[https://nvbugs/5552836][fix] Add flag TLLM_SPAWN_EXTRA_MAIN_PROCESS to disable spawning main process
#8640
Conversation
Signed-off-by: Superjomn <[email protected]>
|
/bot run |
📝 WalkthroughWalkthroughIntroduces a new toggle Changes
Sequence Diagram(s)sequenceDiagram
participant User as User/Launcher
participant MainScript as trtllm-llmapi-launch
participant BGProc as Background Process
participant MainTask as Main Task
participant WorkerNode as mgmn_worker_node
participant MPI as MPI Server
User->>MainScript: Invoke with spawn_extra_main_process=1 (or 0)
alt spawn_extra_main_process=1 (spawn enabled)
MainScript->>MainScript: Clean MPI environment variables
MainScript->>MainScript: Prepare IPC address/keys
MainScript->>BGProc: Spawn background process (rank 0)
par Background execution
BGProc->>MainTask: Execute main task
MainTask-->>BGProc: Return exit code
and Parallel worker execution (rank > 0)
MainScript->>WorkerNode: Execute mgmn_worker_node
WorkerNode-->>MainScript: Return exit code
end
MainScript->>MPI: Stop MGMN leader node
MPI-->>MainScript: Return exit code
MainScript-->>User: Propagate final exit status
else spawn_extra_main_process=0 (spawn disabled)
MainScript->>MainTask: Execute main task (in-process)
MainTask-->>MainScript: Return exit code
alt Rank > 0
MainScript->>WorkerNode: Execute mgmn_worker_node
WorkerNode-->>MainScript: Return exit code
end
MainScript-->>User: Propagate exit status
end
Estimated code review effort🎯 4 (Complex) | ⏱️ ~45 minutes The primary complexity stems from the launch script's new control flow logic with two distinct execution paths, MPI environment variable management, and exit-status propagation semantics. The test changes introduce parametrization and subprocess configuration adjustments. While the overall scope is focused (3 files), the density of logic changes in the main script and the need to verify control-flow correctness across both execution modes warrant careful review. Pre-merge checks and finishing touches❌ Failed checks (1 warning)
✅ Passed checks (2 passed)
✨ Finishing touches
🧪 Generate unit tests (beta)
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. Comment |
![coderabbitai[bot]](https://avatars.githubusercontent.com/in/347564?s=60&v=4){kind=small}
There was a problem hiding this comment.
Actionable comments posted: 0
Caution
Some comments are outside the diff and can’t be posted inline due to platform limitations.
⚠️ Outside diff range comments (1)
tensorrt_llm/llmapi/trtllm-llmapi-launch (1)
31-44: Use python3 consistently and avoid hard dependency on openssl
- The free‑port probe uses “python”, while the rest uses “python3”. On images without “python”, this fails.
- HMAC key generation relies on “openssl”; not always present in minimal containers.
Apply:
- local free_port=$(python -c 'import socket; s=socket.socket(); + local free_port=$(python3 -c 'import socket; s=socket.socket(); port = 10012 while True: try: s.bind(("", port)) break except OSError: port += 1 print(port); s.close()') export TLLM_SPAWN_PROXY_PROCESS_IPC_ADDR="tcp://127.0.0.1:${free_port}" log_stderr "TLLM_SPAWN_PROXY_PROCESS_IPC_ADDR: $TLLM_SPAWN_PROXY_PROCESS_IPC_ADDR" - export TLLM_SPAWN_PROXY_PROCESS_IPC_HMAC_KEY=$(openssl rand -hex 32) + if command -v openssl >/dev/null 2>&1; then + export TLLM_SPAWN_PROXY_PROCESS_IPC_HMAC_KEY="$(openssl rand -hex 32)" + else + export TLLM_SPAWN_PROXY_PROCESS_IPC_HMAC_KEY="$(python3 -c 'import secrets; print(secrets.token_hex(32))')" + fi
🧹 Nitpick comments (8)
examples/llm-api/llm_mgmn_llm_distributed.sh (1)
38-41: Clarify default and when to use this toggleSuggest noting the default is 1 (spawn enabled) and that setting to 0 trades throughput for stability; also mention it’s consumed by trtllm-llmapi-launch. Example:
“TLLM_SPAWN_EXTRA_MAIN_PROCESS=0 disables extra main-process spawning (default is 1). Use 0 only when debugging or if spawning is unstable; keep 1 for high‑throughput streaming.”
