pytorch-lightning: DDP is not working with Pytorch Lightning

I am using DDP in a single machine with 2 GPUs. when I am running the code it stuck forever with the below script. The code is working properly with dp and also with ddp using a single GPU.

GPU available: True, used: True TPU available: False, using: 0 TPU cores LOCAL_RANK: 0 - CUDA_VISIBLE_DEVICES: [0,1] Using native 16bit precision. LOCAL_RANK: 1 - CUDA_VISIBLE_DEVICES: [0,1] Using native 16bit precision. initializing ddp: GLOBAL_RANK: 1, MEMBER: 2/2 initializing ddp: GLOBAL_RANK: 0, MEMBER: 1/2

To reproduce I used the boring code which exactly produce the above script:

import os
import torch
from torch.utils.data import DataLoader, Dataset

from pytorch_lightning import LightningModule, Trainer


class RandomDataset(Dataset):
    def __init__(self, size, length):
        self.len = length
        self.data = torch.randn(length, size)

    def __getitem__(self, index):
        return self.data[index]

    def __len__(self):
        return self.len


class BoringModel(LightningModule):
    def __init__(self):
        super().__init__()
        self.layer = torch.nn.Linear(32, 2)

    def forward(self, x):
        return self.layer(x)

    def training_step(self, batch, batch_idx):
        loss = self(batch).sum()
        self.log("train_loss", loss)
        return {"loss": loss}

    def validation_step(self, batch, batch_idx):
        loss = self(batch).sum()
        self.log("valid_loss", loss)

    def test_step(self, batch, batch_idx):
        loss = self(batch).sum()
        self.log("test_loss", loss)

    def configure_optimizers(self):
        return torch.optim.SGD(self.layer.parameters(), lr=0.1)


def run():
    train_data = DataLoader(RandomDataset(32, 64), batch_size=2)
    val_data = DataLoader(RandomDataset(32, 64), batch_size=2)
    # test_data = DataLoader(RandomDataset(32, 64), batch_size=2)

    model = BoringModel()
    trainer = Trainer(
        default_root_dir=os.getcwd(),
        limit_train_batches=1,
        limit_val_batches=1,
        num_sanity_val_steps=0,
        max_epochs=1,
        gpus=2,
        distributed_backend='ddp',
    )
    trainer.fit(model, train_data, val_data)
    # trainer.test(model,test_data)

if __name__ == "__main__":
    run()

I ran the code with NCCL_DEBUG=INFO and raised with errors below:

u116642:82796:82796 [0] NCCL INFO Bootstrap : Using enp68s0:35.9.130.234<0>
u116642:82796:82796 [0] NCCL INFO NET/Plugin : No plugin found (libnccl-net.so), using internal
implementation
u116642:82796:82796 [0] misc/ibvwrap.cc:63 NCCL WARN Failed to open libibverbs.so[.1]
u116642:82796:82796 [0] NCCL INFO NET/Socket : Using [0]enp68s0:35.9.130.234<0>
[1]veth07bbc25:fe80::5897:33ff:fe5e:15ee%veth07bbc25<0>
[2]veth2d1e326:fe80::54c0:1cff:fe70:8b39%veth2d1e326<0>
u116642:82796:82796 [0] NCCL INFO Using network Socket
NCCL version 2.10.3+cuda11.3
u116642:82845:82845 [1] NCCL INFO Bootstrap : Using enp68s0:35.9.130.234<0>
u116642:82845:82845 [1] NCCL INFO NET/Plugin : No plugin found (libnccl-net.so), using internal
implementation
u116642:82845:82845 [1] misc/ibvwrap.cc:63 NCCL WARN Failed to open libibverbs.so[.1]
u116642:82845:82845 [1] NCCL INFO NET/Socket : Using [0]enp68s0:35.9.130.234<0>
[1]veth07bbc25:fe80::5897:33ff:fe5e:15ee%veth07bbc25<0>
[2]veth2d1e326:fe80::54c0:1cff:fe70:8b39%veth2d1e326<0>
u116642:82845:82845 [1] NCCL INFO Using network Socket
u116642:82796:82908 [0] NCCL INFO Channel 00/04 : 0 1
u116642:82845:82930 [1] NCCL INFO Trees [0] -1/-1/-1->1->0 [1] 0/-1/-1->1->-1 [2] -1/-1/-1->1->0 [3]
0/-1/-1->1->-1
u116642:82796:82908 [0] NCCL INFO Channel 01/04 : 0 1
u116642:82796:82908 [0] NCCL INFO Channel 02/04 : 0 1
u116642:82796:82908 [0] NCCL INFO Channel 03/04 : 0 1
u116642:82796:82908 [0] NCCL INFO Trees [0] 1/-1/-1->0->-1 [1] -1/-1/-1->0->1 [2] 1/-1/-1->0->-1 [3]
-1/-1/-1->0->1
u116642:82796:82908 [0] NCCL INFO Setting affinity for GPU 0 to ffff,ffffffff
u116642:82796:82908 [0] NCCL INFO Channel 00 : 0[1000] -> 1[21000] via P2P/IPC
u116642:82796:82908 [0] NCCL INFO Channel 01 : 0[1000] -> 1[21000] via P2P/IPC
u116642:82796:82908 [0] NCCL INFO Channel 02 : 0[1000] -> 1[21000] via P2P/IPC
u116642:82796:82908 [0] NCCL INFO Channel 03 : 0[1000] -> 1[21000] via P2P/IPC
u116642:82845:82930 [1] NCCL INFO Channel 00 : 1[21000] -> 0[1000] via P2P/IPC
u116642:82845:82930 [1] NCCL INFO Channel 01 : 1[21000] -> 0[1000] via P2P/IPC
u116642:82845:82930 [1] NCCL INFO Channel 02 : 1[21000] -> 0[1000] via P2P/IPC
u116642:82845:82930 [1] NCCL INFO Channel 03 : 1[21000] -> 0[1000] via P2P/IPC
u116642:82796:82908 [0] NCCL INFO Connected all rings
u116642:82796:82908 [0] NCCL INFO Connected all trees
u116642:82796:82908 [0] NCCL INFO threadThresholds 8/8/64 | 16/8/64 | 8/8/512
u116642:82796:82908 [0] NCCL INFO 4 coll channels, 4 p2p channels, 2 p2p channels per peer
u116642:82845:82930 [1] NCCL INFO Connected all rings
u116642:82845:82930 [1] NCCL INFO Connected all trees
u116642:82845:82930 [1] NCCL INFO threadThresholds 8/8/64 | 16/8/64 | 8/8/512
u116642:82845:82930 [1] NCCL INFO 4 coll channels, 4 p2p channels, 2 p2p channels per peer
u116642:82796:82908 [0] NCCL INFO comm 0x7f30ac002fb0 rank 0 nranks 2 cudaDev 0 busId 1000 -
Init COMPLETE
u116642:82796:82796 [0] NCCL INFO Launch mode Parallel

u116642:82845:82930 [1] NCCL INFO comm 0x7f8578002fb0 rank 1 nranks 2 cudaDev 1 busId 21000 -
Init COMPLETE

PyTorch Version :1.9 OS: Linux PyTorch installed with pip Python lightning version: 1.0.0 CUDA/cuDNN version: 11.4 GPU models and configuration: 2 GPU of A 6000

cc @tchaton @justusschock