Program cannot exit after executing the last line. (due to 'from ogb.nodeproppred import PygNodePropPreDataset')

[qiaotonggg]

My code is quit simple:

import torch
import torch_geometric.transforms as T
from ogb.nodeproppred import PygNodePropPredDataset

def main():
    parser = argparse.ArgumentParser(description='OGBN-Products (GNN)')
    parser.add_argument('--device', type=int, default=1)
    args = parser.parse_args()
    print(args)
    device = f'cuda:{args.device}' if torch.cuda.is_available() else 'cpu'
    device = torch.device(device)
    print(device)
    dataset = PygNodePropPredDataset(name='ogbn-products',
                                     transform=T.ToSparseTensor())
    print('Done load.')
    data = dataset[0]
    print('Finish.')

if __name__ == "__main__":
    main()

The program cannot exit after print "Finish.", when i kill it, it will show some problems about threads

Namespace(device=1)
cuda:1
Done load.
Finish.
Exception ignored in: <module 'threading' from '/home/xxxxx/Programs/anaconda3/envs/xxxxx/lib/python3.9/threading.py'>
Traceback (most recent call last):
  File "/home/xxxxx/Programs/anaconda3/envs/profiling-cu113/lib/python3.9/threading.py", line 1477, in _shutdown
    lock.acquire()
KeyboardInterrupt:

Process finished with exit code 0

[qiaotonggg]

I'm sure the ogb is the latest(1.3.5), the system is Ubuntu 20.08.

[qiaotonggg]

When i comment from ogb.nodeproppred import PygNodePropPredDataset, it can exit with code 0 normally.

[weihua916]

Hi! Hmm I cannot reproduce this. I also used ogb 1.3.5, and I was able to run your code without any issue (Ubuntu 20.04). I am a bit confused. When you comment out from ogb.nodeproppred import PygNodePropPredDataset, how did you get PygNodePropPredDataset? Also, given that the code is excuted until Finish, I am not sure if the problem is related to ogb package. Can you double-check?

[qiaotonggg]

Thanks for your reply. I also comment out dataset = PygNodePropPredDataset(name='ogbn-products, transform=T.ToSparseTensor()), when i comment the from ogb.nodeproppred import PygNodePropPredDataset. Moreover, when i keep from ogb.nodeproppred import PygNodePropPredDataset, and only comment the dataset = PygNodePropPredDataset(name='ogbn-products, transform=T.ToSparseTensor()), i can still wittness the problem. So, i think maybe some parallel operations in ogb cause that problem. Thank you again.

[qizou97]

I had a similar problem when I tried to import PygGraphPropPredDataset from ogb.graphproppred. One way to avoid this problem is to cpoy the corresponding PygGraphPropPredDataset and and all functions and classes in ogb related to PygGraphPropPredDataset all functions and classes related to PygGraphProppredDataset cpoy to the local project, and introduce PygGraphProppredDataset from the local implementation

[qiaotonggg]

I had a similar problem when I tried to import PygGraphPropPredDataset from ogb.graphproppred. One way to avoid this problem is to cpoy the corresponding PygGraphPropPredDataset and and all functions and classes in ogb related to PygGraphPropPredDataset all functions and classes related to PygGraphProppredDataset cpoy to the local project, and introduce PygGraphProppredDataset from the local implementation

Thanks, it works anyway!

[qiaotonggg]

https://github.com/snap-stanford/ogb/issues/408#issuecomment-1499150916