray/issues/63623),
[#​63628](https://redirect.github.com/ray-project/ray/issues/63628))
- Bind direct ingress ports to `0.0.0.0` for cross-node HAProxy routing
([#​62515](https://redirect.github.com/ray-project/ray/issues/62515))
- HAProxy ingress request router metrics, enable splice by default,
`TCP_NODELAY` default 1, optional retry knobs,
`RAY_SERVE_HAPROXY_STATS_PORT`
([#​63356](https://redirect.github.com/ray-project/ray/issues/63356),
[#​63531](https://redirect.github.com/ray-project/ray/issues/63531),
[#​63353](https://redirect.github.com/ray-project/ray/issues/63353),
[#​63415](https://redirect.github.com/ray-project/ray/issues/63415),
[#​62979](https://redirect.github.com/ray-project/ray/issues/62979))
- Resolve bundled ray-haproxy binary before
`RAY_SERVE_HAPROXY_BINARY_PATH`; HAProxy abspath env var
([#​63829](https://redirect.github.com/ray-project/ray/issues/63829),
[#​62610](https://redirect.github.com/ray-project/ray/issues/62610))
- Replace socat subprocess with Python socket for HAProxy admin
communication; bump HAProxy to avoid CVE-2025-11230
([#​61897](https://redirect.github.com/ray-project/ray/issues/61897),
[#​62585](https://redirect.github.com/ray-project/ray/issues/62585))
- Expose controller health metrics via `/api/serve/applications/` API;
add `max_replicas_per_node` to response
([#​63556](https://redirect.github.com/ray-project/ray/issues/63556),
[#​63234](https://redirect.github.com/ray-project/ray/issues/63234))
- Run health check on user execution path to detect request-serving
stalls
([#​61621](https://redirect.github.com/ray-project/ray/issues/61621))
- Mark widely-used APIs as stable
([#​62932](https://redirect.github.com/ray-project/ray/issues/62932))
- Retain recently-stopped replica logs in the dashboard
([#​63678](https://redirect.github.com/ray-project/ray/issues/63678))
- Add observability logs for pack scheduling decisions
([#​63603](https://redirect.github.com/ray-project/ray/issues/63603))
- Gate ingress request router body forwarding behind escape hatch
([#​63183](https://redirect.github.com/ray-project/ray/issues/63183))
- Avoid rolling replicas for no-op config overrides
([#​63034](https://redirect.github.com/ray-project/ray/issues/63034))
- Gate replica/deployment creation during shutdown
([#​62761](https://redirect.github.com/ray-project/ray/issues/62761))
- Defer PG creation for TPU Serve deployments to accelerator backend
([#​62941](https://redirect.github.com/ray-project/ray/issues/62941))
- Expose `DeploymentStateManager` APIs for controller access
([#​62950](https://redirect.github.com/ray-project/ray/issues/62950))
- Add tracing support for Windows and gRPC tracing improvements
([#​62821](https://redirect.github.com/ray-project/ray/issues/62821),
[#​63833](https://redirect.github.com/ray-project/ray/issues/63833))
- Split node vs requested resources in deployment scheduler
([#​62778](https://redirect.github.com/ray-project/ray/issues/62778))
- Defer `DEPLOYMENT_TARGETS` broadcast while replicas are RECOVERING
([#​62751](https://redirect.github.com/ray-project/ray/issues/62751))
- Evict per-deployment `LongPollHost` state on deployment delete; enable
logs when client stops its event loop
([#​62820](https://redirect.github.com/ray-project/ray/issues/62820),
[#​63028](https://redirect.github.com/ray-project/ray/issues/63028))
- Add metrics: max replica processing latency, objref resolution
latency, `serve_autoscaling_target_ongoing_requests`
([#​62381](https://redirect.github.com/ray-project/ray/issues/62381),
[#​62355](https://redirect.github.com/ray-project/ray/issues/62355),
[#​62421](https://redirect.github.com/ray-project/ray/issues/62421))
- Filter stale bootstrap observations from `serve_long_poll_latency_ms`
([#​62868](https://redirect.github.com/ray-project/ray/issues/62868))
- Retry `build_serve_application` task on failure
([#​62987](https://redirect.github.com/ray-project/ray/issues/62987))
- Scale down non-matching primary-label replicas first
([#​61488](https://redirect.github.com/ray-project/ray/issues/61488))
- Refactor internal autoscaling policy state extraction into a single
helper
([#​62452](https://redirect.github.com/ray-project/ray/issues/62452))
- Catalog Ray Serve env vars
