conclude_performance() crashes when test_iterations is 1 or 2

Open Beginner friendly
#44 0 comments 0 reactions 0 assignees View on GitHub

Nobody has claimed this yet.

Assessment

Difficulty
1/5
Estimated time
Under an hour
Newbie friendliness
78/100
Issue type
Bug
Clarity
Clearly specified
Activity status
Active
Tech stack
python

Research direction

Start at the trimming block around line 201 of ros2_benchmark/ros2_benchmark/basic_performance_calculator.py in conclude_performance(), and mirror the len(metric_value_list) > 2 guard already used in utils/nvidia_smi_profiler.py. Reproduce with a stubbed rclpy and one or two synthetic iterations through calculate_performance() then conclude_performance(); note that test/benchmark_pol.py uses TIMELINE mode and so never exercises this path. Done means 1, 2 and 3 iterations all report metrics instead of raising. Note the reporter has offered to open a PR, so check for one before starting.

Written by the indexing model from the issue text.

Description

BasicPerformanceCalculator.conclude_performance() always drops the highest and the lowest value of each metric before averaging (ros2_benchmark/ros2_benchmark/basic_performance_calculator.py, around line 201):

# Remove the best and the worst before concluding the metric
metric_value_list.remove(max(metric_value_list))
metric_value_list.remove(min(metric_value_list))

With test_iterations: 1 or 2 there is nothing left after that, so the run fails at the very end, after all iterations have finished, and no results file is written:

  • test_iterations: 1 -> ValueError: min() arg is an empty sequence
  • test_iterations: 2 -> ZeroDivisionError: division by zero (for the mean metrics)
  • test_iterations: 3 and up work

This is hit in LOOPING (the default), SWEEPING and LIVE modes, which call conclude_performance() after the iteration loop in ros2_benchmark_test.py. TIMELINE mode does not call it, which is probably why test/benchmark_pol.py (TIMELINE, test_iterations=1) does not show it.

The resource side already handles this: utils/nvidia_smi_profiler.py only trims when len(metric_value_list) > 2.

To reproduce without ROS, load the module with a stub rclpy, feed one or two iterations of synthetic timestamps to calculate_performance(), then call conclude_performance():

1 FAIL ValueError min() arg is an empty sequence
2 FAIL ZeroDivisionError division by zero
3 ok 111.11

Expected: with one or two iterations, report the metrics over those iterations, same as the resource profiler does. Using the same > 2 guard here would fix it. I can open a small PR for that.

Dominant language
Python
Stars
145
Forks
18
Avg merge
1m
Merged PRs (30d)
1

Getting set up

First steps

  1. Read the whole issue, then the project's contributing guide.
  2. Comment on the issue to say you are picking it up — it saves two people doing the same work.
  3. Fork the repository and make your change on a branch.
  4. Open a pull request that references the issue number.

More from NVIDIA-ISAAC-ROS/ros2_benchmark

All issues in NVIDIA-ISAAC-ROS/ros2_benchmark

Similar issues

More Python issues

Get new issues in your inbox

A short digest of beginner-friendly GitHub issues.