Issues

[Data] A filter that uses a with_column column is not pushed into the read, even its conditions on stored columns

ray-project/ray#67643 · opened

on Oct 30, 2026

[Data] read_json Hive null partitions are written to Parquet as a null-typed column

ray-project/ray#67606 · opened

on Oct 30, 2026

[Data] repartition() warns that shuffle is ignored before rejecting the same argument

ray-project/ray#67604 · opened

on Oct 30, 2026

1

[Tune] A failed BayesOpt warm-up trial prevents further configurations from running

ray-project/ray#67583 · opened

on Oct 30, 2026

[Core] Pool map_async and starmap_async callbacks silently reorder results

ray-project/ray#67581 · opened

on Oct 30, 2026

[Data] read_parquet ignores Parquet bloom filters, so point lookups on unsorted columns read every row group

ray-project/ray#67569 · opened

on Oct 29, 2026

[Data] filter().count() decodes files whose footers already prove every row matches

ray-project/ray#67568 · opened

on Oct 29, 2026

1

[Data] An OR across a partition column and a data column prunes nothing

ray-project/ray#67567 · opened

on Oct 29, 2026

[Data] Arithmetic and string prefixes around a column disable Parquet statistics pruning

ray-project/ray#67565 · opened

on Oct 29, 2026

[Data] min()/max() aren't answered from Parquet footer statistics

ray-project/ray#67564 · opened

on Oct 29, 2026

[Data] Streaming executor busy-spins while output-backpressured and holds the GIL, starving the consuming training loop (iter_batches / iter_torch_batches)

ray-project/ray#67547 · opened

on Oct 29, 2026

1

[Data] Batcher copies the whole multi-chunk input block into worker memory before the first batch

ray-project/ray#67527 · opened

on Oct 29, 2026

[Core] Ray Client ray.put and task arguments over 2 GiB fail with EncodeError: Failed to serialize proto

ray-project/ray#67515 · opened

on Oct 28, 2026

[Serve] Gang scheduling + autoscaling: bounds that aren't multiples of gang_size cause DEPLOY_FAILED or endless scale flapping

ray-project/ray#67494 · opened

on Oct 28, 2026

1

[Autoscaler v2][KubeRay] Instances stay REQUESTED forever after a scale-up is capped at maxReplicas: the re-queue reuses an already-processed request id

ray-project/ray#67477 · opened

on Oct 27, 2026

2

[core] cgroup memory and CPU limits are read at the mount root only, not at the cgroup of the process

ray-project/ray#67473 · opened

on Oct 27, 2026

[Core] Cross-node channel transfer stalls when payload size is an exact multiple of kMaxGrpcPayloadSize in PushMutableObject

ray-project/ray#67471 · opened

on Oct 27, 2026

[Data] Copy batches only once with `zero_copy_batch=False` in the remaining cases

ray-project/ray#67459 · opened

on Oct 27, 2026

2

[Core/Auth] Make `ray.cluster_utils.Cluster` stop depending on `os.environ` for the auth mode

ray-project/ray#67447 · opened

on Oct 26, 2026

[Core] Python worker fork server

ray-project/ray#67446 · opened

on Oct 26, 2026

1

[Data][LLM] Processor silently overwrites repeated stages

ray-project/ray#67419 · opened

on Oct 26, 2026

[Core][Dashboard] `--dashboard-host <specific IP>` advertises the node IP in GCS, so `ray job` / `ray list` / `ray logs` can't connect

ray-project/ray#67416 · opened

on Oct 26, 2026

1

[Data] `write_delta` creates table with pre-transformation schema when a fused transformation produces no rows

ray-project/ray#67412 · opened

on Oct 26, 2026

1

Outdated starlette (1.0.1) pinned in python/requirements.txt — 2 HIGH severity CVEs unpatched

ray-project/ray#67411 · opened

on Oct 26, 2026

[Serve] @serve.batch silently mixes up arguments when callers pass them in a different order

ray-project/ray#67404 · opened

on Oct 25, 2026

Navigation Menu

Repository navigation

Issues

[Data] A filter that uses a with_column column is not pushed into the read, even its conditions on stored columns

[Data] read_json Hive null partitions are written to Parquet as a null-typed column

[Data] repartition() warns that shuffle is ignored before rejecting the same argument

[Tune] A failed BayesOpt warm-up trial prevents further configurations from running

[Core] Pool map_async and starmap_async callbacks silently reorder results

[Data] read_parquet ignores Parquet bloom filters, so point lookups on unsorted columns read every row group

[Data] filter().count() decodes files whose footers already prove every row matches

[Data] An OR across a partition column and a data column prunes nothing

[Data] Arithmetic and string prefixes around a column disable Parquet statistics pruning

[Data] min()/max() aren't answered from Parquet footer statistics

[Data] Streaming executor busy-spins while output-backpressured and holds the GIL, starving the consuming training loop (iter_batches / iter_torch_batches)

[Data] Batcher copies the whole multi-chunk input block into worker memory before the first batch

[Core] Ray Client ray.put and task arguments over 2 GiB fail with EncodeError: Failed to serialize proto

[Serve] Gang scheduling + autoscaling: bounds that aren't multiples of gang_size cause DEPLOY_FAILED or endless scale flapping

[Autoscaler v2][KubeRay] Instances stay REQUESTED forever after a scale-up is capped at maxReplicas: the re-queue reuses an already-processed request id

[core] cgroup memory and CPU limits are read at the mount root only, not at the cgroup of the process

[Core] Cross-node channel transfer stalls when payload size is an exact multiple of kMaxGrpcPayloadSize in PushMutableObject

[Data] Copy batches only once with `zero_copy_batch=False` in the remaining cases

[Core/Auth] Make `ray.cluster_utils.Cluster` stop depending on `os.environ` for the auth mode

[Core] Python worker fork server

[Data][LLM] Processor silently overwrites repeated stages

[Core][Dashboard] `--dashboard-host <specific IP>` advertises the node IP in GCS, so `ray job` / `ray list` / `ray logs` can't connect

[Data] `write_delta` creates table with pre-transformation schema when a fused transformation produces no rows

Outdated starlette (1.0.1) pinned in python/requirements.txt — 2 HIGH severity CVEs unpatched

[Serve] @serve.batch silently mixes up arguments when callers pass them in a different order