tensorrt_llm/llmapi/trtllm-llmapi-launch (5)
8-9: Harden boolean parsing for TLLM_SPAWN_EXTRA_MAIN_PROCESSCurrent test uses -eq 1; non‑numeric values (e.g., “true”) will cause “integer expression expected” and abort under set -e. Normalize common truthy/falsey values and default safely.
Apply:
-# Whether to spawn a additional process for the main process, it will optimize -# the performance of the main process. -spawn_extra_main_process=${TLLM_SPAWN_EXTRA_MAIN_PROCESS:-1} +# Whether to spawn an additional process for the main process (default: 1). +# Accept 1/0/true/false/yes/no/on/off (case-insensitive). +normalize_bool() { + local v="${1:-1}" + v="${v,,}" + case "$v" in + 1|true|yes|on) echo 1 ;; + 0|false|no|off) echo 0 ;; + *) log_stderr "Invalid TLLM_SPAWN_EXTRA_MAIN_PROCESS='$1', defaulting to 1"; echo 1 ;; + esac +} +spawn_extra_main_process="$(normalize_bool "${TLLM_SPAWN_EXTRA_MAIN_PROCESS:-1}")"
100-118: Graceful termination: forward signals to the background subshellIf the job is canceled (SIGTERM/SIGINT), ensure the background task process is also terminated. Add a trap after capturing subshell_pid.
Apply:
subshell_pid=$! log_stderr "Rank${mpi_rank} Subshell PID: $subshell_pid" + # Ensure background subshell is terminated on INT/TERM + trap 'log_stderr "Rank${mpi_rank} received signal; stopping subshell $subshell_pid"; kill -TERM "$subshell_pid" 2>/dev/null || true' INT TERM + log_stderr "Rank${mpi_rank} run mgmn leader node with mpi_world_size: $(mpi_world_size) ..." log_stderr "Rank0 host: $HOSTNAME" python3 -m tensorrt_llm.llmapi.mgmn_leader_nodeI can extend this to capture and terminate any child PIDs from the leader if needed.
Also applies to: 119-124
139-151: Mirror exit logging/propagation in the no‑spawn path (rank 0)For consistency and easier debugging, log and propagate the task’s exit code explicitly.
Apply:
- if [ -z "$mpi_rank" ] || [ "$mpi_rank" -eq 0 ]; then - "${task_with_command[@]}" + if [ -z "$mpi_rank" ] || [ "$mpi_rank" -eq 0 ]; then + set +e + "${task_with_command[@]}" + task_exit_code=$? + log_stderr "Rank${mpi_rank} Task exit code: $task_exit_code" + exit $task_exit_code
10-11: Minor: remove unused native_mpi_ranknative_mpi_rank is assigned but not used.
-native_mpi_rank=$OMPI_COMM_WORLD_RANK
107-109: HOSTNAME may be unset in minimal containersPrefer $(hostname) to avoid empty output.
- log_stderr "Rank0 host: $HOSTNAME" + log_stderr "Rank0 host: $(hostname)"tests/unittest/llmapi/_run_remote_mpi_session.sh (1)
8-8: LGTM; consider echoing the effective branch for assertionsPrinting the numeric toggle is helpful. Optionally add a second line like “mode: spawn” vs “mode: no‑spawn” to assert correct path in tests.
tests/unittest/llmapi/test_mpi_session.py (1)
58-63: Assert the selected mode from process output for stronger coverageYou already print TLLM_SPAWN_EXTRA_MAIN_PROCESS in the helper script. Capture one line (e.g., buffer stderr) and assert it matches the parameter to ensure both code paths are exercised, not just invoked.
