[Anai-Guo]

Problem

deepspeed/nvme/perf_generate_param.py and deepspeed/nvme/perf_run_sweep.py each define a main() that parses arguments and hands them to the worker function, and a __main__ block that skips main() and calls the worker directly with no arguments:

def generate_main(log_dir):
    ...

def main():
    args = parse_arguments()
    if not validate_args(args):
        quit()
    print(f'Generate DeepNVMe configuration from {args.log_dir} logs')
    generate_main(args.log_dir)      # <-- correct

if __name__ == "__main__":
    generate_main()                  # <-- no arguments

def sweep_main(args):
    ...

def main():
    args = parse_sweep_arguments()
    print(f"Running DeepNVMe performance sweep on {args.nvme_dir}")
    sweep_main(args)                 # <-- correct

if __name__ == "__main__":
    sweep_main()                     # <-- no arguments

So running either module directly fails immediately:

$ python -m deepspeed.nvme.perf_generate_param
TypeError: generate_main() missing 1 required positional argument: 'log_dir'

$ python -m deepspeed.nvme.perf_run_sweep
TypeError: sweep_main() missing 1 required positional argument: 'args'

and both main() functions — the ones that actually do the argument parsing and validation — are referenced nowhere else in the tree, so they are dead code today.

Sibling baseline

deepspeed/nvme/parse_nvme_stats.py, in this same package, already does it the right way:

if __name__ == "__main__":
    main()

module	__main__ calls	runs?
parse_nvme_stats.py	main()	✅
perf_generate_param.py	generate_main()	❌ — this PR
perf_run_sweep.py	sweep_main()	❌ — this PR

Scope

bin/ds_nvme_tune is not affected and is not touched: it does its own parsing and already passes the arguments through (sweep_main(args) / generate_main(args.log_dir)). The public deepspeed.nvme exports (sweep_main, generate_main, parse_sweep_arguments, ds_io_main) are unchanged — this only fixes direct module invocation.

Verification

I replayed the argument binding straight out of the AST rather than hand-copying a repro, so it cannot drift from the source: each worker's parameter list and every call site's shape are read from the files themselves, a signature-identical stub is generated, and the binding is replayed.

perf_generate_param.py: generate_main(log_dir)
   line 97   1 arg(s) -> BINDS OK          # inside main()
   line 101  0 arg(s) -> TypeError: generate_main() missing 1 required
                         positional argument: 'log_dir'    # the __main__ block

perf_run_sweep.py: sweep_main(args)
   line 315  1 arg(s) -> BINDS OK          # inside main()
   line 319  0 arg(s) -> TypeError: sweep_main() missing 1 required
                         positional argument: 'args'       # the __main__ block

After the change both __main__ blocks route through main(), which is the call site that already binds correctly.

---

🤖 Generated with Claude Code

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 4f209e117f

ℹ️ About Codex in GitHub

Codex has been enabled to automatically review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

When you sign up for Codex through ChatGPT, Codex can also answer questions or update the PR, like "@codex address that feedback".

[chatgpt-codex-connector]

Add the required sign-off trailer

This non-merge commit has no Signed-off-by trailer, violating the repository's mandatory commit policy and potentially failing DCO enforcement; recreate the commit with --signoff before merging.

AGENTS.md reference: AGENTS.md:L8-L8

Useful? React with 👍 / 👎.