[https://nvbugs/6462303][fix] Import MpiPoolSession under a second private alias (_RealMpiPoolSession)…

[trtllm-agent]

Summary

• Root cause: isinstance(self.mpi_session, MpiPoolSession) used the module-level MpiPoolSession name, which the test-suite session-reuse infra monkey-patches to a factory function; isinstance() against a non-class raises TypeError.
• Fix: Import MpiPoolSession under a second private alias (_RealMpiPoolSession) that the test infra never patches, and use it for the isinstance check; construction still uses the patched name so pool reuse is preserved. Waiver removed.
• 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/6462303

Summary by CodeRabbit

• Bug Fixes
  • Improved executor worker initialization to remain reliable when runtime session components are replaced during testing.
  • Removed an outdated integration-test waiver, allowing the cache eviction and block reuse scenario to run normally.

[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: 56c09db4-31ca-4a73-b037-4194baf78e4a

📥 Commits

Reviewing files that changed from the base of the PR and between b602fa6 and 081d499.

📒 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 worker startup path now checks against the original MpiPoolSession class despite module-level monkey-patching. The waiver for the KV-cache block-reuse eviction test is removed.

Changes

MPI worker initialization

Layer / File(s)	Summary
Preserve real MPI session type and update startup detection tensorrt_llm/executor/proxy.py, tests/integration/test_lists/waives.txt	Aliases the original MpiPoolSession for the _start_executor_workers type check while retaining the runtime-facing import, and removes the KV-cache eviction test waiver.

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

Possibly related PRs

• NVIDIA/TensorRT-LLM#16496: Updates the same worker-startup MpiPoolSession check to use the real class under test monkey-patching.
• NVIDIA/TensorRT-LLM#16489: Modifies the same worker-status handling and waiver-related behavior.
• NVIDIA/TensorRT-LLM#16419: Changes the same session detection path and updates the integration waiver list.

Suggested reviewers: cascade812

🚥 Pre-merge checks | ✅ 5

✅ Passed checks (5 passed)

Check name	Status	Explanation
Title check	✅ Passed	The title is specific and matches the main fix: aliasing MpiPoolSession to avoid monkey-patched isinstance() failures.
Description check	✅ Passed	The description covers the root cause, fix, tests, and bug link, with only minor template differences from the repo format.
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.

✨ Finishing Touches

🧪 Generate unit tests (beta)

• Create PR with unit tests

---

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

[brnguyen2]

Closing this as it's superseded by #16444