[train] Default to disabling Ray Train collective util timeouts#58229
Merged
justinvyu merged 7 commits intoray-project:masterfrom Nov 7, 2025
Merged
[train] Default to disabling Ray Train collective util timeouts#58229justinvyu merged 7 commits intoray-project:masterfrom
justinvyu merged 7 commits intoray-project:masterfrom
Conversation
Signed-off-by: Justin Yu <justinvyu@anyscale.com>
Signed-off-by: Justin Yu <justinvyu@anyscale.com>
Signed-off-by: Justin Yu <justinvyu@anyscale.com>
Contributor
There was a problem hiding this comment.
Code Review
This pull request correctly implements the plan to disable Ray Train collective utility timeouts by default, which should improve user experience by reducing deadlocks and configuration overhead. The renaming of environment variables from REPORT_BARRIER to a more general COLLECTIVE is a good improvement. The logic changes and the addition of a new test case to verify the behavior with no timeout are solid. I've added a couple of suggestions: one to fix a formatting issue in an error message and another to make a new test more explicit and robust.
TimothySeah
approved these changes
Oct 28, 2025
…ve_collective_timeout
Signed-off-by: Justin Yu <justinvyu@anyscale.com>
YoussefEssDS
pushed a commit
to YoussefEssDS/ray
that referenced
this pull request
Nov 8, 2025
…project#58229) Ray Train's framework agnostic collective utilities (`ray.train.collective.barrier`, `ray.train.collective.broadcast_from_rank_zero`) currently timeout after 30 minutes if not all ranks join the operation. `ray.train.report` uses these collective utilities internally, so users who don't call report on every rank can run into deadlocks. For example, the report barrier can deadlock with another worker waiting on others to join a backward pass collective. This PR changes the default Ray Train collective behavior to never timeout and to only log warning messages about the missing ranks. User code typically already has timeouts such as NCCL timeouts (also 30 minutes by default), so the extra timeout here doesn't really help and increases the user burden of keeping track of environment variables to set when debugging hanging jobs. New default: `RAY_TRAIN_COLLECTIVE_TIMEOUT_S=-1` This PR also generalizes the environment variable name: `RAY_TRAIN_REPORT_BARRIER` -> `RAY_TRAIN_COLLECTIVE`. --------- Signed-off-by: Justin Yu <justinvyu@anyscale.com>
landscapepainter
pushed a commit
to landscapepainter/ray
that referenced
this pull request
Nov 17, 2025
…project#58229) Ray Train's framework agnostic collective utilities (`ray.train.collective.barrier`, `ray.train.collective.broadcast_from_rank_zero`) currently timeout after 30 minutes if not all ranks join the operation. `ray.train.report` uses these collective utilities internally, so users who don't call report on every rank can run into deadlocks. For example, the report barrier can deadlock with another worker waiting on others to join a backward pass collective. This PR changes the default Ray Train collective behavior to never timeout and to only log warning messages about the missing ranks. User code typically already has timeouts such as NCCL timeouts (also 30 minutes by default), so the extra timeout here doesn't really help and increases the user burden of keeping track of environment variables to set when debugging hanging jobs. New default: `RAY_TRAIN_COLLECTIVE_TIMEOUT_S=-1` This PR also generalizes the environment variable name: `RAY_TRAIN_REPORT_BARRIER` -> `RAY_TRAIN_COLLECTIVE`. --------- Signed-off-by: Justin Yu <justinvyu@anyscale.com>
Aydin-ab
pushed a commit
to Aydin-ab/ray-aydin
that referenced
this pull request
Nov 19, 2025
…project#58229) Ray Train's framework agnostic collective utilities (`ray.train.collective.barrier`, `ray.train.collective.broadcast_from_rank_zero`) currently timeout after 30 minutes if not all ranks join the operation. `ray.train.report` uses these collective utilities internally, so users who don't call report on every rank can run into deadlocks. For example, the report barrier can deadlock with another worker waiting on others to join a backward pass collective. This PR changes the default Ray Train collective behavior to never timeout and to only log warning messages about the missing ranks. User code typically already has timeouts such as NCCL timeouts (also 30 minutes by default), so the extra timeout here doesn't really help and increases the user burden of keeping track of environment variables to set when debugging hanging jobs. New default: `RAY_TRAIN_COLLECTIVE_TIMEOUT_S=-1` This PR also generalizes the environment variable name: `RAY_TRAIN_REPORT_BARRIER` -> `RAY_TRAIN_COLLECTIVE`. --------- Signed-off-by: Justin Yu <justinvyu@anyscale.com> Signed-off-by: Aydin Abiar <aydin@anyscale.com>
ykdojo
pushed a commit
to ykdojo/ray
that referenced
this pull request
Nov 27, 2025
