Is it appropriate to mix some sort of action call with a resource URI (e.g. /collection/123?action=resendEmail)? Would it be better to specify the action and pass the resource id to it (e.g. /collection/resendEmail?id=123)? Is this the wrong way to be going about it? Traditionally (at least with HTTP) the action being performed is the request method (GET, POST, PUT, DELETE), but those don't really allow for custom actions with a resource.
I'd rather model that in a different way, with a collection of resources representing the emails that are to be sent; the sending will be processed by the internals of the service in due course, at which point the corresponding resource will be removed. (Or the user could DELETE the resource early, causing a canceling of the request to do the send.)
Whatever you do, don't put verbs in the resource name! That's the noun (and the query part is the set of adjectives). Nouning verbs weirds REST!
I use the querystring portion of the URL to filter the set of resources returned when querying a collection (e.g. /collection?someField=someval). Within my API controller I then determine what kind of comparison it is going to do with that field and value. I've found this really doesn't work. I need a way to allow the API user to specify the type of comparison they want to perform.
The best idea I've come up with so far is to allow the API user to specify it as an appendage to the field name (e.g. /collection?someField:gte=someval - to indicate that it should return resources where someField is greater than or equal to (>=) whatever someval is. Is this a good idea? A bad idea? If so, why? Is there a better way to allow the user to specify the type of comparison to perform with the given field and value?
I'd rather specify a general filter clause and have that as an optional query parameter on any request to fetch the contents of the collection. The client can then specify exactly how to restrict the set returned, in whatever way you desire. I'd also worry a bit about the discoverability of the filter/query language; the richer you make it, the harder it is for arbitrary clients to discover. An alternative approach which, at least theoretically, deals with that discoverability issue is to allow making restriction sub-resources of the collection, which clients obtain by POSTing a document describing the restriction to the collection resource. It's still a slight abuse, but at least it's one you can clearly make discoverable!
This sort of discoverability is one of the things that I find least strong with REST.
I often see URI's that look something like /person/123/dogs to get the persons dogs. I generally have avoided something like that because in the end I figure that by creating a URI like that you are actually just accessing a dogs collection filtered by a specific person ID. It would be equivalent to /dogs?person=123. Is there ever really a good reason for a REST URI to be more than two levels deep (/collection/resource_id)?
When the nested collection is truly a sub-feature of the outer collection's member entities, it is reasonable to structure them as a sub-resource. By “sub-feature” I mean something like UML composition relation, where destroying the outer resource naturally means destroying the inner collection.
Other types of collection can be modeled as an HTTP redirect; thus /person/123/dogs can indeed be responded to by doing a 307 that redirects to /dogs?person=123. In this case, the collection isn't actually UML composition, but rather UML aggregation. The difference matters; it is significant!
I assume you are expecting your site have a large amount of traffic. If that assumption is correct, I would say go with a tweaked version of Implementation 3.
If api.site.com and gateway.site.com are on two different servers, I would also have a third server to house the DB and have both apps leverage the centralized DB. This will eliminate the need for api.site.com and gateway.site.com to even know about each other and keep all work in a designated area, making any debugging that much easier.
And if my assumption is incorrect, I'd say go a differently tweaked version of Implementation 3. So you'd build you two separate apps that communicate with the same Db, and simply house everything on the same server. Which is basically what I just said above but all on the same server. This should reduce initial overhead and make scaling a breeze when the time comes.
Best Answer
The ETag mechanism specifies only the communication protocol for optimistic locking. It's the responsibility of the application service to implement the mechanism to detect concurrent updates to enforce the optimistic lock.
In a typical application that uses a database, you'd usually do this by opening a transaction when processing a PUT request. You'd normally read the existing state of the database inside that transaction (to gain a read lock), check your Etag validity, and overwrite the data (in a way that'll cause a write conflict when there's any incompatible concurrent transaction), then commit. If you setup the transaction correctly, then one of the commits should fail because they'll both be trying to update the same data concurrently. You'll then be able to use this transaction failure to either return 412 or retry the request, if it makes sense for the application.