[https://nvbugs/5633340][fix] start disagg workers and servers on free ports

[reasonsolo]

Also print logs at exit for better debugging experience of flaky CI failures.

Summary by CodeRabbit

Release Notes

• Tests
  • Improved test infrastructure with dynamic port allocation and enhanced process lifecycle management for better test isolation and reliability.
  • Re-enabled previously skipped tests for disaggregated server restart scenarios.

_{✏️ Tip: You can customize this high-level summary in your review settings.}

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

/bot [-h] ['run', 'kill', 'skip', 'reuse-pipeline'] ...

Provide a user friendly way for developers to interact with a Jenkins server.

Run /bot [-h|--help] to print this help message.

See details below for each supported subcommand.

Details

run [--reuse-test (optional)pipeline-id --disable-fail-fast --skip-test --stage-list "A10-PyTorch-1, xxx" --gpu-type "A30, H100_PCIe" --test-backend "pytorch, cpp" --add-multi-gpu-test --only-multi-gpu-test --disable-multi-gpu-test --post-merge --extra-stage "H100_PCIe-TensorRT-Post-Merge-1, xxx" --detailed-log --debug(experimental)]

Launch build/test pipelines. All previously running jobs will be killed.

--reuse-test (optional)pipeline-id (OPTIONAL) : Allow the new pipeline to reuse build artifacts and skip successful test stages from a specified pipeline or the last pipeline if no pipeline-id is indicated. If the Git commit ID has changed, this option will be always ignored. The DEFAULT behavior of the bot is to reuse build artifacts and successful test results from the last pipeline.

--disable-reuse-test (OPTIONAL) : Explicitly prevent the pipeline from reusing build artifacts and skipping successful test stages from a previous pipeline. Ensure that all builds and tests are run regardless of previous successes.

--disable-fail-fast (OPTIONAL) : Disable fail fast on build/tests/infra failures.

--skip-test (OPTIONAL) : Skip all test stages, but still run build stages, package stages and sanity check stages. Note: Does NOT update GitHub check status.

--stage-list "A10-PyTorch-1, xxx" (OPTIONAL) : Only run the specified test stages. Examples: "A10-PyTorch-1, xxx". Note: Does NOT update GitHub check status.

--gpu-type "A30, H100_PCIe" (OPTIONAL) : Only run the test stages on the specified GPU types. Examples: "A30, H100_PCIe". Note: Does NOT update GitHub check status.

--test-backend "pytorch, cpp" (OPTIONAL) : Skip test stages which don't match the specified backends. Only support [pytorch, cpp, tensorrt, triton]. Examples: "pytorch, cpp" (does not run test stages with tensorrt or triton backend). Note: Does NOT update GitHub pipeline status.

--only-multi-gpu-test (OPTIONAL) : Only run the multi-GPU tests. Note: Does NOT update GitHub check status.

--disable-multi-gpu-test (OPTIONAL) : Disable the multi-GPU tests. Note: Does NOT update GitHub check status.

--add-multi-gpu-test (OPTIONAL) : Force run the multi-GPU tests in addition to running L0 pre-merge pipeline.

--post-merge (OPTIONAL) : Run the L0 post-merge pipeline instead of the ordinary L0 pre-merge pipeline.

--extra-stage "H100_PCIe-TensorRT-Post-Merge-1, xxx" (OPTIONAL) : Run the ordinary L0 pre-merge pipeline and specified test stages. Examples: --extra-stage "H100_PCIe-TensorRT-Post-Merge-1, xxx".

--detailed-log (OPTIONAL) : Enable flushing out all logs to the Jenkins console. This will significantly increase the log volume and may slow down the job.

--debug (OPTIONAL) : Experimental feature. Enable access to the CI container for debugging purpose. Note: Specify exactly one stage in the stage-list parameter to access the appropriate container environment. Note: Does NOT update GitHub check status.

For guidance on mapping tests to stage names, see docs/source/reference/ci-overview.md
and the scripts/test_to_stage_mapping.py helper.

kill

kill

Kill all running builds associated with pull request.

skip

skip --comment COMMENT

Skip testing for latest commit on pull request. --comment "Reason for skipping build/test" is required. IMPORTANT NOTE: This is dangerous since lack of user care and validation can cause top of tree to break.

reuse-pipeline

reuse-pipeline

Reuse a previous pipeline to validate current commit. This action will also kill all currently running builds associated with the pull request. IMPORTANT NOTE: This is dangerous since lack of user care and validation can cause top of tree to break.