([#​62006](https://redirect.github.com/ray-project/ray/issues/62006))
- Remove or raise clear error for deprecated deployment items; remove
deprecated `DeploymentMode`
([#​63548](https://redirect.github.com/ray-project/ray/issues/63548),
[#​63510](https://redirect.github.com/ray-project/ray/issues/63510))
##### 🔨 Fixes:
- Fix orphaned actors on controller crash during shutdown; drop and
replace replicas surviving a controller crash without rank assignment
([#​62823](https://redirect.github.com/ray-project/ray/issues/62823),
[#​63139](https://redirect.github.com/ray-project/ray/issues/63139))
- Fix deployment actors creating 15K OS threads for sync actor classes
([#​62661](https://redirect.github.com/ray-project/ray/issues/62661))
- Fix gang scheduling PG leak when deployment actors are starting
([#​62469](https://redirect.github.com/ray-project/ray/issues/62469))
- Fix app-level autoscaling policy state cross-deployment contamination
and state loss for skipped deployments
([#​62484](https://redirect.github.com/ray-project/ray/issues/62484))
- Fix Serve autoscaling delay to use wall-clock time
([#​62144](https://redirect.github.com/ray-project/ray/issues/62144))
- Fix race condition in multiplex LRU cache update using `move_to_end()`
([#​62548](https://redirect.github.com/ray-project/ray/issues/62548))
- Normalize multiplexed model ID header to support proxy-transformed
names
([#​61869](https://redirect.github.com/ray-project/ray/issues/61869))
- Fix `AttributeError` when `request_router` is None in
`update_deployment_config`
([#​63180](https://redirect.github.com/ray-project/ray/issues/63180))
- Fix potential `UnboundLocalError` in
`ActorReplicaWrapper.check_stopped()`
([#​63339](https://redirect.github.com/ray-project/ray/issues/63339))
- Fail loud when ingress request router dispatch fails
([#​63215](https://redirect.github.com/ray-project/ray/issues/63215))
- Fix stale `_global_client` cache across driver sessions
([#​62368](https://redirect.github.com/ray-project/ray/issues/62368))
- Fix `start_metrics_pusher` crash when deployment has
`record_autoscaling_stats` but no autoscaling config
([#​62123](https://redirect.github.com/ray-project/ray/issues/62123))
- Fix high-cardinality namespace tag on long poll metrics
([#​62386](https://redirect.github.com/ray-project/ray/issues/62386))
- Fix Java long poll timeout serialization
([#​61875](https://redirect.github.com/ray-project/ray/issues/61875))
- Avoid destructor error when FastAPI ingress init fails
([#​62172](https://redirect.github.com/ray-project/ray/issues/62172))
- Avoid proxy readiness future timeout race
([#​62194](https://redirect.github.com/ray-project/ray/issues/62194))
- Avoid self-cause on non-gRPC replica exceptions
([#​62412](https://redirect.github.com/ray-project/ray/issues/62412))
- Fix HAProxy startup timeout propagation
([#​61752](https://redirect.github.com/ray-project/ray/issues/61752))
- Include `ingress_request_router.lua.tmpl` in `package_data`
([#​63145](https://redirect.github.com/ray-project/ray/issues/63145))
- Revert support for `root_path` parameter across uvicorn versions
([#​62529](https://redirect.github.com/ray-project/ray/issues/62529))
##### 📖 Documentation:
- Add round robin and consistent hashing router documentation
([#​63636](https://redirect.github.com/ray-project/ray/issues/63636))
- Introduce gang scheduling documentation
([#​61737](https://redirect.github.com/ray-project/ray/issues/61737))
- Add deployment scope actor docs
([#​62735](https://redirect.github.com/ray-project/ray/issues/62735))
- Add Kuberay guide for RayService with HAProxy and High Throughput mode
([#​62408](https://redirect.github.com/ray-project/ray/issues/62408))
- Add Ray Serve office hours invite into documentation
([#​62176](https://redirect.github.com/ray-project/ray/issues/62176))
### Ray Train
##### 🎉 New Features
- Add `LoggingConfig` for configuring the `ray.train` logger on
controller and workers
([#​61550](https://redirect.github.com/ray-project/ray/issues/61550))
- Allow `DataParallelTrainer`'s `train_fn` to return data
([#​62021](https://redirect.github.com/ray-project/ray/issues/62021))
- Add async checkpointing/validation with Torch Lightning
([#​62370](https://redirect.github.com/ray-project/ray/issues/62370))
