docker运行

**

一、直接运行Dockerfile出错

**
报错：

E: Unable to locate package libcudnn7
E: Version '2.4.8-1+cuda10.1' for 'libnccl2' was not found
E: Version '2.4.8-1+cuda10.1' for 'libnccl-dev' was not found

分析：
首先下载的docker镜像应该没问题：
FROM nvidia/cuda:10.1-devel-ubuntu18.04
从https://hub.docker.com/r/nvidia/cuda里面找到了对应的系统pull下来
然后按照Dockerfile文件一行行排查，发现是安装libcudnn7出现问题

RUN apt-get update && apt-get install -y --allow-downgrades --allow-change-held-packages --no-install-recommends \
build-essential \
cmake \
g++-4.8 \
git \
curl \
docker.io \
vim \
wget \
ca-certificates \
libcudnn7=${CUDNN_VERSION} \

推测是镜像里面没有libcudnn7下载源，进去镜像查看：

docker ps（发现没有运行）
docker images（找到镜像文件）
REPOSITORY          TAG                      IMAGE ID            CREATED             SIZE
<none>              <none>                   bea2518b9b5d        15 hours ago        2.99GB
<none>              <none>                   6b7222d063f7        22 hours ago        2.88GB
nvidia/cuda         10.1-devel-ubuntu18.04   00e5c760566b        3 weeks ago         2.94GB
<none>              <none>                   efbcebcd213a        6 weeks ago         2.8GB
nvidia/cuda         <none>                   95b9fad2b2bf        2 months ago        2.8GB
nvidia/cuda         10.1-devel-ubuntu16.04   a7223cfec628        5 months ago        2.85GB
hello-world         latest                   feb5d9fea6a5        9 months ago        13.3kB
<none>              <none>                   c665322da80e        13 months ago       7.56GB

docker run -d --name cuda nvidia/cuda:10.1-devel-ubuntu18.04 /bin/sh

错误执行方法：

docker run -d --name cuda nvidia/cuda:10.1-devel-ubuntu18.04 /bin/sh
---2ff5e4726288122258cccda865914ffd7ab35b3a9ea42e7e4ad47edaf902bc18
docker ps
---CONTAINER ID IMAGE COMMAND CREATED STATUS PORTS NAMES
docker ps -l
---CONTAINER ID IMAGE COMMAND CREATED STATUS PORTS NAMES
---2ff5e4726288 nvidia/cuda:10.1-devel-ubuntu18.04 "/bin/sh" 12 seconds ago Exited (0) 12 seconds ago cuda
docker rm 2ff
---2ff

正确：

docker run -itd --name cuda nvidia/cuda:10.1-devel-ubuntu18.04 /bin/sh（执行）
---9d394cbab53a595902c2bca7f647957882e2f138ad39d0e81ecaa021df4e15e1
docker ps（查看）
---CONTAINER ID IMAGE COMMAND CREATED STATUS PORTS NAMES
---9d394cbab53a nvidia/cuda:10.1-devel-ubuntu18.04 "/bin/sh" 4 seconds ago Up 3 seconds cuda
docker exec -it 9d3 /bin/sh（进入docker镜像）
exit（退出docker镜像）

查证：

docker exec -it 9d3 /bin/sh（进入docker镜像）
apt install libcudnn7=7.6.5.32-1+cuda10.1
Reading package lists... Done
Building dependency tree
Reading state information... Done
E: Unable to locate package libcudnn7（发现确实是镜像中没有下载源）

需要在镜像中手动添加下载源，后续研究加入

二、导入已有docker报错

需要用到已经built的docker，直接导入load发现有问题：

docker load < dro-sfm-image.tar
Got permission denied while trying to connect to the Docker daemon socket at unix:///var/run/docker.sock: Post http://%2Fvar%2Frun%2Fdocker.sock/v1.39/images/load?quiet=0: dial unix /var/run/docker.sock: connect: permission denied

解决方法：

sudo chmod a+rw /var/run/docker.sock 

成功：

Loaded image ID: sha256:c665322da80e67af8dfea9f2ca38595838e6dd4eae3a9180eb2b1c0348b82d82