作者：Jack Wu

现象

AWS Ubuntu AMI机器学习服务器无法自动激动 conda 的 tensorflow 环境问题。如我们想自动活动tensorflow_p36，在conda的提示下，于是在.bashrc中追加如下配置：

. /home/ubuntu/anaconda3/etc/profile.d/conda.sh
conda activate tensorflow_p36

但登录后发现使用的Python仍然是conda base的Python:

(tensorflow_p36) ubuntu@ip:~$ which python
which python
/home/ubuntu/anaconda3/bin/python

原因

当登录或tmux或bash嵌套时，脚本调用顺序.profile -> .bashrc。而.profile的内容如下。

# ~/.profile: executed by the command interpreter for login shells.
# This file is not read by bash(1), if ~/.bash_profile or ~/.bash_login
# exists.
# see /usr/share/doc/bash/examples/startup-files for examples.
# the files are located in the bash-doc package.

# the default umask is set in /etc/profile; for setting the umask
# for ssh logins, install and configure the libpam-umask package.
#umask 022

# if running bash
if [ -n "$BASH_VERSION" ]; then
    # include .bashrc if it exists
    if [ -f "$HOME/.bashrc" ]; then
        . "$HOME/.bashrc"
    fi
fi
# for private env
PATH="$HOME/bin:$HOME/.local/bin:$PATH"
export PATH="/home/ubuntu/anaconda3/bin:$PATH"

可以发现，其最后两行的环境变量配置，覆盖了conda activate tensorflow_p36生成的环境变量，通过env查看PATH可看到，

PATH=/home/ubuntu/bin:/home/ubuntu/.local/bin:/home/ubuntu/anaconda3/bin/:/home/ubuntu/anaconda3/envs/tensorflow_p36/bin:/home/ubuntu/bin:/home/ubuntu/.local/bin:/home/ubuntu/anaconda3/bin/:/usr/local/cuda/bin:/usr/local/bin ...

而这时我们已经在tensorflow_p36环境中，再运行conda activate tensorflow_p36也无效果。临时的解决方法是先deactivate再激活tensorflow_p36，以再覆盖一次环境变量PATH。

conda deactivate
conda activate tensorflow_p36

PATH=/home/ubuntu/anaconda3/envs/tensorflow_p36/bin:/home/ubuntu/bin:/home/ubuntu/.local/bin:/home/ubuntu/anaconda3/bin/:/home/ubuntu/anaconda3/envs/tensorflow_p36/bin:/home/ubuntu/bin:/home/ubuntu/.local/bin:/home/ubuntu/anaconda3/bin/:/usr/local/cuda/bin:/usr/local/bin ...

如何分析出如上的原因呢，可以在.bashrc头加上set -x，使bash打印每次运行的脚本的第一行信息：

#Source .dlamirc in .bashrc
set -x
source ~/.dlamirc
# ~/.bashrc: executed by bash(1) for non-login shells.
# see /usr/share/doc/bash/examples/startup-files (in the package bash-doc)
# for examples
...

解决

从上面可以看出，比较偷懒的解决方法是，修改.bashrc中的 tensorflow_p36 激活脚本为如下：

. /home/ubuntu/anaconda3/etc/profile.d/conda.sh
conda deactivate
conda activate tensorflow_p36