##### 💫 Enhancements
- Report time spent syncing and transferring checkpoints to storage in
`ray.train.report(checkpoint)`
([#​62027](https://redirect.github.com/ray-project/ray/issues/62027))
- Block until `create_or_update_train_run` completes on Train
initialization
([#​63432](https://redirect.github.com/ray-project/ray/issues/63432))
- Implement `DatasetManager`
([#​63309](https://redirect.github.com/ray-project/ray/issues/63309))
- Forward `label_selector` to `AutoscalingCoordinator`
([#​63287](https://redirect.github.com/ray-project/ray/issues/63287))
- Add log line before launching training function
([#​62911](https://redirect.github.com/ray-project/ray/issues/62911))
- Allow `contextlib.redirect_stdout()` to bypass print redirect to logs
([#​61075](https://redirect.github.com/ray-project/ray/issues/61075))
- Add timeouts to validation functions of `ray.train.report`
([#​62916](https://redirect.github.com/ray-project/ray/issues/62916))
- `ray.train.report` does not hang across replica group restarts; Ray
Train manages replica group restarts
([#​62651](https://redirect.github.com/ray-project/ray/issues/62651),
[#​61475](https://redirect.github.com/ray-project/ray/issues/61475))
- Swallow `RayTaskError` during `BackendSetupCallback` shutdown
([#​63143](https://redirect.github.com/ray-project/ray/issues/63143))
- Improve `JaxTrainer` TPU multi-slice fault tolerance and reservation
ergonomics
([#​62893](https://redirect.github.com/ray-project/ray/issues/62893))
- Export default data execution options
([#​62784](https://redirect.github.com/ray-project/ray/issues/62784))
- Consolidate Train run metadata sanitization and improve readability
([#​63182](https://redirect.github.com/ray-project/ray/issues/63182))
- Fix `PlacementGroupCleaner` race condition: drain queue before cleanup
on controller death
([#​62754](https://redirect.github.com/ray-project/ray/issues/62754))
- Harden against unsafe pickle deserialization
([#​62807](https://redirect.github.com/ray-project/ray/issues/62807))
- Raise error when checkpoint is within experiment directory and
`delete_local_checkpoint_after_upload=True`
([#​62555](https://redirect.github.com/ray-project/ray/issues/62555))
- Add `timeout_s` to `ray.train.get_all_reported_checkpoints`
([#​61761](https://redirect.github.com/ray-project/ray/issues/61761))
- Change remaining `pytorch_lightning` imports
([#​61291](https://redirect.github.com/ray-project/ray/issues/61291))
- Make controller resilient to errors in all lifecycle hooks
([#​60900](https://redirect.github.com/ray-project/ray/issues/60900))
- Remove `Predictor` from Train v1
([#​63461](https://redirect.github.com/ray-project/ray/issues/63461))
##### 🔨 Fixes
- Fix missing comma in `DataBatchType` Union type
([#​63872](https://redirect.github.com/ray-project/ray/issues/63872))
- Handle Arrow-backed pandas dtypes in LightGBM examples
([#​63427](https://redirect.github.com/ray-project/ray/issues/63427))
- Fix `exclude_resources` regression for V1 Train + V2 cluster
autoscaler
([#​62827](https://redirect.github.com/ray-project/ray/issues/62827))
- Add missing `%s` to `logger.debug`
([#​63039](https://redirect.github.com/ray-project/ray/issues/63039))
- Increase `get_actor` timeout
([#​62516](https://redirect.github.com/ray-project/ray/issues/62516))
##### 📖 Documentation
- Document S3-compatible storage
([#​63103](https://redirect.github.com/ray-project/ray/issues/63103))
- Add Azure Files to persistent storage docs
([#​63406](https://redirect.github.com/ray-project/ray/issues/63406))
- Uncomment `Result.from_path` in docs
([#​62887](https://redirect.github.com/ray-project/ray/issues/62887))
- Document how to tune async validation
([#​62227](https://redirect.github.com/ray-project/ray/issues/62227))
- Document why validation runs need unique names
([#​62224](https://redirect.github.com/ray-project/ray/issues/62224))
### Ray Tune
##### 💫 Enhancements
- Fix Tune search for Python 3.14
([#​63575](https://redirect.github.com/ray-project/ray/issues/63575))
- Modernize `AxSearch` for Ax Platform 1.0.0+
([#​60522](https://redirect.github.com/ray-project/ray/issues/60522))
- Use built-in `inspect` for argument capture
