Issues

[Data] Dataset completes with missing rows after STREAMING_GENERATOR_REPLAY_INCONSISTENT on a map_groups task (Ray 2.58.0)

ray-project/ray#66917 · opened

on Oct 10, 2026

1

[Data] Filters on synthesized path columns can return incorrect rows

ray-project/ray#66916 · opened

on Oct 10, 2026

1

[Data] Partition pruning is inconsistent with the Dataset's logical schema

ray-project/ray#66915 · opened

on Oct 10, 2026

1

[Data] Parquet V2 can silently drop rows when root files coexist with Hive partitions

ray-project/ray#66914 · opened

on Oct 10, 2026

1

[Core][Serve] Serve controller inefficiently fetches resources info from GCS and degrades performance

ray-project/ray#66912 · opened

on Oct 10, 2026

[Data] Actor-pool scale-down leaves primary object-store pins, so worker nodes never go idle

ray-project/ray#66903 · opened

on Oct 9, 2026

1

[Core] Option to retry tasks only after node preemption

ray-project/ray#66901 · opened

on Oct 9, 2026

[core][rdt] Add execution-ucx (Axon) as an RDT backend

ray-project/ray#66890 · opened

on Oct 9, 2026

1

[Data] Include input file paths in JSON/JSONL parsing errors

ray-project/ray#66877 · opened

on Oct 9, 2026

[core] Resource leak on Raylet when CommitBundleResources RPC fails and retries

ray-project/ray#66873 · opened

on Oct 9, 2026

[Data] Support projection, filter, and limit pushdown for read_hive table reads

ray-project/ray#66843 · opened

on Oct 8, 2026

[Data] Make null ordering consistent between Arrow/Pandas and Polars sort

ray-project/ray#66841 · opened

on Oct 8, 2026

1

[Data] Support optional query-schema inference in read_hive

ray-project/ray#66840 · opened

on Oct 8, 2026

2

[data] Runtime Bloom-filter pushdown to Join

ray-project/ray#66839 · opened

on Oct 8, 2026

[data] Broadcast hash join

ray-project/ray#66838 · opened

on Oct 8, 2026

[data] Plan-level common subtree reuse (CSE on the logical DAG)

ray-project/ray#66837 · opened

on Oct 8, 2026

1

[data] Eager aggregation: push partial aggregates below joins

ray-project/ray#66836 · opened

on Oct 8, 2026

1

[data] Evaluate residual (non-equi) join predicates at probe time

ray-project/ray#66835 · opened

on Oct 8, 2026

[data] Join condition slot + automatic equi-join key extraction

ray-project/ray#66834 · opened

on Oct 8, 2026

[data] Derive required columns through Join in ProjectionPushdown

ray-project/ray#66833 · opened

on Oct 8, 2026

[data] Predicate pushdown through joins: split conjuncts, fix fusion ordering, derive one-sided OR weakenings

ray-project/ray#66832 · opened

on Oct 8, 2026

[data] Push Limit through Sort (Top-N execution)

ray-project/ray#66831 · opened

on Oct 8, 2026

1

[data] Track output partitioning as a physical property and eliminate redundant shuffles

ray-project/ray#66830 · opened

on Oct 8, 2026

1

[data] Umbrella: query optimizer gaps identified by TPC-H benchmarking

ray-project/ray#66829 · opened

on Oct 8, 2026

[Serve] gRPC servers keep SO_REUSEPORT on with enable_so_reuseport=False, so direct ingress ports can be shared

ray-project/ray#66815 · opened

on Oct 7, 2026

Navigation Menu

Repository navigation

Issues

[Data] Dataset completes with missing rows after STREAMING_GENERATOR_REPLAY_INCONSISTENT on a map_groups task (Ray 2.58.0)

[Data] Filters on synthesized path columns can return incorrect rows

[Data] Partition pruning is inconsistent with the Dataset's logical schema

[Data] Parquet V2 can silently drop rows when root files coexist with Hive partitions

[Core][Serve] Serve controller inefficiently fetches resources info from GCS and degrades performance

[Data] Actor-pool scale-down leaves primary object-store pins, so worker nodes never go idle

[Core] Option to retry tasks only after node preemption

[core][rdt] Add execution-ucx (Axon) as an RDT backend

[Data] Include input file paths in JSON/JSONL parsing errors

[core] Resource leak on Raylet when CommitBundleResources RPC fails and retries

[Data] Support projection, filter, and limit pushdown for read_hive table reads

[Data] Make null ordering consistent between Arrow/Pandas and Polars sort

[Data] Support optional query-schema inference in read_hive

[data] Runtime Bloom-filter pushdown to Join

[data] Broadcast hash join

[data] Plan-level common subtree reuse (CSE on the logical DAG)

[data] Eager aggregation: push partial aggregates below joins

[data] Evaluate residual (non-equi) join predicates at probe time

[data] Join condition slot + automatic equi-join key extraction

[data] Derive required columns through Join in ProjectionPushdown

[data] Predicate pushdown through joins: split conjuncts, fix fusion ordering, derive one-sided OR weakenings

[data] Push Limit through Sort (Top-N execution)

[data] Track output partitioning as a physical property and eliminate redundant shuffles

[data] Umbrella: query optimizer gaps identified by TPC-H benchmarking

[Serve] gRPC servers keep SO_REUSEPORT on with enable_so_reuseport=False, so direct ingress ports can be shared