Indexing API Draft

[rgbkrk]

We'll want to expose a simplistic API for nbviewer to send notebooks to. Under the hood we'll hit elasticsearch, but that will be abstracted away from the notebook viewer.

Submitting a notebook

POST /api/index/{ uuid.uuid5(uuid.NAMESPACE_URL, notebook_url) }
{
  "public": true,
  "content": { Full notebook document },
  "notebook_url": "http://jakevdp.github.com/downloads/notebooks/XKCD_plots.ipynb"
}

Retrieving the stored contents

GET /api/index/{ uuid.uuid5(uuid.NAMESPACE_URL, notebook_url) }

Returns the trimmed version of the body from the post. The notebook document e.g. may or may not have the base64 encoded images stripped out.

Delete an entry from the index

DELETE /api/index/{ uuid.uuid5(uuid.NAMESPACE_URL, notebook_url) }

I'm on the fence about if we should slightly adhere to the contents API for consistency or just choose the fields that make sense to us.

[yuvipanda]

/me performs necromancy

Do you think APIs should be versioned? :)

[rgbkrk]

Yes!

...not that I always do it at the start

Nice to see you @yuvipanda!

Since this came from nowhere, do you prefer the API version in the path /api/v1/index/ or as part of an explicit Accept header application/vnd.jupyter.v3.raw+json?

[yuvipanda]

Definitely prefer it in the URL path - much more obvious, IMO.

[minrk]

+1 to versioned URLs. I don't like using Accept headers.