…project#58229) Ray Train's framework agnostic collective utilities (`ray.train.collective.barrier`, `ray.train.collective.broadcast_from_rank_zero`) currently timeout after 30 minutes if not all ranks join the operation. `ray.train.report` uses these collective utilities internally, so users who don't call report on every rank can run into deadlocks. For example, the report barrier can deadlock with another worker waiting on others to join a backward pass collective. This PR changes the default Ray Train collective behavior to never timeout and to only log warning messages about the missing ranks. User code typically already has timeouts such as NCCL timeouts (also 30 minutes by default), so the extra timeout here doesn't really help and increases the user burden of keeping track of environment variables to set when debugging hanging jobs. New default: `RAY_TRAIN_COLLECTIVE_TIMEOUT_S=-1` This PR also generalizes the environment variable name: `RAY_TRAIN_REPORT_BARRIER` -> `RAY_TRAIN_COLLECTIVE`. --------- Signed-off-by: Justin Yu <justinvyu@anyscale.com> Signed-off-by: YK <1811651+ykdojo@users.noreply.github.com>
SheldonTsen
pushed a commit
to SheldonTsen/ray
that referenced
this pull request
Dec 1, 2025
…project#58229) Ray Train's framework agnostic collective utilities (`ray.train.collective.barrier`, `ray.train.collective.broadcast_from_rank_zero`) currently timeout after 30 minutes if not all ranks join the operation. `ray.train.report` uses these collective utilities internally, so users who don't call report on every rank can run into deadlocks. For example, the report barrier can deadlock with another worker waiting on others to join a backward pass collective. This PR changes the default Ray Train collective behavior to never timeout and to only log warning messages about the missing ranks. User code typically already has timeouts such as NCCL timeouts (also 30 minutes by default), so the extra timeout here doesn't really help and increases the user burden of keeping track of environment variables to set when debugging hanging jobs. New default: `RAY_TRAIN_COLLECTIVE_TIMEOUT_S=-1` This PR also generalizes the environment variable name: `RAY_TRAIN_REPORT_BARRIER` -> `RAY_TRAIN_COLLECTIVE`. --------- Signed-off-by: Justin Yu <justinvyu@anyscale.com>
Future-Outlier
pushed a commit
to Future-Outlier/ray
that referenced
this pull request
Dec 7, 2025
…project#58229) Ray Train's framework agnostic collective utilities (`ray.train.collective.barrier`, `ray.train.collective.broadcast_from_rank_zero`) currently timeout after 30 minutes if not all ranks join the operation. `ray.train.report` uses these collective utilities internally, so users who don't call report on every rank can run into deadlocks. For example, the report barrier can deadlock with another worker waiting on others to join a backward pass collective. This PR changes the default Ray Train collective behavior to never timeout and to only log warning messages about the missing ranks. User code typically already has timeouts such as NCCL timeouts (also 30 minutes by default), so the extra timeout here doesn't really help and increases the user burden of keeping track of environment variables to set when debugging hanging jobs. New default: `RAY_TRAIN_COLLECTIVE_TIMEOUT_S=-1` This PR also generalizes the environment variable name: `RAY_TRAIN_REPORT_BARRIER` -> `RAY_TRAIN_COLLECTIVE`. --------- Signed-off-by: Justin Yu <justinvyu@anyscale.com> Signed-off-by: Future-Outlier <eric901201@gmail.com>
peterxcli
pushed a commit
to peterxcli/ray
that referenced
this pull request
Feb 25, 2026
…project#58229) Ray Train's framework agnostic collective utilities (`ray.train.collective.barrier`, `ray.train.collective.broadcast_from_rank_zero`) currently timeout after 30 minutes if not all ranks join the operation. `ray.train.report` uses these collective utilities internally, so users who don't call report on every rank can run into deadlocks. For example, the report barrier can deadlock with another worker waiting on others to join a backward pass collective. This PR changes the default Ray Train collective behavior to never timeout and to only log warning messages about the missing ranks. User code typically already has timeouts such as NCCL timeouts (also 30 minutes by default), so the extra timeout here doesn't really help and increases the user burden of keeping track of environment variables to set when debugging hanging jobs. New default: `RAY_TRAIN_COLLECTIVE_TIMEOUT_S=-1` This PR also generalizes the environment variable name: `RAY_TRAIN_REPORT_BARRIER` -> `RAY_TRAIN_COLLECTIVE`. --------- Signed-off-by: Justin Yu <justinvyu@anyscale.com> Signed-off-by: peterxcli <peterxcli@gmail.com>
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Description
Ray Train's framework agnostic collective utilities (
ray.train.collective.barrier,ray.train.collective.broadcast_from_rank_zero) currently timeout after 30 minutes if not all ranks join the operation.ray.train.reportuses these collective utilities internally, so users who don't call report on every rank can run into deadlocks. For example, the report barrier can deadlock with another worker waiting on others to join a backward pass collective.This PR changes the default Ray Train collective behavior to never timeout and to only log warning messages about the missing ranks. User code typically already has timeouts such as NCCL timeouts (also 30 minutes by default), so the extra timeout here doesn't really help and increases the user burden of keeping track of environment variables to set when debugging hanging jobs.
New default:
RAY_TRAIN_COLLECTIVE_TIMEOUT_S=-1This PR also generalizes the environment variable name:
RAY_TRAIN_REPORT_BARRIER->RAY_TRAIN_COLLECTIVE.