Skip to content

[https://nvbugs/5979673][fix] Pin nixl-cu13==0.9.0 instead of nixl meta-package - #14567

Closed
Shixiaowei02 wants to merge 2 commits into
NVIDIA:mainfrom
Shixiaowei02:user/xiaoweis/pin-nixl-cu13
Closed

[https://nvbugs/5979673][fix] Pin nixl-cu13==0.9.0 instead of nixl meta-package#14567
Shixiaowei02 wants to merge 2 commits into
NVIDIA:mainfrom
Shixiaowei02:user/xiaoweis/pin-nixl-cu13

Conversation

@Shixiaowei02

@Shixiaowei02 Shixiaowei02 commented May 26, 2026

Copy link
Copy Markdown
Collaborator

The nixl PyPI package became a meta-package whose only dependency is nixl-cu12 with no version constraint, so nixl==0.9.0 resolves to nixl-cu12==<latest> (currently 1.1.0). The 1.1.0 bindings reference a new 3-arg overload of nixlAgent::prepXferDlist (with nixlAgentOptionalArgs) that does not exist in the source-built 0.9.0 libnixl.so shipped in the container, causing the test_agent unit test to fail at import time with an undefined-symbol error.

Pin nixl-cu13==0.9.0 directly so:

  • the wheel's _bindings.so only depends on the 4-arg prepXferDlist, which matches both the wheel's bundled libnixl and the source-built 0.9.0 libnixl;
  • the wheel variant aligns with the CUDA 13 container instead of cu12.

Summary by CodeRabbit

  • Chores
    • Updated a development dependency to its CUDA 13–specific build to align dev tooling with CUDA 13 environments.

Review Change Stack

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)

  • If PR introduces API changes, an appropriate PR label is added - either api-compatible or api-breaking. For api-breaking, include BREAKING in the PR title.

  • 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.

@Shixiaowei02
Shixiaowei02 requested a review from a team as a code owner May 26, 2026 09:53
@Shixiaowei02
Shixiaowei02 force-pushed the user/xiaoweis/pin-nixl-cu13 branch from 4e43b56 to c510e75 Compare May 26, 2026 09:54
@Shixiaowei02

Copy link
Copy Markdown
Collaborator Author

/bot run --disable-fail-fast

@coderabbitai

coderabbitai Bot commented May 26, 2026

Copy link
Copy Markdown
Contributor
📝 Walkthrough

Walkthrough

Replaces the development dependency nixl==0.9.0 with the CUDA 13-specific package nixl-cu13==0.9.0 in requirements-dev.txt.

Changes

CUDA 13 Dependency Update

Layer / File(s) Summary
CUDA 13-specific nixl dependency
requirements-dev.txt
Package dependency updated from nixl==0.9.0 to nixl-cu13==0.9.0.

Estimated code review effort

🎯 1 (Trivial) | ⏱️ ~2 minutes

Suggested reviewers

  • limin2021
  • chzblych
  • lfr-0531
🚥 Pre-merge checks | ✅ 4 | ❌ 1

❌ Failed checks (1 warning)

Check name Status Explanation Resolution
Description check ⚠️ Warning PR description provides detailed context and solution, but does not follow the provided template with clearly filled-in sections like 'Description' and 'Test Coverage'. Fill in the template sections: add a brief explanation in the 'Description' section and list relevant tests in 'Test Coverage' section that validate this dependency pinning change.
✅ Passed checks (4 passed)
Check name Status Explanation
Title check ✅ Passed The title clearly summarizes the main change: pinning nixl-cu13==0.9.0 instead of the nixl meta-package, which directly addresses the dependency resolution issue described in the PR objectives.
Docstring Coverage ✅ Passed No functions found in the changed files to evaluate docstring coverage. Skipping docstring coverage check.
Linked Issues check ✅ Passed Check skipped because no linked issues were found for this pull request.
Out of Scope Changes check ✅ Passed Check skipped because no linked issues were found for this pull request.

✏️ Tip: You can configure your own custom pre-merge checks in the settings.

✨ Finishing Touches
🧪 Generate unit tests (beta)
  • Create PR with unit tests

Comment @coderabbitai help to get the list of available commands and usage tips.

