pipcook run error: socket.timeout: The read operation timed out when download torch
AnselDai opened this issue · comments
@AnselDai You could append a mirror parameter --tuna to avoid this error.
I use --tuna parameter but the problem still exist. Can i specify the source when download torch?
Sorry @AnselDai, current Pipcook don't have the ability to specify it in the user interface, but we config the index page at https://github.com/alibaba/pipcook/blob/master/packages/cli/src/config.ts#L9, could you please try to change that at your local machine(/usr/local/lib/node_modules/@pipcook/pipcook-cli/lib/config.js)?
And could you please provide the followings:
- logs
- macOS or Linux
- Node.js version
- the country where your PC is located
Thank you!
root@instance-pppo0843:~/myfile/pipcook_demo# pipcook run pipeline.json --tuna
✔ plugin @pipcook/plugins-object-detection-pascalvoc-data-collect@1.1.0 has already been installed
✔ plugin @pipcook/plugins-coco-data-access@1.1.1 has already been installed
ℹ PYTHON: Looking in indexes: https://pypi.tuna.tsinghua.edu.cn/simple
ℹ PYTHON: Collecting git+https://gitee.com/pipcook/detectron2.git (from -r /root/.pipcook/plugins/conda_envs/@pipcook/plugins-detectron-fasterrcnn-model-define@1.1.0/requirements.txt (line 6))
ℹ PYTHON: Cloning https://gitee.com/pipcook/detectron2.git to /tmp/pip-req-build-kki_8eby
⚠ NODE: npm WARN plugins@1.0.0 No description
⚠ NODE: npm WARN plugins@1.0.0 No repository field.
⚠ NODE:
ℹ NODE: + @pipcook/plugins-detectron-fasterrcnn-model-define@1.1.0
ℹ NODE: added 2 packages and audited 150 packages in 5.138s
ℹ NODE:
ℹ NODE: 3 packages are looking for funding
ℹ NODE: run npm fund for details
ℹ NODE:
ℹ NODE: found 0 vulnerabilities
ℹ NODE:
ℹ PYTHON: Complete output from command python setup.py egg_info:
ℹ PYTHON: Collecting torch==1.4.0
ℹ PYTHON: Downloading https://files.pythonhosted.org/packages/1a/3b/fa92ece1e58a6a48ec598bab327f39d69808133e5b2fb33002ca754e381e/torch-1.4.0-cp37-cp37m-manylinux1_x86_64.whl (753.4MB)
ℹ PYTHON: Exception:
ℹ PYTHON: Traceback (most recent call last):
ℹ PYTHON: File "/root/.pipcook/plugins/conda_envs/@pipcook/plugins-detectron-fasterrcnn-model-define@1.1.0/lib/python3.7/site-packages/pip/_vendor/urllib3/response.py", line 360, in _error_catcher
ℹ PYTHON: yield
ℹ PYTHON: File "/root/.pipcook/plugins/conda_envs/@pipcook/plugins-detectron-fasterrcnn-model-define@1.1.0/lib/python3.7/site-packages/pip/_vendor/urllib3/response.py", line 442, in read
ℹ PYTHON: data = self._fp.read(amt)
ℹ PYTHON: File "/root/.pipcook/plugins/conda_envs/@pipcook/plugins-detectron-fasterrcnn-model-define@1.1.0/lib/python3.7/site-packages/pip/_vendor/cachecontrol/filewrapper.py", line 62, in read
ℹ PYTHON: data = self.__fp.read(amt)
ℹ PYTHON: File "/root/.pipcook/server/node_modules/@pipcook/boa/.miniconda/lib/python3.7/http/client.py", line 457, in read
ℹ PYTHON: n = self.readinto(b)
ℹ PYTHON: File "/root/.pipcook/server/node_modules/@pipcook/boa/.miniconda/lib/python3.7/http/client.py", line 501, in readinto
ℹ PYTHON: n = self.fp.readinto(b)
ℹ PYTHON: File "/root/.pipcook/server/node_modules/@pipcook/boa/.miniconda/lib/python3.7/socket.py", line 589, in readinto
ℹ PYTHON: return self._sock.recv_into(b)
ℹ PYTHON: File "/root/.pipcook/server/node_modules/@pipcook/boa/.miniconda/lib/python3.7/ssl.py", line 1071, in recv_into
ℹ PYTHON: return self.read(nbytes, buffer)
ℹ PYTHON: File "/root/.pipcook/server/node_modules/@pipcook/boa/.miniconda/lib/python3.7/ssl.py", line 929, in read
ℹ PYTHON: return self._sslobj.read(len, buffer)
ℹ PYTHON: socket.timeout: The read operation timed out
In Linux
Node 12.18
In China
I use wget to test the link of pytorch, the download speed is really slow if i don't use a VPN.
@FeelyChau does It mean the PyIndex config is not working?
If the version of daemon is 1.1.2, the bug has been fixed by 1.2.0, please upgrade the daemon:
$ npm install @pipcook/pipcook-cli -g
$ pipcook init --tunaroot@instance-pppo0843:/lib# pipcook -v
Pipcook Tools v1.2.0 /usr/lib/node_modules/@pipcook/pipcook-cli/
Pipcook Daemon v1.2.0 /root/.pipcook/server/node_modules/@pipcook/daemon
I do use --tuna to init pipcook. The registry of npm I change to taobao source. Dose the taobao source bring the problem?
I do use
--tunato init pipcook. The registry of npm I change to taobao source. Dose the taobao source bring the problem?
It looks like the detectron2 package depends on the torch, and install it without the tuna flag.
It seems like pipcook use conda to create a virtual env for each plugins, so the torch installed on my machine doesn't work.
It seems like pipcook use conda to create a virtual env for each plugins, so the torch installed on my machine doesn't work.
A temporary workaround is to manually download the whl file and store it under ~/.cache/pip
This way pip will look for the cache first instead of downloading it directly.
It doesn't work. I put torch file both under ~/.cache/pip and ~/.pipcook/plugins/.pip, but it still try to download it from the web
Confirmed this is a bug at our detectron2 repository which doesn't use the specific index mirror to download.