[coderabbitai]

📝 Walkthrough

Walkthrough

Refactors test infrastructure to use dynamic port allocation via a new get_free_unused_port mechanism instead of hard-coded ports. Introduces disagg_port and work_dir fixtures, creates a ProcessWrapper class to manage subprocess lifecycle and logging, and updates all worker/server launch functions to support per-process configuration files and graceful termination with log tailing. Removes two test skip entries from the waives list.

Changes

Cohort / File(s)	Summary
Dynamic port allocation and subprocess lifecycle management tests/integration/defs/disaggregated/test_auto_scaling.py	Adds get_free_unused_port() mechanism with global USED_PORTS tracking; introduces ProcessWrapper class to encapsulate subprocess execution with logging and port metadata; adds disagg_port and work_dir fixtures for dynamic port supply and temporary working directories; refactors _run_worker(), run_ctx_worker(), run_gen_worker(), run_disagg_server() signatures to accept work_dir and use auto-assign port (0); updates service_discovery fixture to use dynamic port and disk-backed data dir; modifies verify_cluster_info() to default port to 0 with validation; reworks terminate() to tail logs and release ports; updates all test cases (test_service_discovery, test_minimal_instances, test_worker_restart, test_disagg_server_restart) to use new port/work_dir flow.
Test skip removals tests/integration/test_lists/waives.txt	Removes two SKIP entries for disaggregated/test_auto_scaling.py::test_disagg_server_restart[etcd-round_robin] and disaggregated/test_auto_scaling.py::test_disagg_server_restart[http-round_robin].

Estimated code review effort

🎯 4 (Complex) | ⏱️ ~45 minutes

Areas requiring extra attention:

• ProcessWrapper class design and lifecycle management guarantees
• Port allocation strategy and USED_PORTS state management across concurrent test execution
• Fixture dependency chain (disagg_port → work_dir → service_discovery, disagg_server_config)
• Function signature changes across _run_worker(), run_ctx_worker(), run_gen_worker(), run_disagg_server() and their integration with the new port/work_dir parameters
• Termination logic refactoring, particularly log tailing on failure and port cleanup
• Consistency of dynamic port usage across all test scenarios and helper function calls
• Configuration file path construction and per-process isolation under work_dir

Pre-merge checks and finishing touches

❌ Failed checks (2 warnings)

Check name	Status	Explanation	Resolution
Description check	⚠️ Warning	The PR description is incomplete. It mentions printing logs at exit but lacks detailed explanations of the issue, solution, test coverage, and verification steps required by the repository template.	Add a clear Description section explaining what port allocation issue was fixed and why, list specific test cases covered (particularly test_disagg_server_restart tests), and verify all checklist items are addressed.
Docstring Coverage	⚠️ Warning	Docstring coverage is 0.00% which is insufficient. The required threshold is 80.00%.	You can run @coderabbitai generate docstrings to improve docstring coverage.

✅ Passed checks (1 passed)

Check name	Status	Explanation
Title check	✅ Passed	The title clearly and specifically describes the main change: starting disagg workers and servers on free ports, with an NVBugs ticket reference and [fix] type marker following the repository template.

✨ Finishing touches

• 📝 Generate docstrings

🧪 Generate unit tests (beta)

• Create PR with unit tests
• Post copyable unit tests in a comment

---

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

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

[coderabbitai]

Actionable comments posted: 1

Caution

Some comments are outside the diff and can’t be posted inline due to platform limitations.

⚠️ Outside diff range comments (2)

tests/integration/defs/disaggregated/test_auto_scaling.py (2)

225-239: Ensure verify_cluster_info is always called with a valid port.

• Adding port as an explicit parameter and asserting port > 0 is a good guard against accidental misuse.
• However, test_disagg_server_restart currently calls verify_cluster_info(False, 1, 1, expected_code=500) without passing port, so this function fails on the assertion rather than by actually contacting the server. That weakens what the test is asserting.