([#​60049](https://redirect.github.com/ray-project/ray/issues/60049))
##### 🔨 Fixes
- Fix import count in CIFAR PyTorch tutorial
([#​62756](https://redirect.github.com/ray-project/ray/issues/62756))
### Ray LLM
##### 🎉 New Features
- Major Ray Serve LLM performance improvement with direct streaming
([#​63167](https://redirect.github.com/ray-project/ray/issues/63167),
[#​63468](https://redirect.github.com/ray-project/ray/issues/63468),
[#​63779](https://redirect.github.com/ray-project/ray/issues/63779))
- TPU support: Add `topology` field to `LLMConfig` for multi-host TPU
support
([#​61906](https://redirect.github.com/ray-project/ray/issues/61906))
- Add per-host bundles default and fix fractional TPUs for
`TPUAccelerator`
([#​63177](https://redirect.github.com/ray-project/ray/issues/63177))
- Enable Ray Serve LLM session-stickiness routing policy via
`RAY_SERVE_SESSION_ID_HEADER_KEY`
([#​63362](https://redirect.github.com/ray-project/ray/issues/63362))
##### 💫 Enhancements
- Upgrade `vLLM` to 0.22.0
([#​63730](https://redirect.github.com/ray-project/ray/issues/63730),
[#​63396](https://redirect.github.com/ray-project/ray/issues/63396),
[#​62970](https://redirect.github.com/ray-project/ray/issues/62970),
[#​62349](https://redirect.github.com/ray-project/ray/issues/62349))
- Co-locate DP rank 0 worker with advertised master address
([#​63803](https://redirect.github.com/ray-project/ray/issues/63803))
- Add pick-only fast path to `AsyncioRouter` for LLM ingress
([#​63517](https://redirect.github.com/ray-project/ray/issues/63517))
- Replace LLM ingress router replica selection with `choose_replica`;
don't fetch `LLMConfig` from replicas at startup
([#​63280](https://redirect.github.com/ray-project/ray/issues/63280),
[#​63065](https://redirect.github.com/ray-project/ray/issues/63065))
- Promote `max_tasks_in_flight_per_actor` to a first-class config field
and adjust defaults
([#​63214](https://redirect.github.com/ray-project/ray/issues/63214))
- Validate `accelerator_type` against CPU-only configs; replace
`GPUType` alias with `AcceleratorType`
([#​62139](https://redirect.github.com/ray-project/ray/issues/62139),
[#​62978](https://redirect.github.com/ray-project/ray/issues/62978))
- Add rate-limiter for per-request traceback spam
([#​62440](https://redirect.github.com/ray-project/ray/issues/62440))
- Promote SGLang integration to user guide and move engine to
`_internal`
([#​62570](https://redirect.github.com/ray-project/ray/issues/62570))
- Lazy-load batch stage/processor submodules and make boto3/botocore
imports lazy
([#​62861](https://redirect.github.com/ray-project/ray/issues/62861),
[#​62383](https://redirect.github.com/ray-project/ray/issues/62383))
- LLM telemetry bugfixes
([#​63782](https://redirect.github.com/ray-project/ray/issues/63782))
##### 🔨 Fixes
- Fix flaky GPU-0 worker and NIXL port collisions
([#​63810](https://redirect.github.com/ray-project/ray/issues/63810))
- Fix P/D direct streaming OpenAI routing
([#​63679](https://redirect.github.com/ray-project/ray/issues/63679))
- Remove `guided_decoding`, `truncate_prompt_tokens`,
`build_llm_processor`
([#​63569](https://redirect.github.com/ray-project/ray/issues/63569))
- Fix misleading `ImportError` when `vLLM` is installed but fails to
import
([#​63305](https://redirect.github.com/ray-project/ray/issues/63305))
- Fix `max_pending_requests` default to track `vLLM`'s GPU-dependent
`max_num_seqs`
([#​62918](https://redirect.github.com/ray-project/ray/issues/62918))
- Fix HF config loading for models with custom `rope_scaling`
([#​62464](https://redirect.github.com/ray-project/ray/issues/62464))
- Wait for request router init in `LLMRouter` constructor
([#​63206](https://redirect.github.com/ray-project/ray/issues/63206))
- Materialize chat completion message `content` in sanitizer
([#​63119](https://redirect.github.com/ray-project/ray/issues/63119))
- Fix `lora_request` not forwarded to `vLLM` engine + add regression
tests
([#​62609](https://redirect.github.com/ray-project/ray/issues/62609))
- Fix `SGLangEngineProcessor` telemetry for `trust_remote_code` models
([#​62102](https://redirect.github.com/ray-project/ray/issues/62102))
- Fix `TOKENIZER_ONLY` downloads missing `chat_template` for S3-backed
models
