Is locust considered as a benchmark tool also ?

[user1500177]

Here’s how you can properly phrase your questions for the discussion:

1. Is Locust considered a benchmark tool in addition to being a load testing tool?
2. When running Locust in CMD mode, what is the best way to store the output in a file so that the file is not overwritten on each run? Is there a way to specify just a folder path and have Locust automatically generate unique output files for each run?
3. I noticed there is a .conf file for Locust— is it preferred to use a configuration file for managing Locust settings?

[cyberw]

• What is your definition of a benchmark tool? Locust is primarily used for client-server applications - I don't think it is good fit for microbenchmarks or anything else that runs in the same process.
• I'd use your shell to generated a timestamp for the file. Something like
  locust --json --headless -t 10 > "$(date +%Y%m%d_%H%M%S).json"
• Command line arguments, env vars and .conf files are all equally supported. Whatever works best for you!

[user1500177]

Thanks for the clarification.

By benchmarking, I specifically mean macro benchmarks — measuring the performance of a running API end-to-end over HTTP.

The API I’m benchmarking is very lightweight:

• Stateless
• No database interactions
• Primarily runs an algorithm and returns a response

My current approach is:

• Start the API as a real HTTP server (e.g. Uvicorn) as a subprocess
• The subprocess is started inside Locust startup logic/hooksusing events.test_start, so no external orchestration is needed.
• Use Locust purely as an external HTTP client
• Stop the server subprocess cleanly on Locust shutdown
• Run everything in CI/CD by executing a single Locust command and exporting CSV/JSON/HTML results

The goal of this approach is to keep CI/CD integration simple: the pipeline only needs to run Locust, and Locust itself manages starting and stopping the test server.

From your response, it sounds like this aligns with Locust’s intended usage as long as it’s treated as macro benchmarking of a client-server API, rather than in-process benchmarking.

Please let me know if starting and stopping the test server as a subprocess from within Locust hooks is considered an acceptable pattern, or if there are recommended alternatives for CI usage.

Thanks!

[cyberw]

Hi! You'll have to make some effort yourself first, feel free to come back with more specific questions later!