Recommend updating that call to:

-        with pytest.raises(Exception):
-            verify_cluster_info(False, 1, 1, expected_code=500)
+        with pytest.raises(Exception):
+            verify_cluster_info(False, 1, 1, port=disagg_port, expected_code=500)

so the failure path is due to the server actually being down (connection error / HTTP code), not a local assertion on the argument.

---

468-506: test_disagg_server_restart flow is correct modulo verify_cluster_info port usage.

• Using the common helpers (run_*, wait_for_disagg_server_ready, request_completion, terminate) with dynamic ports and work_dir is consistent with the other tests.
• Restarting the disagg server with the same disagg_port after killing it is reasonable and exercises server resilience.

As noted earlier, the only functional concern is the verify_cluster_info(False, 1, 1, expected_code=500) call missing port=disagg_port, which should be updated so the assertion is about server response rather than argument validation. Once that is fixed, the test looks solid.

🧹 Nitpick comments (9)

tests/integration/defs/disaggregated/test_auto_scaling.py (9)

1-20: Add NVIDIA copyright header and consider namespaced import style.

• This Python file is missing the NVIDIA copyright header required for *.py files; please add the standard header with the current year at the top.
• Also consider using a namespaced import for _utils (e.g., import tensorrt_llm._utils as trt_utils and trt_utils.get_free_port()) to better align with the guideline of maintaining module namespaces for imports.

---

23-39: Port-reservation helper is reasonable; consider a custom exception and shorter message.

The USED_PORTS tracking and get_free_unused_port() loop look correct for a single-process test environment and provide an extra safeguard on top of get_free_port(), which matches prior learnings about port allocation in these tests.
For cleanliness and linter friendliness, you could improve:

-    max_attempts = 100
+    max_attempts = 100

@@
-    raise Exception(
-        f"Failed to find a free unused port after {max_attempts} attempts")
+    class PortAllocationError(RuntimeError):
+        """Raised when a free unused port cannot be allocated."""
+
+    raise PortAllocationError(
+        f"Failed to find a free unused port after {max_attempts} attempts")

---

47-71: disagg_port and work_dir fixtures look good; minor consideration for service_discovery.

• disagg_port + get_free_unused_port() and work_dir = tempfile.mkdtemp() are appropriate for isolating tests and avoiding port clashes.
• In the etcd branch of service_discovery, disagg_port is injected but unused; that's harmless but you could drop it from the signature if not needed elsewhere, or add a brief comment that it’s part of the common fixture interface.

---

125-132: ProcessWrapper is a useful abstraction; consider documenting its contract.

The wrapper cleanly encapsulates process, log_file, log_path, and port. To aid future maintainers, consider a short class docstring clarifying that it’s intended for use with terminate() and that port must correspond to an entry in USED_PORTS.

---

134-167: Worker launcher logic is sound; small robustness improvements possible.

• Dynamic port selection when port == 0 and per-worker config files in work_dir look correct.
• Logging to role-specific files and returning ProcessWrapper is a good fit for later termination & log tailing.

Consider:

-        log_path = os.path.join(work_dir, f"output_{role}.log")
+        log_path = os.path.join(work_dir, f"output_{role}_{port}.log")

to distinguish logs when multiple workers with the same role are started over time.

Also, if there’s any chance device could be None, guarding the CUDA_VISIBLE_DEVICES set with if device is not None and device != -1: would be slightly more defensive.

---

169-190: Server launcher mostly consistent, but argument naming is confusing.

The mechanics of run_disagg_server (writing the config under work_dir, spawning trtllm-serve disaggregated, and wrapping in ProcessWrapper) look fine. Two nits:

• The parameter name disagg_cluster_config is misleading here; it actually receives the full server config dict (disagg_server_config fixture). Renaming it would improve readability.
• Similar to workers, you might want to include port in the log filename to better distinguish restarts:

-    log_path = os.path.join(work_dir, "disagg_server.log")
+    log_path = os.path.join(work_dir, f"disagg_server_{port}.log")

---

242-251: tail() helper is fine for CI; minor cleanups possible.

Using the system tail to show the end of log files is practical for debugging flaky CI. Minor optional improvements:

• Call proc.wait() after reading to avoid leaving a zombie process.
• If you’d like to avoid the external dependency on tail, you could reimplement this directly in Python (e.g., seek from end), though that’s not strictly necessary.

---

321-364: test_minimal_instances logic is consistent with new helpers.