Example:
- with Popen(command, + with Popen(command, env=envs, stdout=PIPE, stderr=PIPE, bufsize=1, start_new_session=True, universal_newlines=True, cwd=os.path.dirname(os.path.abspath(__file__))) as process: - # Function to read from a stream and write to output + captured = [] + # Function to read from a stream and write to output def read_stream(stream, output_stream): for line in stream: output_stream.write(line) output_stream.flush() + captured.append(line) ... return_code = process.wait() ... if return_code != 0: raise subprocess.CalledProcessError(return_code, command) + # Basic check that the selected mode was honored + expected = f"TLLM_SPAWN_EXTRA_MAIN_PROCESS: {'1' if spawn_extra_main_process else '0'}" + assert any(expected in ln for ln in captured), f"Missing toggle line: {expected}"Also applies to: 70-73, 73-81
📜 Review details
Configuration used: Path: .coderabbit.yaml
Review profile: CHILL
Plan: Pro
📒 Files selected for processing (4)
examples/llm-api/llm_mgmn_llm_distributed.sh(1 hunks)tensorrt_llm/llmapi/trtllm-llmapi-launch(2 hunks)tests/unittest/llmapi/_run_remote_mpi_session.sh(1 hunks)tests/unittest/llmapi/test_mpi_session.py(1 hunks)
🧰 Additional context used
📓 Path-based instructions (3)
**/*.{h,hpp,hh,hxx,cpp,cxx,cc,cu,cuh,py}
📄 CodeRabbit inference engine (CODING_GUIDELINES.md)
Use only spaces, no tabs; indent with 4 spaces.
Files:
tests/unittest/llmapi/test_mpi_session.py
**/*.py
📄 CodeRabbit inference engine (CODING_GUIDELINES.md)
**/*.py: Python code must target Python 3.8+.
Indent Python code with 4 spaces; do not use tabs.
Maintain module namespace when importing; prefer 'from package.subpackage import foo' then 'foo.SomeClass()' instead of importing the class directly.
Python filenames should be snake_case (e.g., some_file.py).
Python classes use PascalCase names.
Functions and methods use snake_case names.
Local variables use snake_case; prefix 'k' for variables that start with a number (e.g., k_99th_percentile).
Global variables use upper SNAKE_CASE prefixed with 'G' (e.g., G_MY_GLOBAL).
Constants use upper SNAKE_CASE (e.g., MY_CONSTANT).
Avoid shadowing variables from an outer scope.
Initialize all externally visible members of a class in the constructor.
Prefer docstrings for interfaces that may be used outside a file; comments for in-function or file-local interfaces.
Use Google-style docstrings for classes and functions (Sphinx-parsable).
Document attributes and variables inline so they render under the class/function docstring.
Avoid reflection when a simpler, explicit approach suffices (e.g., avoid dict(**locals()) patterns).
In try/except, catch the most specific exceptions possible.
For duck-typing try/except, keep the try body minimal and use else for the main logic.
Files:
tests/unittest/llmapi/test_mpi_session.py
**/*.{cpp,cxx,cc,h,hpp,hh,hxx,cu,cuh,py}
📄 CodeRabbit inference engine (CODING_GUIDELINES.md)
Prepend the NVIDIA Apache-2.0 copyright header with current year to the top of all source files (e.g., .cpp, .h, .cu, .py).
