Issues

[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

[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] Dataset.join(validate_schemas=True) mishandles asymmetric and multi-key joins

ray-project/ray#67415 · opened

on Oct 26, 2026

4

[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

Outdated starlette (1.0.1) pinned in python/requirements.txt — 2 HIGH severity CVEs unpatched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

[Data] Dataset.mix does not consistently reject non-finite weights

ray-project/ray#67387 · opened

on Oct 25, 2026

4

[Core] Redundant replacement actor lease can retain resources after an earlier grant is adopted

ray-project/ray#67383 · opened

on Oct 25, 2026

[Autoscaler][v2] InstanceManager drops transition details from history and status output

ray-project/ray#67380 · opened

on Oct 25, 2026

[Serve] graceful_shutdown_wait_loop_s=0 busy-loops the drain and floods replica logs

ray-project/ray#67379 · opened

on Oct 25, 2026

5

[Data] map_batches loses explicit Pandas extension dtype for all-null columns

ray-project/ray#67372 · opened

on Oct 24, 2026

1

[Core] GCS resource-load puller fetches unused fields add costs to GCS and Raylet main thread

ray-project/ray#67369 · opened

on Oct 24, 2026

[Data] map_batches output with string_view/binary_view columns grows with the number of batches

ray-project/ray#67368 · opened

on Oct 24, 2026

2

[Data] zero_copy_batch=False doesn't give writable NumPy batches in map_groups and for list columns

ray-project/ray#67367 · opened

on Oct 24, 2026

[Data] Hive null partitions are read back as the string "__HIVE_DEFAULT_PARTITION__"

ray-project/ray#67362 · opened

on Oct 24, 2026

[Core] Process-group cleanup can call killpg(1) (= kill(-1)) when a worker reports a pid from another PID namespace

ray-project/ray#67348 · opened

on Oct 23, 2026

[Data] DownstreamCapacityBackpressurePolicy applies no backpressure when the downstream operator has 0 bytes in flight, so an operator's output queue can fill the whole object-store budget

ray-project/ray#67340 · opened

on Oct 23, 2026

1

[Data] read_sql can silently drop rows with NULL shard keys

ray-project/ray#67318 · opened

on Oct 23, 2026

1

[RLlib] Restored multi-agent EnvRunner retains stale routing after weights recover

ray-project/ray#67317 · opened

on Oct 23, 2026

Navigation Menu

Repository navigation

Issues

[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] Dataset.join(validate_schemas=True) mishandles asymmetric and multi-key joins

[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

[Data] Dataset.mix does not consistently reject non-finite weights

[Core] Redundant replacement actor lease can retain resources after an earlier grant is adopted

[Autoscaler][v2] InstanceManager drops transition details from history and status output

[Serve] graceful_shutdown_wait_loop_s=0 busy-loops the drain and floods replica logs

[Data] map_batches loses explicit Pandas extension dtype for all-null columns

[Core] GCS resource-load puller fetches unused fields add costs to GCS and Raylet main thread

[Data] map_batches output with string_view/binary_view columns grows with the number of batches

[Data] zero_copy_batch=False doesn't give writable NumPy batches in map_groups and for list columns

[Data] Hive null partitions are read back as the string "__HIVE_DEFAULT_PARTITION__"

[Core] Process-group cleanup can call killpg(1) (= kill(-1)) when a worker reports a pid from another PID namespace

[Data] DownstreamCapacityBackpressurePolicy applies no backpressure when the downstream operator has 0 bytes in flight, so an operator's output queue can fill the whole object-store budget

[Data] read_sql can silently drop rows with NULL shard keys

[RLlib] Restored multi-agent EnvRunner retains stale routing after weights recover