[TRTLLM-10688][fix] fix cross-node rollout issues in verl#11924
[TRTLLM-10688][fix] fix cross-node rollout issues in verl#11924hchings merged 8 commits intoNVIDIA:mainfrom
Conversation
Signed-off-by: Erin Ho <14718778+hchings@users.noreply.github.com>
hchings
changed the title
hchings
changed the title
📝 WalkthroughWalkthroughThese changes improve Ray worker environment isolation by filtering node-local environment variables in the executor and establishing per-worker DeepGemm JIT cache directories to prevent file conflicts among co-located workers. Changes
Estimated code review effort🎯 2 (Simple) | ⏱️ ~10 minutes 🚥 Pre-merge checks | ✅ 1 | ❌ 2❌ Failed checks (2 warnings)
✅ Passed checks (1 passed)
✏️ Tip: You can configure your own custom pre-merge checks in the settings. ✨ Finishing Touches🧪 Generate unit tests (beta)
📝 Coding Plan
Comment Tip You can customize the high-level summary generated by CodeRabbit.Configure the |
![coderabbitai[bot]](https://avatars.githubusercontent.com/in/347564?s=60&v=4){kind=small}
There was a problem hiding this comment.
Actionable comments posted: 1
🤖 Prompt for all review comments with AI agents
Verify each finding against the current code and only fix it if needed.
Inline comments:
In `@tensorrt_llm/executor/ray_gpu_worker.py`:
- Around line 62-63: The DG_JIT_CACHE_DIR value currently uses only rank and
self.gpu and can collide across separate Ray jobs; update the assignment in
ray_gpu_worker.py to generate a unique per-process/job directory (e.g., use
tempfile.mkdtemp or append a UUID/pid) and set os.environ["DG_JIT_CACHE_DIR"] to
that path; also add import tempfile to the stdlib imports (or import
uuid/os.getpid if using those) so each executor gets a non-colliding cache
directory (refer to the DG_JIT_CACHE_DIR assignment and use of rank and self.gpu
to locate the change).
ℹ️ Review info
⚙️ Run configuration
Configuration used: Path: .coderabbit.yaml
Review profile: CHILL
Plan: Pro
Run ID: 98ec96f6-e59c-48d5-9852-b5bbca2cefd7
📒 Files selected for processing (2)
tensorrt_llm/executor/ray_executor.pytensorrt_llm/executor/ray_gpu_worker.py
Signed-off-by: Erin Ho <14718778+hchings@users.noreply.github.com>
|
/bot run |
|
PR_Github #38398 [ run ] triggered by Bot. Commit: |
|
PR_Github #38398 [ run ] completed with state
|
Removed redundant line break in JIT cache directory setup. Signed-off-by: Erin <14718778+hchings@users.noreply.github.com>
|
/bot run |
|
PR_Github #38536 [ run ] triggered by Bot. Commit: |
|
PR_Github #38536 [ run ] completed with state
|
|
/bot run |
|
PR_Github #38621 [ run ] triggered by Bot. Commit: |
|
PR_Github #38621 [ run ] completed with state
|
|
/bot run |
|
PR_Github #38633 [ run ] triggered by Bot. Commit: |
|
PR_Github #38633 [ run ] completed with state
|
|
/bot run |
|
PR_Github #38784 [ run ] triggered by Bot. Commit: |
|
PR_Github #38784 [ run ] completed with state |
|
/bot run |
|
PR_Github #38838 [ run ] triggered by Bot. Commit: |
|
PR_Github #38838 [ run ] completed with state
|
|
/bot run |
|
PR_Github #38906 [ run ] triggered by Bot. Commit: |
|
PR_Github #38906 [ run ] completed with state
|
|
/bot run |
|
PR_Github #39130 [ run ] triggered by Bot. Commit: |
|
PR_Github #39130 [ run ] completed with state
|
|
/bot run |
|
PR_Github #39303 [ run ] triggered by Bot. Commit: |
|
PR_Github #39303 [ run ] completed with state
|
|
/bot run |
|
PR_Github #39556 [ run ] triggered by Bot. Commit: |
|
PR_Github #39556 [ run ] completed with state |
3 participants
Summary by CodeRabbit
Description
Test Coverage
PR Checklist
Please review the following before submitting your PR:
PR description clearly explains what and why. If using CodeRabbit's summary, please make sure it makes sense.
PR Follows TRT-LLM CODING GUIDELINES to the best of your knowledge.
Test cases are provided for new code paths (see test instructions)
Any new dependencies have been scanned for license and vulnerabilities
CODEOWNERS updated if ownership changes
Documentation updated as needed
Update tava architecture diagram if there is a significant design change in PR.
The reviewers assigned automatically/manually are appropriate for the PR.
Please check this after reviewing the above items as appropriate for this PR.
GitHub Bot Help
To see a list of available CI bot commands, please comment
/bot help.