([#​62121](https://redirect.github.com/ray-project/ray/issues/62121))
- Fix SGLang chat tokenize to respect `add_generation_prompt`
([#​61688](https://redirect.github.com/ray-project/ray/issues/61688))
- Fix bool serialization in `benchmark_vllm` CLI builder
([#​63516](https://redirect.github.com/ray-project/ray/issues/63516))
##### 📖 Documentation
- Document multimodal pixel-budget gotchas and `vLLM` compatibility
([#​63593](https://redirect.github.com/ray-project/ray/issues/63593))
- Add tokenization disaggregation documentation
([#​62494](https://redirect.github.com/ray-project/ray/issues/62494))
- Add benchmark docs and refactor into submodules
([#​62204](https://redirect.github.com/ray-project/ray/issues/62204))
- Remove `VLLM_USE_V1` from docs and examples
([#​63001](https://redirect.github.com/ray-project/ray/issues/63001))
- Fix wrong documented default for `max_tasks_in_flight_per_actor`
([#​62917](https://redirect.github.com/ray-project/ray/issues/62917))
### Ray RLlib
##### 🎉 New Features
- Add `custom_resources_per_learner` config and
`custom_resources_for_main_process` to `AlgorithmConfig`
([#​63303](https://redirect.github.com/ray-project/ray/issues/63303),
[#​62475](https://redirect.github.com/ray-project/ray/issues/62475))
- Add Importance Sampling `APPO` metrics to the torch learner
([#​63675](https://redirect.github.com/ray-project/ray/issues/63675))
##### 💫 Enhancements
- Put only one copy of weights into the object store
([#​63529](https://redirect.github.com/ray-project/ray/issues/63529))
- Handle the all-evaluation-workers-unhealthy case uniformly across
modes
([#​63128](https://redirect.github.com/ray-project/ray/issues/63128))
- Stop `IMPALA`/`APPO` learner thread gracefully to avoid misleading
error messages
([#​62763](https://redirect.github.com/ray-project/ray/issues/62763))
- Improve invalid input error messages
([#​62324](https://redirect.github.com/ray-project/ray/issues/62324))
##### 🔨 Fixes
- Fix two substantial edge cases in `PPO`'s value target calculation
([#​59958](https://redirect.github.com/ray-project/ray/issues/59958))
- Fix `EnvRunner` crash loops
([#​62884](https://redirect.github.com/ray-project/ray/issues/62884))
- Fix extra model outputs hanging val indexing
([#​62960](https://redirect.github.com/ray-project/ray/issues/62960))
- Fix `ValueError` in `MultiAgentEpisode.get_rewards()` when an agent is
inactive for all requested env steps
([#​62907](https://redirect.github.com/ray-project/ray/issues/62907))
- Preserve Torch optimizer param-group scalar types on restore
([#​61937](https://redirect.github.com/ray-project/ray/issues/61937))
- Fix wrong assert variable in `_update_env_seed_if_necessary`
([#​61823](https://redirect.github.com/ray-project/ray/issues/61823))
- Maintain value in `EMAStat`
([#​63064](https://redirect.github.com/ray-project/ray/issues/63064))
##### 📖 Documentation
- Clarify extra model output docstrings
([#​63524](https://redirect.github.com/ray-project/ray/issues/63524))
### Ray Core
##### 🎉 New Features
- Add support for Furiosa AI NPU
([#​63035](https://redirect.github.com/ray-project/ray/issues/63035))
and `register_collective_backend` API for custom collective backends
([#​60701](https://redirect.github.com/ray-project/ray/issues/60701))
- In-place pod resizing (IPPR) on Kubernetes 1.35: initial
implementation and standalone KubeRay IPPR provider
([#​55961](https://redirect.github.com/ray-project/ray/issues/55961),
[#​62369](https://redirect.github.com/ray-project/ray/issues/62369),
[#​62215](https://redirect.github.com/ray-project/ray/issues/62215))
- Label locality support: GPU-domain-aware placement groups, autoscaler
proto changes, and state API observability
([#​61442](https://redirect.github.com/ray-project/ray/issues/61442),
[#​61614](https://redirect.github.com/ray-project/ray/issues/61614),
[#​62487](https://redirect.github.com/ray-project/ray/issues/62487),
[#​62533](https://redirect.github.com/ray-project/ray/issues/62533))
- Publish platform events via Ray Event Recorder and support
single-event emission in the Python layer
([#​63329](https://redirect.github.com/ray-project/ray/issues/63329),
[#​60858](https://redirect.github.com/ray-project/ray/issues/60858))
- Autoscaler v2: priority-based worker group selection
