Static file mounting

[rgbkrk]

This begins the adventure of both:

• Sticking nginx in a container
• Mounting static assets from the userland container to the nginx container

This also tosses the static.tar legacy file.

[rgbkrk]

Alright, this is ready for review!

[rgbkrk]

This works! It's alive!

[smashwilson]

There's a trick you can use to register whether or not a newer image was actually pulled, which can save you extraneous restarts:

- name: pull the latest docker build
  command: docker pull myimage
  register: someimage
  changed_when: someimage.stdout.find('Downloaded newer image') != -1
  sudo: yes

[rgbkrk]

Hmmm, with that I'll need to start a new nginx container too then (so that it links to new static assets).

I'm not as worried about upgrading nodes in place since we can end up with a version mismatch here:

1. User creates a notebook server
2. (Some) assets are cached by their browser
3. We upgrade the static files served by nginx
4. They open a link to their server on a different browser, computer, etc.
5. The notebook now uses different js/css than is served by nginx

[smashwilson]

Hmmm, with that I'll need to start a new nginx container too then (so that it links to new static assets).

Yes, the idea is that later you can do:

- name: Restart nginx
  docker: image=nginx state=restarted
  when: someimage | changed

I'm not as worried about upgrading nodes in place since we can end up with a version mismatch here

Oh, right. I really wish the assets could be hash-fingerprinted, it would solve a lot of problems like this one...

[rgbkrk]

Even if we hash-fingerprint them, we're still serving a specific set from nginx rather than what any given container is necessarily running.

If we wanted to though, we could finish up the piece from tmpnb's tornado app that handles static asset serving. Right now it only serves static.tar when it could serve static for each path (making sure to key on image id).

[rgbkrk]

The better option, that I still haven't tried, is to try loading IPython's static assets over a CDN.

This involves setting the static_url setting for IPython's tornado server.