[https://nvbugs/6463812][fix] Import MpiPoolSession from its source module (tensorrt_llm.llmapi.mpi_session)…

[trtllm-agent]

Summary

• Root cause: tests/test_common/session_reuse.py rebinds proxy.MpiPoolSession to a factory function, so isinstance(self.mpi_session, MpiPoolSession) raises TypeError.
• Fix: Import MpiPoolSession from its source module (tensorrt_llm.llmapi.mpi_session) inline at the check site so isinstance always receives the concrete class, and remove the corresponding waiver.
• Automated fix generated by repair-bot

Test plan

• Verify fix on the same GPU type as the original failure
• Check for regressions in related tests

Links

• Bug: https://nvbugs/6463812

Summary by CodeRabbit

• Bug Fixes

  • Improved executor startup reliability when running in environments with customized test configurations.

• Tests

  • Re-enabled an end-to-end text generation integration test, expanding validation coverage for encoder-decoder models.

[coderabbitai]

No actionable comments were generated in the recent review. 🎉

ℹ️ Recent review info

⚙️ Run configuration

Configuration used: Path: .coderabbit.yaml

Review profile: CHILL

Plan: Enterprise

Run ID: d9ccde4f-a8c0-4356-a0f4-5859c55073be

📥 Commits

Reviewing files that changed from the base of the PR and between b602fa6 and 01c457d.

📒 Files selected for processing (2)

• tensorrt_llm/executor/proxy.py
• tests/integration/test_lists/waives.txt

💤 Files with no reviewable changes (1)

• tests/integration/test_lists/waives.txt

---

📝 Walkthrough

Walkthrough

The executor proxy now re-imports MpiPoolSession before its worker-startup type check. The BART PyTorch encoder-decoder integration test is removed from the skipped-test waiver list.

Changes

MPI session startup and BART integration coverage

Layer / File(s)	Summary
Reload MPI session class for worker startup tensorrt_llm/executor/proxy.py	A local MpiPoolSession import makes the isinstance check use the real class when the module-level reference is monkey-patched.
Remove BART integration waiver tests/integration/test_lists/waives.txt	Removes the test_bart_pytorch_generate_encoder_decoder_end_to_end skip waiver associated with NVBUG 6463812.

Estimated code review effort: 2 (Simple) | ~10 minutes

Possibly related PRs

• NVIDIA/TensorRT-LLM#16438: Applies the same local re-import approach for MpiPoolSession in the executor proxy.
• NVIDIA/TensorRT-LLM#16490: Changes the related MPI session detection logic in the executor proxy.
• NVIDIA/TensorRT-LLM#16465: Updates the BART integration-test waiver entries.

Suggested reviewers: brnguyen2

🚥 Pre-merge checks | ✅ 4 | ❌ 1

❌ Failed checks (1 warning)

Check name	Status	Explanation	Resolution
Docstring Coverage	⚠️ Warning	Docstring coverage is 0.00% which is insufficient. The required threshold is 80.00%.	Write docstrings for the functions missing them to satisfy the coverage threshold.

✅ Passed checks (4 passed)

Check name	Status	Explanation
Title check	✅ Passed	The title is specific and matches the main fix: importing MpiPoolSession at the check site for the nvbugged TypeError.
Description check	✅ Passed	It covers the issue, fix, test plan, and bug link, though it doesn't use the template's exact Description/Test Coverage/PR Checklist headings.
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.

✨ Finishing Touches

🧪 Generate unit tests (beta)

• Create PR with unit tests

---

_{Comment @coderabbitai help to get the list of available commands.}

[BowenFu]

Heads up: this is one of three competing fixes for the same one-line session check in proxy.py — #16417, #16490, #16496 (same line, would conflict; only one should land). This re-import approach fixes the symptom but not the root fragility (it still breaks if tensorrt_llm.llmapi.mpi_session.MpiPoolSession itself is patched). #16490's polymorphic self.mpi_session.is_comm_session() is immune to that. Suggest consolidating on #16490 and closing this one, folding its nvbug un-waive into #16490.

[brnguyen2]

Closing as it's superseded by #16444