([#​62997](https://redirect.github.com/ray-project/ray/issues/62997))
and `noDriverTimeoutSeconds` for KubeRay cluster termination
([#​63465](https://redirect.github.com/ray-project/ray/issues/63465))
- RDT: concurrent one-sided transfers for multiple `ObjectRef`s in
`ray.get`
([#​61773](https://redirect.github.com/ray-project/ray/issues/61773)),
retry support
([#​62842](https://redirect.github.com/ray-project/ray/issues/62842)),
and NIXL memory deregistration via `deregister_nixl_memory`
([#​62341](https://redirect.github.com/ray-project/ray/issues/62341))
- Support `.tar.gz` archives for remote `working_dir` URIs
([#​62813](https://redirect.github.com/ray-project/ray/issues/62813))
- Add IPv6 localhost and all-interfaces support
([#​60023](https://redirect.github.com/ray-project/ray/issues/60023))
##### 💫 Enhancements
- Resource isolation: event-based memory monitor, multi-memory-monitor
factory, time-based group killing policy, idle-worker prioritization,
system/user slice bounds, and OOM policy tuning
([#​62060](https://redirect.github.com/ray-project/ray/issues/62060),
[#​62705](https://redirect.github.com/ray-project/ray/issues/62705),
[#​62643](https://redirect.github.com/ray-project/ray/issues/62643),
[#​62378](https://redirect.github.com/ray-project/ray/issues/62378),
[#​62168](https://redirect.github.com/ray-project/ray/issues/62168),
[#​63521](https://redirect.github.com/ray-project/ray/issues/63521),
[#​63324](https://redirect.github.com/ray-project/ray/issues/63324),
[#​63067](https://redirect.github.com/ray-project/ray/issues/63067),
[#​62957](https://redirect.github.com/ray-project/ray/issues/62957))
- Compute per-component memory usage in MiB
([#​63932](https://redirect.github.com/ray-project/ray/issues/63932))
and add host vs container memory distinction to memory panels
([#​63111](https://redirect.github.com/ray-project/ray/issues/63111))
- Consider cgroup limit when fetching CPU
([#​63685](https://redirect.github.com/ray-project/ray/issues/63685))
and correct worker OOM score adjustment logic
([#​62470](https://redirect.github.com/ray-project/ray/issues/62470))
- Replace `NodeAffinitySchedulingStrategy` with Label Selector API when
`soft=False`
([#​54940](https://redirect.github.com/ray-project/ray/issues/54940))
- Improve `SlicePlacementGroup` lifecycle and support explicit
`bundle_label_selector` for TPUs
([#​63171](https://redirect.github.com/ray-project/ray/issues/63171));
add TPU head resource for Ironwood TPU
([#​62786](https://redirect.github.com/ray-project/ray/issues/62786)),
`chips_per_vm` arg
([#​62526](https://redirect.github.com/ray-project/ray/issues/62526)),
and v6e single-host fixes
([#​62306](https://redirect.github.com/ray-project/ray/issues/62306))
- Batch placement group bundle removal RPCs
([#​63839](https://redirect.github.com/ray-project/ray/issues/63839));
remove PG resource deduction from GCS in favor of resource broadcast
([#​63723](https://redirect.github.com/ray-project/ray/issues/63723))
- Migrate Raylet/GCS timing logic to a shared `ClockInterface` with a
fake clock for testing
([#​62562](https://redirect.github.com/ray-project/ray/issues/62562),
[#​62502](https://redirect.github.com/ray-project/ray/issues/62502),
[#​62476](https://redirect.github.com/ray-project/ray/issues/62476))
- Refactor asio build targets and add `IOContextMonitor`; run GCS health
check on `io_service`
([#​63042](https://redirect.github.com/ray-project/ray/issues/63042),
[#​63166](https://redirect.github.com/ray-project/ray/issues/63166),
[#​62608](https://redirect.github.com/ray-project/ray/issues/62608),
[#​62374](https://redirect.github.com/ray-project/ray/issues/62374))
- Autoscaler v2 performance: skip serializations for debug logs
([#​63778](https://redirect.github.com/ray-project/ray/issues/63778));
accept fractional resource values in `request_resources`
([#​63306](https://redirect.github.com/ray-project/ray/issues/63306))
- Reduce traffic: halve task arg pubsub by skipping redundant raylet
pull
([#​62583](https://redirect.github.com/ray-project/ray/issues/62583)),
avoid extra memcpy when spilling fused objects
([#​63653](https://redirect.github.com/ray-project/ray/issues/63653)),
