Open
Description
Branch
1.x branch (1.x version, such as v1.0.0rc2, or dev-1.x branch)
Prerequisite
- I have searched Issues and Discussions but cannot get the expected help.
- I have read the documentation but cannot get the expected help.
- The bug has not been fixed in the latest version.
Environment
sys.platform: linux
Python: 3.8.13 | packaged by conda-forge | (default, Mar 25 2022, 06:04:10) [GCC 10.3.0]
CUDA available: True
numpy_random_seed: 2147483648
GPU 0,1,2,3: Tesla V100-SXM2-16GB
CUDA_HOME: /usr/local/cuda
NVCC: Cuda compilation tools, release 11.3, V11.3.109
GCC: gcc (Ubuntu 9.4.0-1ubuntu1~20.04.1) 9.4.0
PyTorch: 1.12.1+cu113
PyTorch compiling details: PyTorch built with:
- GCC 9.3
- C++ Version: 201402
- Intel(R) Math Kernel Library Version 2020.0.0 Product Build 20191122 for Intel(R) 64 architecture applications
- Intel(R) MKL-DNN v2.6.0 (Git Hash 52b5f107dd9cf10910aaa19cb47f3abf9b349815)
- OpenMP 201511 (a.k.a. OpenMP 4.5)
- LAPACK is enabled (usually provided by MKL)
- NNPACK is enabled
- CPU capability usage: AVX2
- CUDA Runtime 11.3
- NVCC architecture flags: -gencode;arch=compute_37,code=sm_37;-gencode;arch=compute_50,code=sm_50;-gencode;arch=compute_60,code=sm_60;-gencode;arch=compute_70,code=sm_70;-gencode;arch=compute_75,code=sm_75;-gencode;arch=compute_80,code=sm_80;-gencode;arch=compute_86,code=sm_86
- CuDNN 8.3.2 (built against CUDA 11.5)
- Magma 2.5.2
- Build settings: BLAS_INFO=mkl, BUILD_TYPE=Release, CUDA_VERSION=11.3, CUDNN_VERSION=8.3.2, CXX_COMPILER=/opt/rh/devtoolset-9/root/usr/bin/c++, CXX_FLAGS= -fabi-version=11 -Wno-deprecated -fvisibility-inlines-hidden -DUSE_PTHREADPOOL -fopenmp -DNDEBUG -DUSE_KINETO -DUSE_FBGEMM -DUSE_QNNPACK -DUSE_PYTORCH_QNNPACK -DUSE_XNNPACK -DSYMBOLICATE_MOBILE_DEBUG_HANDLE -DEDGE_PROFILER_USE_KINETO -O2 -fPIC -Wno-narrowing -Wall -Wextra -Werror=return-type -Wno-missing-field-initializers -Wno-type-limits -Wno-array-bounds -Wno-unknown-pragmas -Wno-unused-parameter -Wno-unused-function -Wno-unused-result -Wno-unused-local-typedefs -Wno-strict-overflow -Wno-strict-aliasing -Wno-error=deprecated-declarations -Wno-stringop-overflow -Wno-psabi -Wno-error=pedantic -Wno-error=redundant-decls -Wno-error=old-style-cast -fdiagnostics-color=always -faligned-new -Wno-unused-but-set-variable -Wno-maybe-uninitialized -fno-math-errno -fno-trapping-math -Werror=format -Werror=cast-function-type -Wno-stringop-overflow, LAPACK_INFO=mkl, PERF_WITH_AVX=1, PERF_WITH_AVX2=1, PERF_WITH_AVX512=1, TORCH_VERSION=1.12.1, USE_CUDA=ON, USE_CUDNN=ON, USE_EXCEPTION_PTR=1, USE_GFLAGS=OFF, USE_GLOG=OFF, USE_MKL=ON, USE_MKLDNN=OFF, USE_MPI=ON, USE_NCCL=ON, USE_NNPACK=ON, USE_OPENMP=ON, USE_ROCM=OFF,
TorchVision: 0.13.1+cu113
OpenCV: 4.6.0
MMEngine: 0.3.2
MMCV: 2.0.0rc3
MMCV Compiler: GCC 9.3
MMCV CUDA Compiler: 11.3
MMSelfSup: 1.0.0rc3+6db0433
Describe the bug
I'm attempting to train MAE with MPI across 4 nodes (AWS P3dn.24xlarge instances, so 8 32GB V100s each). It runs successfully for a while, but errors out after about 10 epochs. I've tried resuming a few times in a row, and it always seems to crash after 10-12 epochs.
