PyTorch使用tensorboard的SummaryWriter报错

PyTorch使用tensorboard可以显示网络运行情况，但偶尔使用SummaryWriter时遇到Segmentation fault错误。
利用python3的faulthandler，可定位到出错的代码行，具体操作有两种方式如下：
(1) 在代码中写入faulthandler

import faulthandler
# 在import之后直接添加以下启用代码即可
faulthandler.enable()
# 后边正常写你的代码

1
2
3
4
5

(2)直接通过命令行来启用，运行时添加-X faulthandler参数即可：

python -X faulthandler your_script.py
1

错误代码：

import xxx
from torch.utils.tensorboard import SummaryWriter
import yyy

log_dir = './run_logs'
self.writer = SummaryWriter(log_dir)
1
2
3
4
5
6

报错如下：

 File "/opt/conda/lib/python3.7/site-packages/tensorboard/compat/__init__.py", line 45 in tf
  File "/opt/conda/lib/python3.7/site-packages/tensorboard/lazy.py", line 50 in load_once
  File "/opt/conda/lib/python3.7/site-packages/tensorboard/lazy.py", line 97 in wrapper
  File "/opt/conda/lib/python3.7/site-packages/tensorboard/lazy.py", line 65 in __getattr__
  File "/opt/conda/lib/python3.7/site-packages/tensorboard/summary/writer/event_file_writer.py", line 72 in __init__
  File "/opt/conda/lib/python3.7/site-packages/torch/utils/tensorboard/writer.py",line 66 in __init__
  File "/opt/conda/lib/python3.7/site-packages/torch/utils/tensorboard/writer.py",line 256 in _get_file_writer
  File "/opt/conda/lib/python3.7/site-packages/torch/utils/tensorboard/writer.py", line 225 in __init__


1
2
3
4
5
6
7
8
9
10

经过查询，发现极有可能是导包(from torch.utils.tensorboard import SummaryWriter)顺序导致的问题。参考链接：[TensorBoard] The different order of import SummaryWriter may cause Segmentation fault

将from torch.utils.tensorboard import SummaryWriter顺序调前，可以顺利导包和实例化了。

import xxx
from torch.utils.tensorboard import SummaryWriter
import yyy

1
2
3
4

改为

from torch.utils.tensorboard import SummaryWriter
import xxx
import yyy

1
2
3
4