@tensorrt-cicd

Copy link
Copy Markdown
Collaborator

PR_Github #50341 [ run ] triggered by Bot. Commit: c510e75 Link to invocation

@tensorrt-cicd

Copy link
Copy Markdown
Collaborator

PR_Github #50341 [ run ] completed with state SUCCESS. Commit: c510e75
/LLM/main/L0_MergeRequest_PR pipeline #39868 completed with status: 'FAILURE'

CI Report

⚠️ Action Required:

  • Please check the failed tests and fix your PR
  • If you cannot view the failures, ask the CI triggerer to share details
  • Once fixed, request an NVIDIA team member to trigger CI again

CI Agent Failure Analysis

Link to invocation

@Shixiaowei02
Shixiaowei02 force-pushed the user/xiaoweis/pin-nixl-cu13 branch from c510e75 to 5cacbca Compare May 27, 2026 01:32
@Shixiaowei02

Copy link
Copy Markdown
Collaborator Author

/bot run --stage-list "A10-PyTorch-1" --disable-fail-fast

@tensorrt-cicd

Copy link
Copy Markdown
Collaborator

PR_Github #50417 [ run ] triggered by Bot. Commit: 5cacbca Link to invocation

@Shixiaowei02
Shixiaowei02 force-pushed the user/xiaoweis/pin-nixl-cu13 branch from 5cacbca to 4ab94b7 Compare May 27, 2026 03:16
@Shixiaowei02

Copy link
Copy Markdown
Collaborator Author

/bot run --disable-fail-fast

@tensorrt-cicd

Copy link
Copy Markdown
Collaborator

PR_Github #50438 [ run ] triggered by Bot. Commit: 4ab94b7 Link to invocation

@tensorrt-cicd

Copy link
Copy Markdown
Collaborator

PR_Github #50438 [ run ] completed with state FAILURE. Commit: 4ab94b7

Link to invocation

@Shixiaowei02
Shixiaowei02 force-pushed the user/xiaoweis/pin-nixl-cu13 branch from 4ab94b7 to 120a83f Compare May 27, 2026 05:04
@Shixiaowei02

Copy link
Copy Markdown
Collaborator Author

/bot run --disable-fail-fast

@tensorrt-cicd

Copy link
Copy Markdown
Collaborator

PR_Github #50461 [ run ] triggered by Bot. Commit: 120a83f Link to invocation

@tensorrt-cicd

Copy link
Copy Markdown
Collaborator

PR_Github #50461 [ run ] completed with state SUCCESS. Commit: 120a83f
/LLM/main/L0_MergeRequest_PR pipeline #39977 completed with status: 'FAILURE'

CI Report

⚠️ Action Required:

  • Please check the failed tests and fix your PR
  • If you cannot view the failures, ask the CI triggerer to share details
  • Once fixed, request an NVIDIA team member to trigger CI again

CI Agent Failure Analysis

Link to invocation

@Shixiaowei02
Shixiaowei02 requested a review from a team as a code owner May 29, 2026 03:51
@tensorrt-cicd

Copy link
Copy Markdown
Collaborator

PR_Github #50976 [ run ] triggered by Bot. Commit: 747e3be Link to invocation

@tensorrt-cicd

Copy link
Copy Markdown
Collaborator

PR_Github #50976 [ run ] completed with state SUCCESS. Commit: 747e3be
/LLM/main/L0_MergeRequest_PR pipeline #40430 completed with status: 'FAILURE'

CI Report

⚠️ Action Required:

  • Please check the failed tests and fix your PR
  • If you cannot view the failures, ask the CI triggerer to share details
  • Once fixed, request an NVIDIA team member to trigger CI again

CI Agent Failure Analysis

Link to invocation

@Shixiaowei02
Shixiaowei02 force-pushed the user/xiaoweis/pin-nixl-cu13 branch 2 times, most recently from c842a11 to b768245 Compare June 1, 2026 09:46
@Shixiaowei02

Copy link
Copy Markdown
Collaborator Author

/bot run --disable-fail-fast

The trtllm dev environment used to install the meta `nixl` package,
which provides a top-level `nixl/__init__.py` shim that redirects to
`nixl_cu13` / `nixl_cu12`. After
NVIDIA#14436 switched the
dependency to `nixl-cu13` directly, the meta shim is no longer present,
so `from nixl import nixl_agent, ...` in
`tensorrt_llm/_torch/disaggregation/nixl/_agent_py.py` raises
`ModuleNotFoundError: No module named 'nixl'` at import time, which in
turn breaks
`test_agent_multi_backends::test_run_with_different_env[1]` during
collection of the spawned subprocess pytest.

Resolve the symbols from `nixl_cu13` (or `nixl_cu12`) first and fall
back to the meta-package only when neither CUDA-specific package is
installed.