• Local initialization of worker/server variables to None is good defensive practice.
• Using dynamic ports for workers and disagg_port for the server is coherent.
• The two-phase cluster population (1 ctx/1 gen → failure, then 2 ctx/2 gen → success) still reads correctly under the new abstractions.

The use of pytest.raises(Exception) is broad but acceptable in these integration tests; tightening the expected exception type could be a future improvement.

---

372-461: test_worker_restart scenarios are correct; note comment drift and rely on fixed terminate.

• The restart scenario (kill gen, add another gen; kill ctx, add another ctx; then add two more workers) is wired correctly with run_*_worker, wait_for_worker_ready, and verify_cluster_info(..., port=disagg_port).
• Using port=0 when adding replacement workers and relying on dynamic allocation is a good change; however, the comment “restart ctx1 and gen1 with the same ports” is now inaccurate since ports are auto-assigned. Consider updating the comment to reflect the behavior (e.g., “restart ctx and gen workers with new ports, ending up with 2 ctx and 2 gen workers”).
• Once terminate is made idempotent as suggested earlier, the mid-test terminate(gen_worker1) and terminate(ctx_worker1) plus final terminate(...) in finally will behave correctly.

📜 Review details

Configuration used: Path: .coderabbit.yaml

Review profile: CHILL

Plan: Pro

📥 Commits

Reviewing files that changed from the base of the PR and between f9aa86d and e3b3671.

📒 Files selected for processing (2)

• tests/integration/defs/disaggregated/test_auto_scaling.py (14 hunks)
• tests/integration/test_lists/waives.txt (0 hunks)

💤 Files with no reviewable changes (1)

• tests/integration/test_lists/waives.txt

🧰 Additional context used

📓 Path-based instructions (2)

**/*.py

📄 CodeRabbit inference engine (CODING_GUIDELINES.md)

**/*.py: The code developed for TensorRT-LLM should conform to Python 3.8+
Indent Python code with 4 spaces; do not use tabs
Always maintain the namespace when importing in Python, even if only one class or function from a module is used (e.g., use from package.subpackage import foo and then foo.SomeClass() instead of from package.subpackage.foo import SomeClass)
Python filenames should use snake_case (e.g., some_file.py)
Python class names should use PascalCase (e.g., class SomeClass)
Python function and method names should use snake_case (e.g., def my_awesome_function():)
Python local variable names should use snake_case, with prefix k for variable names that start with a number (e.g., k_99th_percentile = ...)
Python global variables should use upper snake_case with prefix G (e.g., G_MY_GLOBAL = ...)
Python constants should use upper snake_case (e.g., MY_CONSTANT = ...)
Avoid shadowing variables declared in an outer scope in Python
Initialize all externally visible members of a Python class in the constructor
For Python interfaces that may be used outside a file, prefer docstrings over comments
Python comments should be reserved for code within a function, or interfaces that are local to a file
Use Google style docstrings for Python classes and functions, which can be parsed by Sphinx
Python attributes and variables can be documented inline with type and description (e.g., self.x = 5 followed by """<type>: Description of 'x'""" )
Avoid using reflection in Python when functionality can be easily achieved without reflection
When using try-except blocks in Python, limit the except clause to the smallest set of specific errors possible instead of catching all exceptions
When using try-except blocks in Python to handle multiple possible variable types (duck-typing), keep the body of the try as small as possible and use the else block to implement the logic

Files:

• tests/integration/defs/disaggregated/test_auto_scaling.py

**/*.{cpp,h,cu,py}

📄 CodeRabbit inference engine (CODING_GUIDELINES.md)

All TensorRT-LLM Open Source Software code files should contain an NVIDIA copyright header that includes the current year at the top

Files:

• tests/integration/defs/disaggregated/test_auto_scaling.py

🧠 Learnings (2)

📓 Common learnings

Learnt from: Shixiaowei02
Repo: NVIDIA/TensorRT-LLM PR: 9582
File: tests/integration/defs/accuracy/test_disaggregated_serving.py:73-83
Timestamp: 2025-12-02T03:40:40.572Z
Learning: In the disaggregated serving tests (tests/integration/defs/accuracy/test_disaggregated_serving.py), calling get_free_port() multiple times in succession is acceptable because the tests run in a controlled single-process environment where race conditions for port allocation are not a concern.

