generate_data_parallel.py hangs with pool.apply_async(), but not when called in the main thread.

[djl11]

By naively following the installation instructions, I'm finding that this line hangs when running with multiprocessing, but not when running in the main thread.

Would it be possible to list the full conda environment you used, by running conda list -e > requirements.txt?
For example, I would be interested to know which versions of python, Open3D and multiprocessing you used.

Any help appreciated :)

[Steve-Tod]

Hi, I add the conda environment description here. BTW, I suggest debugging the data generation process with --num-proc 1 argument to set it to single process.

[djl11]

Thanks for sending over! Unfortunately using your conda environment didn't help, and setting --num-proc 1 still causes it to hang. The only way I can get it to run is remove mp.Pool() entirely and call main() explicitly. I'll leave this issue open and further investigate, and then close with the solution once I work out what's going on.

[Steve-Tod]

Sorry about that. What script are you running? generate_data_parallel.py or construct_dataset_parallel.py?

[djl11]

generate_data_parallel.py. Also, sorry setting to --num-proc to 1 does indeed work for debugging. I mistakenly thought I had added the if args.num_proc > 1 check myself.

still might just be something silly on my end, will post my findings either way!

[djl11]

Multiprocessing issues with Open3D seem relatively common: a, b, c.

I took a look at how Open3D apply parallelism for batch-processing in their own examples, and they use joblib.

Replacing lines 195-202 with:

from joblib import Parallel, delayed
Parallel(n_jobs=args.num_proc)(delayed(main)(args, i) for i in range(args.num_proc))

Fixed the problem for me. Hopefully this helps others who may get stuck with similar issues.

[Andyyoung0507]

Hi, I also try python scripts/generate_data_parallel.py --scene pile --object-set pile/train --num-grasps 4000000 --num-proc 40 --save-scene ./data/pile/data_pile_train_random_raw_4M to generate the data, but the terminal show nothing but pybullet build time:……, I also set --num-proc to 1 to debug, but it didn't work. Then I use vscode to debug, find that is maybe caused by multiprocess. I tried @djl11 's solution, but nothing changed. So i reinstall the conda env, The script runs, but gets stuck somewhere after generating some data. May I have your help please?
Thanks for help. @djl11 @Steve-Tod