Signed-off-by: Shixiaowei02 <39303645+Shixiaowei02@users.noreply.github.com>
This case (TRTLLM_USE_PY_NIXL_KVCACHE=1) was waived under nvbugs/5979673
because the meta `nixl` package pulled the latest `nixl-cu12` wheel,
whose `_bindings.so` referenced symbols absent from the source-built
0.9.0 `libnixl.so`. With `nixl-cu13==0.9.0` directly pinned in
requirements-dev.txt, the bindings and the libnixl in the container are
ABI-consistent again, so the test is expected to pass.

Signed-off-by: Shixiaowei02 <39303645+Shixiaowei02@users.noreply.github.com>
@tensorrt-cicd

Copy link
Copy Markdown
Collaborator

PR_Github #51355 [ run ] triggered by Bot. Commit: b768245 Link to invocation

@Shixiaowei02
Shixiaowei02 force-pushed the user/xiaoweis/pin-nixl-cu13 branch from b768245 to 7b73fda Compare June 1, 2026 09:57
@Shixiaowei02

Copy link
Copy Markdown
Collaborator Author

/bot run --disable-fail-fast

@tensorrt-cicd

Copy link
Copy Markdown
Collaborator

PR_Github #51356 [ run ] triggered by Bot. Commit: 7b73fda Link to invocation

@tensorrt-cicd

Copy link
Copy Markdown
Collaborator

PR_Github #51355 [ run ] completed with state ABORTED. Commit: b768245

Link to invocation

@tensorrt-cicd

Copy link
Copy Markdown
Collaborator

PR_Github #51356 [ run ] completed with state SUCCESS. Commit: 7b73fda
/LLM/main/L0_MergeRequest_PR pipeline #40767 completed with status: 'FAILURE'

CI Report

⚠️ Action Required:

  • Please check the failed tests and fix your PR
  • If you cannot view the failures, ask the CI triggerer to share details
  • Once fixed, request an NVIDIA team member to trigger CI again

CI Agent Failure Analysis

Link to invocation

@Shixiaowei02
Shixiaowei02 force-pushed the user/xiaoweis/pin-nixl-cu13 branch from 7b73fda to dcf3fde Compare June 1, 2026 11:20
@Shixiaowei02

Copy link
Copy Markdown
Collaborator Author

/bot run --disable-fail-fast

1 similar comment
@Shixiaowei02

Copy link
Copy Markdown
Collaborator Author

/bot run --disable-fail-fast

@Shixiaowei02
Shixiaowei02 force-pushed the user/xiaoweis/pin-nixl-cu13 branch from dcf3fde to af5beae Compare June 2, 2026 03:02
@Shixiaowei02

Copy link
Copy Markdown
Collaborator Author

/bot run --disable-fail-fast

@tensorrt-cicd

Copy link
Copy Markdown
Collaborator

PR_Github #51485 [ run ] triggered by Bot. Commit: af5beae Link to invocation

@tensorrt-cicd

Copy link
Copy Markdown
Collaborator

PR_Github #51485 [ run ] completed with state SUCCESS. Commit: af5beae
/LLM/main/L0_MergeRequest_PR pipeline #40891 completed with status: 'SUCCESS'
Pipeline passed with automatic retried tests. Check the rerun report for details.

CI Report

Link to invocation

@Shixiaowei02

Copy link
Copy Markdown
Collaborator Author

/bot skip --comment "CI had already passed before handling the waive file conflicts."

@Shixiaowei02
Shixiaowei02 enabled auto-merge (squash) June 3, 2026 01:38
@Shixiaowei02
Shixiaowei02 marked this pull request as draft June 3, 2026 01:40
auto-merge was automatically disabled June 3, 2026 01:40

Pull request was converted to draft

@tensorrt-cicd

Copy link
Copy Markdown
Collaborator

PR_Github #51720 [ skip ] triggered by Bot. Commit: b5da4cf Link to invocation

@tensorrt-cicd

Copy link
Copy Markdown
Collaborator

PR_Github #51720 [ skip ] completed with state SUCCESS. Commit: b5da4cf
Skipping testing for commit b5da4cf

Link to invocation

@Shixiaowei02
Shixiaowei02 force-pushed the user/xiaoweis/pin-nixl-cu13 branch from b5da4cf to 7b73fda Compare June 4, 2026 05:21
@Shixiaowei02

Copy link
Copy Markdown
Collaborator Author

Moved to #14939

@Shixiaowei02
Shixiaowei02 deleted the user/xiaoweis/pin-nixl-cu13 branch June 4, 2026 05:31
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

6 participants