📚 Learning: 2025-12-02T03:40:40.572Z

Learnt from: Shixiaowei02
Repo: NVIDIA/TensorRT-LLM PR: 9582
File: tests/integration/defs/accuracy/test_disaggregated_serving.py:73-83
Timestamp: 2025-12-02T03:40:40.572Z
Learning: In the disaggregated serving tests (tests/integration/defs/accuracy/test_disaggregated_serving.py), calling get_free_port() multiple times in succession is acceptable because the tests run in a controlled single-process environment where race conditions for port allocation are not a concern.

Applied to files:

• tests/integration/defs/disaggregated/test_auto_scaling.py

🪛 Ruff (0.14.7)

tests/integration/defs/disaggregated/test_auto_scaling.py

38-39: Create your own exception

(TRY002)

---

38-39: Avoid specifying long messages outside the exception class

(TRY003)

---

160-160: subprocess call: check for execution of untrusted input

(S603)

---

188-188: subprocess call: check for execution of untrusted input

(S603)

---

244-244: subprocess call: check for execution of untrusted input

(S603)

---

244-244: Starting a process with a partial executable path

(S607)

---

247-247: Do not catch blind exception: Exception

(BLE001)

---

261-261: Do not catch blind exception: Exception

(BLE001)

---

274-274: Do not catch blind exception: Exception

(BLE001)

---

293-293: Unused function argument: router

(ARG001)

---

293-293: Unused function argument: service_discovery

(ARG001)

---

322-322: Unused function argument: router

(ARG001)

---

322-322: Unused function argument: service_discovery

(ARG001)

---

347-347: Do not assert blind exception: Exception

(B017)

---

372-372: Unused function argument: router

(ARG001)

---

372-372: Unused function argument: service_discovery

(ARG001)

---

401-401: Do not assert blind exception: Exception

(B017)

---

427-427: Do not assert blind exception: Exception

(B017)

---

469-469: Unused function argument: router

(ARG001)

---

469-469: Unused function argument: service_discovery

(ARG001)

⏰ Context from checks skipped due to timeout of 90000ms. You can increase the timeout in your CodeRabbit configuration to a maximum of 15 minutes (900000ms). (1)

• GitHub Check: Pre-commit Check

🔇 Additional comments (3)

tests/integration/defs/disaggregated/test_auto_scaling.py (3)

90-106: disagg_server_config wiring is consistent with dynamic ports.

Using disagg_port here to set the server port and passing disagg_cluster_config through looks correct and consistent with later calls to run_disagg_server. Nothing blocking.

---

278-285: request_completion helper is straightforward and correct.

The OpenAI client construction with a dynamic base_url and explicit model, prompt, and minimal generation args looks good for these tests. No issues.

---

292-312: test_service_discovery wiring with dynamic ports and ProcessWrapper looks good.

• Workers and server are started with the new helpers and work_dir, using dynamic ports.
• wait_for_disagg_server_ready(disagg_port) + verify_cluster_info(True, 1, 1, port=disagg_port) + request_completion(..., port=disagg_port) is consistent.
• terminate(ctx_worker1, gen_worker1, disagg_server) correctly funnels teardown through the new helper (subject to the terminate fix above).

[reasonsolo]

/bot run

[tensorrt-cicd]

PR_Github #26916 [ run ] triggered by Bot. Commit: 7168756

[tensorrt-cicd]

PR_Github #26916 [ run ] completed with state ABORTED. Commit: 7168756
/LLM/main/L0_MergeRequest_PR pipeline #20508 completed with status: 'FAILURE'

[reasonsolo]

/bot run --disable-fail-fast

[tensorrt-cicd]

PR_Github #26953 [ run ] triggered by Bot. Commit: 7168756

[tensorrt-cicd]

PR_Github #26953 [ run ] completed with state SUCCESS. Commit: 7168756
/LLM/main/L0_MergeRequest_PR pipeline #20539 completed with status: 'FAILURE'

[pcastonguay]

/bot run --disable-fail-fast

[tensorrt-cicd]

PR_Github #27002 [ run ] triggered by Bot. Commit: 7168756

[tensorrt-cicd]

PR_Github #27002 [ run ] completed with state SUCCESS. Commit: 7168756
/LLM/main/L0_MergeRequest_PR pipeline #20589 completed with status: 'SUCCESS'