REST summary
- every resource has to be discoverable via links
- basically a developer can learn the API with a browser
- also, with links, you can mingle resources from any REST api
- verbs
- GET - retrieve a specific resource
- DELETE - delete a specific resource
- PUT - update a specific resource
- POST - create a new resource and return the ID - neither safe nor idempotent
- you can repeat calls to GET/DELETE/PUT all day long if they fail
- to provide complex functions (eg query and result) as resources...
- create a query resource with a TTL
- add to it
- execute it (and auto-release?)
- release it (or let this happen automatically with server-side TTL cleanup)
- how do we handle huge lists? like this?
- wya/users
- returns the first 10 users with a link to next 10
- wya/users/range/11-20
- {resource}/edit should return an edit form (cool) - I think recess already does this?
- authentication - use SSL with HTTP Basic Authentication, or SHA1 signature (ala Amazon S3)
REST Examples
- The best example I have to follow is the Twitter API. It *obviously* scales. It uses Basic Authentication, which sucks (everything is plaintext, holy shit). BETTER force it to use SSL.
- The other excellent example is the Amazon S3 API. Authentication is brilliant - it uses a SHA1 signature of the request, which MUST include a timestamp within the last 15 minutes. There is a set of rules for turning the request URL into the request string that is signed.
Signature = Base64( HMAC-SHA1( UTF-8-Encoding-Of( YourSecretAccessKeyID, StringToSign ) ) );