Files:
tests/unittest/llmapi/test_mpi_session.py
🧬 Code graph analysis (1)
tests/unittest/llmapi/test_mpi_session.py (2)
tests/integration/defs/trt_test_alternative.py (1)
Popen(153-168)tests/unittest/llmapi/apps/_test_disagg_serving_multi_nodes.py (1)
env(61-68)
🪛 Ruff (0.14.1)
tests/unittest/llmapi/test_mpi_session.py
73-73: subprocess call: check for execution of untrusted input
(S603)
⏰ 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)
tensorrt_llm/llmapi/trtllm-llmapi-launch (2)
156-158: Env var casing consistency for tllm_mpi_sizeYou export tllm_mpi_size (lowercase). If downstream expects uppercase naming (typical for env vars), consider TLLM_MPI_SIZE or confirm current usage.
Please confirm the consumer name; if uppercase is expected, adjust:
-export tllm_mpi_size=$(mpi_world_size) -log_stderr "tllm_mpi_size: $tllm_mpi_size" +export TLLM_MPI_SIZE="$(mpi_world_size)" +log_stderr "TLLM_MPI_SIZE: $TLLM_MPI_SIZE"
63-77: Verify whether KMP_ inclusion in blacklist is intentional or overly broad.The review comment's concern is valid. Test file
_test_disagg_serving_multi_nodes.py(line 67) strips onlyPMI_,OMPI_,PMIX_,SLURM_, while bothtrtllm-llmapi-launchandmpi_session.pyincludeKMP_in their broader blacklist. Since KMP_ controls OpenMP/MKL threading (not MPI initialization), and the launch script's documented purpose is to "clean the MPI environment" before spawning child processes, the inclusion of KMP_ warrants review to confirm it's intentional and not unintended side-effect cleanup.tests/unittest/llmapi/test_mpi_session.py (1)
1-7: Security hint (Ruff S603) is OK hereInputs are controlled: task_type is a Literal (“submit”|“submit_sync”) and the script path is computed locally. Keeping shell=False and arg list avoids injection.
If desired, run with a third, invalid task_type to confirm failure is clean and not shell‑interpreted.
Also applies to: 73-81
|
PR_Github #22398 [ run ] triggered by Bot. Commit: |
|
/bot run |
|
PR_Github #22409 [ run ] triggered by Bot. Commit: |
|
PR_Github #22398 [ run ] completed with state |
|
PR_Github #22409 [ run ] completed with state |
Superjomn
force-pushed
the
fix.turn-off-spawn-main-process
branch
from
f245ee6 to
1bad84f
Compare
|
/bot run |
|
PR_Github #22431 [ run ] triggered by Bot. Commit: |
Signed-off-by: Superjomn <[email protected]>
Superjomn
force-pushed
the
fix.turn-off-spawn-main-process
branch
from
1bad84f to
c82731d
Compare
Superjomn
changed the title
|
/bot run |
Superjomn
changed the title
TLLM_SPAWN_EXTRA_MAIN_PROCESS to disable spawning main process
|
PR_Github #22436 [ run ] triggered by Bot. Commit: |
|
PR_Github #22431 [ run ] completed with state |
|
PR_Github #22436 [ run ] completed with state |
|
/bot run |
|
PR_Github #22499 [ run ] triggered by Bot. Commit: |
|
PR_Github #22499 [ run ] completed with state |
|
/bot run |
|
PR_Github #22517 [ run ] triggered by Bot. Commit: |
|
PR_Github #22517 [ run ] completed with state |
|
|
||
| export_free_tcp_addr_for_spawn_proxy_process | ||
|
|
||
| if [ -z "$mpi_rank" ] || [ "$mpi_rank" -eq 0 ]; then |
There was a problem hiding this comment.
I assume the following parts are directly moved to run_with_spawn_extra_main_process without changes?
4 participants
Summary by CodeRabbit
New Features
Documentation
Tests
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
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.
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 thestage-listparameter 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.mdand the
scripts/test_to_stage_mapping.pyhelper.kill
killKill all running builds associated with pull request.
skip
skip --comment COMMENTSkip 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-pipelineReuse 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.