Installing Anaconda on Amazon Elastic Beanstalk

Question

I've added deploy commands to my Elastic Beanstalk deployment which download the Anaconda installer, and install it into /anaconda. Everything goes well, but I cannot seem to correctly modify the PATH of my instance to include /anaconda/bin as suggested by the Anaconda installation page. If I SSH into an instance and manually add it, everything works fine. But this is obviously not the correct approach, as machines will be added automatically by EB.

So my question is: how can I use Anaconda in my script?

A couple more details:

• I've tried adding /anaconda/bin to the system PATH all ways I can think of. Pre/post deploy scripts, custom environment variables, etc. It seems that no matter what I do, the modifications don't persist to when the application is run.
• I've tried to include Anaconda via adding it to sys.path: sys.path.append('/anaconda/bin')
  to no avail. Using the following: sys.path.append('/anaconda/lib/python2.7/site-packages') allows me to import some packages but fails on import pandas. Strangely enough, if I SSH into the instance and run the application with their python (/opt/python/run/venv/bin/python2.7) it runs fine. Am I going crazy? Why does it fail on a specific import statement when run via EB?

Answer 1

Found the answer: import pandas was failing because matplotlib was failing to initialize, because it was trying to get the current user's home directory. Since the application is run via WSGI, the HOME variable is set to /home/wsgi but this directory doesn't exist. So, creating this directory via deployment command fixed this issue.

---

My overall setup to use Anaconda on Elastic Beanstalk is as follows:
.ebextensions/options.config contains:

commands:
  00_download_conda:
    command: 'wget http://repo.continuum.io/archive/Anaconda-2.0.1-Linux-x86_64.sh'
    test: test ! -d /anaconda
  01_install_conda:
    command: 'bash Anaconda-2.0.1-Linux-x86_64.sh -b -f -p /anaconda'
    test: test ! -d /anaconda
  02_create_home:
    command: 'mkdir -p /home/wsgi'

00_download_conda simply downloads Anaconda. See here for latest Anaconda version download link. The test commands are EB's way of letting you only execute the command if the test fails...Just prevents double downloading when in development.
01_install_conda installs Anaconda with options -b -f -p /anaconda which allows it to be installed in the specified directory, without user input, and skips installation if it has already been installed.
02_create_home creates the missing directory.

And finally - to use Anaconda inside your python application: sys.path.append('/anaconda/lib/python2.7/site-packages')
Cheers!