REST APIs: custom HTTP headers vs URL parameters

When do you use custom HTTP headers in the request part of a REST API ?

Example:

Would you ever use

GET /orders/view 
(custom HTTP header) CLIENT_ID: 23

instead of

GET /orders/view/client_id/23 or 
GET /orders/view/?client_id=23

The URL indicates the resource itself. A "client" is a resource that can be acted upon, so should be part of the base url: /orders/view/client/23.

Parameters are just that, to parameterize access to the resource. This especially comes into play with posts and searches: /orders/find?q=blahblah&sort=foo. There's a fine line between parameters and sub-resources: /orders/view/client/23/active versus /orders/view/client/23?show=active. I recommend the sub-resource style and reserve parameters for searches.

Since each endpoint REpresents a State Transfer (to mangle the mnemonic), custom headers should only be used for things that don't involve the name of the resource (the url), the state of the resource (the body), or parameters directly affecting the resource (parameters). That leaves true metadata about the request for custom headers.

HTTP has a very wide selection of headers that cover most everything you'll need. Where I've seen custom headers come up is in a system to system request operating on behalf of a user. The proxy system will validate the user and add "X-User: userid" to the headers and use the system credentials to hit the endpoint. The receiving system validates that the system credentials are authorized to act on behalf of the user, then validate that the user is authorized to perform the action.

I would only use a custom header when there is no other way to pass information by standard or convention. Darren102 is explaining the typical way to pass that value. Your Api will be much more friendly by using typical patterns verse using custom headers.That's not to say you won't have a case to use them, just that they should be the last resort and something not already handled by the HTTP spec.

Use HTTP Headers for sending,

• Directives ( read the input in JSON format )

• Metadata ( who is making the request )

• Common data to be sent on every request ( like Authentication, session )

Use Path Parameters to make,

• Specific requests on a resource ( /country/state/city )

  They must form a logical hierarchy

Use Query Parameters for sending,

• Action requests on a resource ( like pagination, filters )

> When do you use...HTTP headers in the request part of a REST API?

Authentication: GUIDs, basic authentication, custom tokens, etc. e.g., https://stackoverflow.com/questions/12086041/basic-authentication-with-a-guid-token-for-rest-api-instead-of-username-password

If you get involved in passing tokens or other authentication-like information between domains covered by PCI-DSS or other security rules you may also have to bury parameters because some regulations explicitly require authentication elements to stay out of URLs that could be trivially replayed (from browser histories, proxy logs, etc.).

Custom headers have the following advantages:

• Keeps urls free from security stuff (safer, not in browser/proxy caches)

Personally I would only use those internally between my own web code and my own web server in case I need something special.

There is no standard for REST however the accepted way would be

GET /orders/view/23

Not using the custom headers and hence the 23 after view assumes to be the id hence you would have a function that takes in the id and hence produces just that information.

I wouldn't use custom headers as you don't know if any proxies will pass those on. URL based is the way to go.

> GET /orders/view/client/23

Definitely OK:

GET /orders/view/client_id/23 or 
GET /orders/view/?client_id=23

Also OK:

GET /orders/view/23 or 

I would think this would be OK, too:

POST /orders/view 
(custom HTTP header) CLIENT_ID: 23

You can use custom headers to include more information about a partially processed request considering that Enveloping is not a good practice. The headers are secure.