Reproduces the problem - code sample
Run command below
Reproduces the problem - command or script
mpirun --host algo-3:8,algo-1:8,algo-2:8,algo-4:8 -np 32 --allow-run-as-root --tag-output --oversubscribe -mca btl_tcp_if_include eth0 -mca oob_tcp_if_include eth0 -mca plm_rsh_no_tree_spawn 1 -mca pml ob1 -mca btl ^openib -mca orte_abort_on_non_zero_status 1 -mca btl_vader_single_copy_mechanism none -mca plm_rsh_num_concurrent 4 -x NCCL_SOCKET_IFNAME=eth0 -x NCCL_DEBUG=INFO -x LD_LIBRARY_PATH -x PATH -x SMDATAPARALLEL_USE_HOMOGENEOUS=1 -x FI_PROVIDER=efa -x RDMAV_FORK_SAFE=1 -x LD_PRELOAD=/opt/conda/lib/python3.8/site-packages/gethostname.cpython-38-x86_64-linux-gnu.so -verbose -x NCCL_DEBUG=VERSION -x FI_EFA_USE_DEVICE_RDMA=1 -x NCCL_PROTO=simple -x SMDATAPARALLEL_SERVER_ADDR=algo-3 -x SMDATAPARALLEL_SERVER_PORT=7592 -x SAGEMAKER_INSTANCE_TYPE=ml.p3dn.24xlarge smddprun /opt/conda/bin/python3.8 -m mpi4py sagemaker_train.py --amp True --cfg-options seed=0 log_level='INFO' visualizer.save_dir='/opt/ml/checkpoints' default_hooks.logger.interval=50 train_dataloader.dataset.data_root='/opt/ml/input/data/dataset' train_dataloader.batch_size=256 auto_scale_lr.base_batch_size=512auto_scale_lr.enable=True --config configs/selfsup/mae/mae_vit-base-p16_8xb512-amp-coslr-300e_in1k.py
Reproduces the problem - error message
[1,mpirank:1,algo-1]<stdout>: File "sagemaker_train.py", line 172, in main
[1,mpirank:1,algo-1]<stdout>: runner.train()
[1,mpirank:1,algo-1]<stdout>: File "/opt/conda/lib/python3.8/site-packages/mmengine/runner/runner.py", line 1684, in train
[1,mpirank:1,algo-1]<stdout>: model = self.train_loop.run() # type: ignore
[1,mpirank:1,algo-1]<stdout>: File "/opt/conda/lib/python3.8/site-packages/mmengine/runner/loops.py", line 90, in run
[1,mpirank:1,algo-1]<stdout>: self.run_epoch()
[1,mpirank:1,algo-1]<stdout>: File "/opt/conda/lib/python3.8/site-packages/mmengine/runner/loops.py", line 105, in run_epoch
[1,mpirank:1,algo-1]<stdout>: for idx, data_batch in enumerate(self.dataloader):
[1,mpirank:1,algo-1]<stdout>: File "/opt/conda/lib/python3.8/site-packages/torch/utils/data/dataloader.py", line 688, in __next__
[1,mpirank:1,algo-1]<stdout>: (data, worker_id) = self._next_data()
[1,mpirank:1,algo-1]<stdout>: File "/opt/conda/lib/python3.8/site-packages/torch/utils/data/dataloader.py", line 1449, in _next_data
[1,mpirank:1,algo-1]<stdout>: return (self._process_data(data), w_id)
[1,mpirank:1,algo-1]<stdout>: File "/opt/conda/lib/python3.8/site-packages/torch/utils/data/dataloader.py", line 1475, in _process_data
[1,mpirank:1,algo-1]<stdout>: data.reraise()
[1,mpirank:1,algo-1]<stdout>: File "/opt/conda/lib/python3.8/site-packages/torch/_utils.py", line 461, in reraise
[1,mpirank:1,algo-1]<stdout>: raise exception
[1,mpirank:1,algo-1]<stdout>:OSError: Caught OSError in DataLoader worker process 0.
[1,mpirank:1,algo-1]<stdout>:Original Traceback (most recent call last):
[1,mpirank:1,algo-1]<stdout>: File "/opt/conda/lib/python3.8/site-packages/mmengine/fileio/backends/local_backend.py", line 34, in get
[1,mpirank:1,algo-1]<stdout>: value = f.read()
[1,mpirank:1,algo-1]<stdout>:ConnectionAbortedError: [Errno 103] Software caused connection abort
[1,mpirank:1,algo-1]<stdout>:
Additional information
No response