and resolve task dependencies synchronously when objects exist
([#​62561](https://redirect.github.com/ray-project/ray/issues/62561))
- Improve `inspect_serializability` messages and traversal context
([#​63501](https://redirect.github.com/ray-project/ray/issues/63501),
[#​63373](https://redirect.github.com/ray-project/ray/issues/63373),
[#​63258](https://redirect.github.com/ray-project/ray/issues/63258));
better worker startup error messages
([#​63714](https://redirect.github.com/ray-project/ray/issues/63714))
- Warn when `runtime_env` package approaches upload size limit
([#​63404](https://redirect.github.com/ray-project/ray/issues/63404));
harden zip extraction path containment
([#​63786](https://redirect.github.com/ray-project/ray/issues/63786),
[#​62813](https://redirect.github.com/ray-project/ray/issues/62813))
- Include owner node ID in `OwnerDiedError`
([#​63727](https://redirect.github.com/ray-project/ray/issues/63727));
add dependency info to taskspec debug string
([#​62316](https://redirect.github.com/ray-project/ray/issues/62316))
- Add unexpected worker failure metric and dashboard panel
([#​62297](https://redirect.github.com/ray-project/ray/issues/62297));
group observability APIs in `ray` CLI help
([#​62748](https://redirect.github.com/ray-project/ray/issues/62748))
- Normalize OTel metric labels before Prometheus export
([#​63744](https://redirect.github.com/ray-project/ray/issues/63744))
and retry/log when Prometheus queries fail
([#​63578](https://redirect.github.com/ray-project/ray/issues/63578));
add GPU usage instance filter
([#​62214](https://redirect.github.com/ray-project/ray/issues/62214))
- Move observability and control-plane pubsub to dedicated services and
rename `InternalPubSub*` to `ControlPlanePubSub*`
([#​62806](https://redirect.github.com/ray-project/ray/issues/62806),
[#​63044](https://redirect.github.com/ray-project/ray/issues/63044),
[#​62461](https://redirect.github.com/ray-project/ray/issues/62461))
- AMD GPU: replace `rocm-smi` ctypes binding with `amd-smi` Python
interface
([#​62393](https://redirect.github.com/ray-project/ray/issues/62393));
detect NVIDIA Blackwell consumer GPUs
([#​63322](https://redirect.github.com/ray-project/ray/issues/63322))
- Add task retry delay for `ACTOR_UNAVAILABLE` retries
([#​62330](https://redirect.github.com/ray-project/ray/issues/62330));
improve State API filter key handling
([#​63638](https://redirect.github.com/ray-project/ray/issues/63638))
- Patch `setproctitle` to skip launch services IPC calls
([#​63366](https://redirect.github.com/ray-project/ray/issues/63366));
add timeout for first redis probe
([#​63148](https://redirect.github.com/ray-project/ray/issues/63148))
- Clarify head node commands in `ray up` output
([#​63409](https://redirect.github.com/ray-project/ray/issues/63409));
pass `logging_config` through Ray Client `ray.init`
([#​62192](https://redirect.github.com/ray-project/ray/issues/62192))
- Print subprocess log tails with exit codes on unexpected exit
([#​61905](https://redirect.github.com/ray-project/ray/issues/61905));
add warning log when GPU profiling command times out
([#​63706](https://redirect.github.com/ray-project/ray/issues/63706))
- Add unique suffix to log filenames
([#​62365](https://redirect.github.com/ray-project/ray/issues/62365));
disable profiling endpoints by default
([#​62531](https://redirect.github.com/ray-project/ray/issues/62531))
- Remove pydantic v1 support
([#​62716](https://redirect.github.com/ray-project/ray/issues/62716));
update Starlette to v1.0.1
([#​63722](https://redirect.github.com/ray-project/ray/issues/63722))
- Deprecate `DAGNode.execute()`
([#​63716](https://redirect.github.com/ray-project/ray/issues/63716));
remove experimental `_owner` support for `ray.put`
([#​63520](https://redirect.github.com/ray-project/ray/issues/63520))
##### 🔨 Fixes
- Fix `ray.get` hanging forever when an object's owner dies during pull
([#​63694](https://redirect.github.com/ray-project/ray/issues/63694));
resolve `ReferenceCounter` race on `WORKER_REF_REMOVED_CHANNEL`
([#​60495](https://redirect.github.com/ray-project/ray/issues/60495))
- Fix resource leaks in subprocess management
([#​63878](https://redirect.github.com/ray-project/ray/issues/63878))
