Torch.distributed.barrier doesn't work with pytorch 2.0 and Backend=NCCL

I still have the same problem with this (torch.distributed.barrier Bug with pytorch 2.0 and Backend=NCCL · Issue #98763 · pytorch/pytorch · GitHub)

import os
import torch
import torch.distributed as dist
import torch.multiprocessing as mp

def run(rank, size):
    """ Distributed function to be implemented later. """
    print('Rank {} of {}.'.format(rank, size))
    torch.distributed.barrier()
    print('after barrier')
    pass

def init_process(rank, size, fn, backend='nccl'):
    """ Initialize the distributed environment. """
    os.environ['MASTER_ADDR'] = '127.0.0.1'
    os.environ['MASTER_PORT'] = '29500'
    torch.cuda.set_device(rank)
    dist.init_process_group(backend, rank=rank, world_size=size)
    fn(rank, size)


if __name__ == "__main__":
    size = 6
    processes = []
    mp.set_start_method("spawn")
    for rank in range(size):
        p = mp.Process(target=init_process, args=(rank, size, run))
        p.start()
        processes.append(p)

    for p in processes:
        p.join()

It stuck at barrier. It doesn’t print “after barrier”.
And when it is in the barrier, GPU-util is almost 100%.
What can be the problem? I already changed torch version, nvidia driver, cuda version, cudnn version so many times. Can you give me any recommendation that I can do?
I already checked this code in another computer. It works well on the other computers.

Versions

Collecting environment information…
PyTorch version: 2.0.1+cu117
Is debug build: False
CUDA used to build PyTorch: 11.7
ROCM used to build PyTorch: N/A

OS: Ubuntu 20.04.6 LTS (x86_64)
GCC version: (Ubuntu 9.4.0-1ubuntu1~20.04.2) 9.4.0
Clang version: Could not collect
CMake version: version 3.27.7
Libc version: glibc-2.31

Python version: 3.10.13 (main, Sep 11 2023, 13:44:35) [GCC 11.2.0] (64-bit runtime)
Python platform: Linux-5.15.0-86-generic-x86_64-with-glibc2.31
Is CUDA available: True
CUDA runtime version: 11.7.99
CUDA_MODULE_LOADING set to: LAZY
GPU models and configuration:
GPU 0: NVIDIA RTX A6000
GPU 1: NVIDIA RTX A6000
GPU 2: NVIDIA RTX A6000
GPU 3: NVIDIA RTX A6000
GPU 4: NVIDIA RTX A6000
GPU 5: NVIDIA RTX A6000

Nvidia driver version: 535.113.01
cuDNN version: Could not collect
HIP runtime version: N/A
MIOpen runtime version: N/A
Is XNNPACK available: True

CPU:
Architecture: x86_64
CPU op-mode(s): 32-bit, 64-bit
Byte Order: Little Endian
Address sizes: 46 bits physical, 48 bits virtual
CPU(s): 40
On-line CPU(s) list: 0-39
Thread(s) per core: 2
Core(s) per socket: 10
Socket(s): 2
NUMA node(s): 2
Vendor ID: GenuineIntel
CPU family: 6
Model: 85
Model name: Intel(R) Xeon(R) Silver 4114 CPU @ 2.20GHz
Stepping: 4
CPU MHz: 800.000
CPU max MHz: 3000.0000
CPU min MHz: 800.0000
BogoMIPS: 4400.00
Virtualization: VT-x
L1d cache: 640 KiB
L1i cache: 640 KiB
L2 cache: 20 MiB
L3 cache: 27.5 MiB
NUMA node0 CPU(s): 0-9,20-29
NUMA node1 CPU(s): 10-19,30-39
Vulnerability Gather data sampling: Mitigation; Microcode
Vulnerability Itlb multihit: KVM: Mitigation: VMX disabled
Vulnerability L1tf: Mitigation; PTE Inversion; VMX conditional cache flushes, SMT vulnerable
Vulnerability Mds: Mitigation; Clear CPU buffers; SMT vulnerable
Vulnerability Meltdown: Mitigation; PTI
Vulnerability Mmio stale data: Mitigation; Clear CPU buffers; SMT vulnerable
Vulnerability Retbleed: Mitigation; IBRS
Vulnerability Spec rstack overflow: Not affected
Vulnerability Spec store bypass: Mitigation; Speculative Store Bypass disabled via prctl and seccomp
Vulnerability Spectre v1: Mitigation; usercopy/swapgs barriers and __user pointer sanitization
Vulnerability Spectre v2: Mitigation; IBRS, IBPB conditional, STIBP conditional, RSB filling, PBRSB-eIBRS Not affected
Vulnerability Srbds: Not affected
Vulnerability Tsx async abort: Mitigation; Clear CPU buffers; SMT vulnerable
Flags: fpu vme de pse tsc msr pae mce cx8 apic sep mtrr pge mca cmov pat pse36 clflush dts acpi mmx fxsr sse sse2 ss ht tm pbe syscall nx pdpe1gb rdtscp lm constant_tsc art arch_perfmon pebs bts rep_good nopl xtopology nonstop_tsc cpuid aperfmperf pni pclmulqdq dtes64 monitor ds_cpl vmx smx est tm2 ssse3 sdbg fma cx16 xtpr pdcm pcid dca sse4_1 sse4_2 x2apic movbe popcnt tsc_deadline_timer aes xsave avx f16c rdrand lahf_lm abm 3dnowprefetch cpuid_fault epb cat_l3 cdp_l3 invpcid_single pti intel_ppin ssbd mba ibrs ibpb stibp tpr_shadow vnmi flexpriority ept vpid ept_ad fsgsbase tsc_adjust bmi1 hle avx2 smep bmi2 erms invpcid rtm cqm mpx rdt_a avx512f avx512dq rdseed adx smap clflushopt clwb intel_pt avx512cd avx512bw avx512vl xsaveopt xsavec xgetbv1 xsaves cqm_llc cqm_occup_llc cqm_mbm_total cqm_mbm_local dtherm ida arat pln pts pku ospke md_clear flush_l1d arch_capabilities

Versions of relevant libraries:
[pip3] numpy==1.26.1
[pip3] open-clip-torch==2.22.0
[pip3] torch==2.0.1+cu117
[pip3] torchaudio==2.0.2+cu117
[pip3] torchvision==0.15.2+cu117
[pip3] triton==2.0.0
[conda] numpy 1.26.1 pypi_0 pypi
[conda] open-clip-torch 2.22.0 pypi_0 pypi
[conda] torch 2.0.1+cu117 pypi_0 pypi
[conda] torchaudio 2.0.2+cu117 pypi_0 pypi
[conda] torchvision 0.15.2+cu117 pypi_0 pypi
[conda] triton 2.0.0 pypi_0 pypi

### Tasks

After setting NCCL_P2P_DISABLE=1, it was solved. Sorry for my silly question. I hope this can help the people who have the same problems.

1 Like

That is interesting, disabling nccl p2p should not be necessary. If you reenable p2p and set NCCL_DEBUG=INFO I am curious about what the logs output, this is still something that we should fix. I am wondering why it doesn’t work on your specific machine.

I encountered a similar issue. When using 3090 or A5000, there were no problems, but when I ran the same code on A6000, it got stuck at the barrier. Looking at NCCL_DEBUG, it showed ‘Init COMPLETE’ and then went into an infinite loop without any further output. Anyway, it works after setting NCCL_P2P_DISABLE. Thank you!