Failure when build docker image

1

Hi all,
I am trying to build a docker image following the document. However, when I run make -f docker.Makefile, it failed with below error message

#25 35.44 cmake -DBUILD_PYTHON=True -DBUILD_TEST=True -DCMAKE_BUILD_TYPE=Release -DCMAKE_INSTALL_PREFIX=/opt/pytorch/torch -DCMAKE_PREFIX_PATH=/opt/conda/lib/python3.7/site-packages;/opt/conda/bin/../ -DNUMPY_INCLUDE_DIR=/opt/conda/lib/python3.7/site-packages/numpy/core/include -DPYTHON_EXECUTABLE=/opt/conda/bin/python -DPYTHON_INCLUDE_DIR=/opt/conda/include/python3.7m -DPYTHON_LIBRARY=/opt/conda/lib/libpython3.7m.so.1.0 -DTORCH_BUILD_VERSION=1.13.0a0+git9be97ea -DUSE_NUMPY=True /opt/pytorch
------
failed to solve with frontend dockerfile.v0: failed to solve with frontend gateway.v0: rpc error: code = Unknown desc = failed to build LLB: executor failed running [/bin/sh -c TORCH_CUDA_ARCH_LIST="3.5 5.2 6.0 6.1 7.0+PTX 8.0" TORCH_NVCC_FLAGS="-Xfatbin -compress-all"     CMAKE_PREFIX_PATH="$(dirname $(which conda))/../"     python setup.py install]: runc did not terminate sucessfully
docker.Makefile:48: recipe for target 'devel-image' failed

My docker version is Docker version 19.03.14, build 5eb3275d40, OS with Ubuntu 18.04.1 and CUDA version with 11.2. Please let me know if there is any other information useful I can provide.

Any help or idea would be appreciated!

2

Are you seeing any errors in the build log file?
The current error is only explaining that the docker container build failed due to a previous error.

3

Thank you for replying. I noticed there are some .bzl file does not exist errors, but I am using the master branch of pytorch, so files should be complete. Part of the error log is shown below due to character limit.

#25 30.98 CMake Error: File /opt/pytorch/build_variables.bzl does not exist.
#25 30.98 CMake Error at cmake/Codegen.cmake:351 (configure_file):
#25 30.98   configure_file Problem configuring file
#25 30.98 Call Stack (most recent call first):
#25 30.98   aten/src/ATen/CMakeLists.txt:163 (append_filelist)
#25 30.98 
#25 30.98 
#25 30.98 CMake Error: File /opt/pytorch/build_variables.bzl does not exist.
#25 30.98 CMake Error at cmake/Codegen.cmake:351 (configure_file):
#25 30.98   configure_file Problem configuring file
#25 30.98 Call Stack (most recent call first):
#25 30.98   aten/src/ATen/CMakeLists.txt:164 (append_filelist)
#25 30.98 
#25 30.98 
#25 30.98 CMake Error: File /opt/pytorch/build_variables.bzl does not exist.
#25 30.98 CMake Error at cmake/Codegen.cmake:351 (configure_file):
#25 30.98   configure_file Problem configuring file
#25 30.98 Call Stack (most recent call first):
#25 30.98   aten/src/ATen/CMakeLists.txt:245 (append_filelist)
#25 30.98 
#25 30.98 
#25 30.98 CMake Error: File /opt/pytorch/build_variables.bzl does not exist.
#25 30.98 CMake Error at cmake/Codegen.cmake:351 (configure_file):
#25 30.98   configure_file Problem configuring file
#25 30.98 Call Stack (most recent call first):
#25 30.98   aten/src/ATen/CMakeLists.txt:246 (append_filelist)
#25 30.98 
#25 30.98 
#25 30.98 CMake Error: File /opt/pytorch/build_variables.bzl does not exist.
#25 30.98 CMake Error at cmake/Codegen.cmake:351 (configure_file):
#25 30.98   configure_file Problem configuring file
#25 30.98 Call Stack (most recent call first):
#25 30.98   aten/src/ATen/CMakeLists.txt:247 (append_filelist)
4

I’m not familiar enough with bazel as I’m not using it to build PyTorch from source, but a quick search yields this PR which has moved the file to the root folder so check if this file if indeed there.

1 Like
5

Thanks for pointing out. I checked the root folder and the build_variables.bzl is indeed in my root folder. My file structure is below

.
|-- BUCK.oss
|-- BUILD.bazel
|-- CITATION
|-- CMakeLists.txt
|-- CODEOWNERS
|-- CODE_OF_CONDUCT.md
|-- CONTRIBUTING.md
|-- Dockerfile
|-- GLOSSARY.md
|-- LICENSE
|-- MANIFEST.in
|-- Makefile
|-- NOTICE
|-- README.md
|-- RELEASE.md
|-- SECURITY.md
|-- WORKSPACE
|-- android
|-- aten
|-- aten.bzl
|-- benchmarks
|-- binaries
|-- buckbuild.bzl
|-- build
|-- build.bzl
|-- build_variables.bzl
|-- c10
|-- c2_defs.bzl
|-- c2_test_defs.bzl
|-- caffe2
|-- cmake
|-- defs.bzl
|-- defs_gpu.bzl
|-- defs_hip.bzl
|-- docker.Makefile
|-- docs
|-- img.txt
|-- ios
|-- key.asc
|-- make_log.txt
|-- modules
|-- mypy-strict.ini
|-- mypy.ini
|-- mypy_plugins
|-- pt_defs.oss.bzl
|-- pt_template_srcs.bzl
|-- pyproject.toml
|-- pytest.ini
|-- requirements-flake8.txt
|-- requirements.txt
|-- scripts
|-- setup.py
|-- test
|-- third_party
|-- tools
|-- torch
|-- torchgen
|-- tree.txt
|-- ubsan.supp
|-- ufunc_defs.bzl
`-- version.txt
18 directories, 43 files

I noticed that Dockerfile is trying to copy this directory to the image through

FROM dev-base as submodule-update
WORKDIR /opt/pytorch
COPY . .
RUN ls /opt/pytorch
RUN git submodule update --init --recursive --jobs 0

However, when I use RUN ls /opt/pytorch to print the pasted contents, there is no build_variables.bzl in the image. The weird thing is that some files do exist, such like defs.bzl, defs_gpu.bzl, etc,.