and `runtime_env` cache not detecting changes in `-r`-referenced
requirements files
([#​63403](https://redirect.github.com/ray-project/ray/issues/63403))
- Fix replica actor zombie process after GCS restart
([#​63764](https://redirect.github.com/ray-project/ray/issues/63764));
fix actor creation race condition
([#​62994](https://redirect.github.com/ray-project/ray/issues/62994));
fix actor state counter bug
([#​63647](https://redirect.github.com/ray-project/ray/issues/63647))
- Fix placement groups with label domain stuck on the infeasible queue
([#​62483](https://redirect.github.com/ray-project/ray/issues/62483));
log status for failed PG `PrepareResources`/`CommitResources`
([#​62836](https://redirect.github.com/ray-project/ray/issues/62836))
- Fix env var expansion in `ray job submit` CLI via `shlex.join`
([#​63797](https://redirect.github.com/ray-project/ray/issues/63797))
and `--working-dir` for local zip files and `http://` URLs
([#​62843](https://redirect.github.com/ray-project/ray/issues/62843))
- Surface WebSocket close codes and errors in job log streaming
([#​63364](https://redirect.github.com/ray-project/ray/issues/63364));
fix `ray stop` failing to terminate dashboard/runtime\_env agents on
Windows
([#​62428](https://redirect.github.com/ray-project/ray/issues/62428))
- Fix `ray down` not stopping Docker containers on worker nodes for
local clusters
([#​62169](https://redirect.github.com/ray-project/ray/issues/62169));
fix delayed/missing worker logs in Jupyter by flushing stdout/stderr
([#​63599](https://redirect.github.com/ray-project/ray/issues/63599))
- Fix Python log monitor handling for same-inode truncated files
([#​63720](https://redirect.github.com/ray-project/ray/issues/63720));
avoid `os.getcwd()` on import by lazily evaluating `scratch_dir`
([#​63040](https://redirect.github.com/ray-project/ray/issues/63040))
- Fix accelerator detection on NVIDIA Blackwell consumer GPUs
([#​63322](https://redirect.github.com/ray-project/ray/issues/63322));
avoid `FabricManager` stall on NVLink systems in `GpuProfilingManager`
([#​63312](https://redirect.github.com/ray-project/ray/issues/63312))
- Fix POSIX semaphore crash in experimental mutable objects
([#​62328](https://redirect.github.com/ray-project/ray/issues/62328));
fix overflow on exponential backoff multiplication
([#​62366](https://redirect.github.com/ray-project/ray/issues/62366))
- Fix OOM kill message wrong threshold with resource isolation
([#​62948](https://redirect.github.com/ray-project/ray/issues/62948));
fix `OpenTelemetryMetricRecorder` singleton init guard
([#​63081](https://redirect.github.com/ray-project/ray/issues/63081))
- Fix `MarkFootprintAsBusy` clearing saved idle state for unrelated
items
([#​62588](https://redirect.github.com/ray-project/ray/issues/62588));
fix `HandleIsLocalWorkerDead` for drivers
([#​62688](https://redirect.github.com/ray-project/ray/issues/62688))
- Fix `AttributeError` on `trace` in client mode
([#​62955](https://redirect.github.com/ray-project/ray/issues/62955));
fix `IndexError` in legacy post-mortem debugging ([#&#
> ✂ **Note**
>
> PR body was truncated to here.
</details>
---
### Configuration
📅 **Schedule**: (UTC)
- Branch creation
- At any time (no schedule defined)
- Automerge
- At any time (no schedule defined)
🚦 **Automerge**: Disabled by config. Please merge this manually once you
are satisfied.
♻ **Rebasing**: Whenever PR becomes conflicted, or you tick the
rebase/retry checkbox.
🔕 **Ignore**: Close this PR and you won't be reminded about these
updates again.
---
- [ ] <!-- rebase-check -->If you want to rebase/retry this PR, check
this box
---
This PR was generated by [Mend Renovate](https://mend.io/renovate/).
View the [repository job
log](https://developer.mend.io/github/vortex-data/vortex).
<!--renovate-debug:eyJjcmVhdGVkSW5WZXIiOiI0My4yNTkuMiIsInVwZGF0ZWRJblZlciI6IjQzLjI1OS4yIiwidGFyZ2V0QnJhbmNoIjoiZGV2ZWxvcCIsImxhYmVscyI6WyJjaGFuZ2Vsb2cvY2hvcmUiXX0=-->
Co-authored-by: renovate[bot] <29139614+renovate[bot]@users.noreply.github.com>1 parent 30fff30 commit 452e0f8
1 file changed
Lines changed: 49 additions & 49 deletions
Some generated files are not rendered by default. Learn more about customizing how changed files appear on GitHub.
0 commit comments