search

MiniBullLab / easy_docker

1 stars 1 forks source link

镜像大小优化-替换底层镜像 #49

Closed vitahlin closed 3 years ago

vitahlin commented 3 years ago
nvidia/cuda版本 base ai_lib lib runtime workspace
10.0-base-ubuntu18.04 1.74G
10.0-runtime-ubuntu18.04 2.5G
10.0-devel-ubuntu18.04 3.67G

方案说明：替换cuda的基础镜像，采用更小的base镜像，cuda-base, cuda-runtime, cuda-devel区别：https://stackoverflow.com/questions/56405159/what-is-the-difference-between-devel-and-runtime-tag-for-a-docker-container

vitahlin commented 3 years ago

caffe编译错误：

PROTOC src/caffe/proto/caffe.proto
CXX .build_release/src/caffe/proto/caffe.pb.cc
CXX src/caffe/syncedmem.cpp
In file included from ./include/caffe/common.hpp:19:0,
                 from src/caffe/syncedmem.cpp:1:
./include/caffe/util/device_alternate.hpp:34:10: fatal error: cublas_v2.h: No such file or directory
 #include <cublas_v2.h>
          ^~~~~~~~~~~~~
compilation terminated.
Makefile:593: recipe for target '.build_release/src/caffe/syncedmem.o' failed
make: *** [.build_release/src/caffe/syncedmem.o] Error 1
ERROR: Service 'lib' failed to build: The command '/bin/sh -c echo "\n    Begin install caffe..." &&     cd /opt &&     curl -o caffe.zip http://118.31.19.101:8080/software/amba/docker_caffe.zip &&     unzip caffe.zip &&     cd /opt/caffe/python &&     pip3 install -i https://mirrors.aliyun.com/pypi/simple -r requirements.txt &&     cd /opt/caffe &&     make all &&     make pycaffe &&     echo "export PYTHONPATH=/opt/caffe/python:\$PYTHONPATH">>/etc/bash.bashrc &&     /bin/bash -c "source /etc/bash.bashrc" &&     rm -rf /opt/caffe.zip &&     echo "\n    Install caffe success"' returned a non-zero code: 2
Error response from daemon: No such image: easy_lib:latest

@lpj0822

vitahlin commented 3 years ago 
src/caffe/layers/cudnn_lcn_layer.cpp:49:15: note: 'CUDA_CHECK' declared here, later in the translation unit
     CUDA_CHECK(cudaMalloc(&tempData1, totalSizeInBytes));
     ~~~~~~~~~~^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
src/caffe/layers/cudnn_lcn_layer.cpp: In instantiation of 'caffe::CuDNNLCNLayer<Dtype>::~CuDNNLCNLayer() [with Dtype = double]':
src/caffe/layers/cudnn_lcn_layer.cpp:70:1:   required from here
src/caffe/layers/cudnn_lcn_layer.cpp:59:44: error: 'cudnnDestroyTensorDescriptor' was not declared in this scope
   cudnnDestroyTensorDescriptor(bottom_desc_);
                                            ^
src/caffe/layers/cudnn_lcn_layer.cpp:60:41: error: 'cudnnDestroyTensorDescriptor' was not declared in this scope, and no declarations were found by argument-dependent lookup at the point of instantiation [-fpermissive]
   cudnnDestroyTensorDescriptor(top_desc_);
                                         ^
src/caffe/layers/cudnn_lcn_layer.cpp:59:44: note: 'cudnnDestroyTensorDescriptor' declared here, later in the translation unit
   cudnnDestroyTensorDescriptor(bottom_desc_);
                                            ^
src/caffe/layers/cudnn_lcn_layer.cpp:63:23: error: 'cudnnDestroy' was not declared in this scope
   cudnnDestroy(handle_);
                       ^
src/caffe/layers/cudnn_lcn_layer.cpp:66:11: error: 'cudaFree' was not declared in this scope
   cudaFree(tempData1);
   ~~~~~~~~^~~~~~~~~~~
src/caffe/layers/cudnn_lcn_layer.cpp:67:11: error: 'cudaFree' was not declared in this scope, and no declarations were found by argument-dependent lookup at the point of instantiation [-fpermissive]
   cudaFree(tempData2);
   ~~~~~~~~^~~~~~~~~~~
src/caffe/layers/cudnn_lcn_layer.cpp:66:11: note: 'cudaFree' declared here, later in the translation unit
   cudaFree(tempData1);
   ~~~~~~~~^~~~~~~~~~~
Makefile:593: recipe for target '.build_release/src/caffe/layers/cudnn_lcn_layer.o' failed
vitahlin commented 3 years ago

目前来看，替换底层镜像的话，会出现比较多的编译问题，好像不是因为底层镜像问题。

base镜像标签没有处理，更新标签，继续测试

Mac上base caffe编译通过，opencv 失败

kingwangxiang commented 3 years ago

先不改base，看看lab。

vitahlin commented 3 years ago

测试，base和runtime镜像caffe编译都会出问题

vitahlin commented 3 years ago

不更换cuda版本，研究其他的镜像大小压缩方案