When in doubt, consult the documentation. Reviewing the W3C definitions for HTTP Status codes, gives us this:
200 OK - The request has succeeded. The information returned with the response is dependent on the method used in the request.
404 Not Found - The server has not found anything matching the Request-URI.
In the context of your API, it very much depends on how queries are created and how objects are retrieved. But, my interpretation has always been that:
- If I ask for a particular object, and it exists return
200 code, if it doesn't exist return the correct 404 code.
- But, if I ask for a set of objects that match a query, a null set is a valid response and I want that returned with a
200 code. The rationale for this is that the query was valid, it succeeded and the query returned nothing.
So in this case you are correct, the service isn't searching for "a specific thing" it is requesting a particular thing, if that thing isn't found say that clearly.
I think Wikipedia puts it best:
200 OK - ... The actual response will depend on the request method used. In a GET request, the response will contain an entity corresponding to the requested resource.
404 Not Found - The requested resource could not be found but may be available again in the future. Subsequent requests by the client are permissible.
Seems pretty clear to me.
Regarding the example requests
/GoalTree/GetByDate?versionDate=...
/GoalTree/GetById?versionId=...
For the format, you said, you always return the nearest revision to that date. It will never not return an object, so it should always be returning 200 OK. Even if this were able to take a date range, and the logic were to return all objects within that timeframe returning 200 OK - 0 Results is ok, as that is what the request was for - the set of things that met that criteria.
However, the latter is different as you are asking for a specific object, presumably unique, with that identity. Returning 200 OK in this case is wrong as the requested resource doesn't exist and is not found.
Regarding choosing status codes
- 2xx codes Tell a User Agent (UA) that it did the right thing, the request worked. It can keep doing this in the future.
- 3xx codes Tell a UA what you asked probably used to work, but that thing is now elsewhere. In future the UA might consider just going to the redirect.
- 4xx codes Tell a UA it did something wrong, the request it constructed isn't proper and shouldn't try it again, without at least some modification.
- 5xx codes Tell a UA the server is broken somehow. But hey that query could work in the future, so there is no reason not to try it again. (except for 501, which is more of a 400 issue).
You mentioned in a comment using a 5xx code, but your system is working. It was asked a query that doesn't work and needs to communicate that to the UA. No matter how you slice it, this is 4xx territory.
Consider an alien querying our solar system
Alien: Computer, please tell me all planets that humans inhabit.
Computer: 1 result found. Earth
Alien: Computer, please tell me about Earth.
Computer: Earth - Mostly Harmless.
Alien: Computer, please tell me about all planets humans inhabit, outside the asteroid belt.
Computer: 0 results found.
Alien: Computer, please destroy Earth.
Computer: 200 OK.
Alien: Computer, please tell me about Earth.
Computer: 404 - Not Found
Alien: Computer, please tell me all planets that humans inhabit.
Computer: 0 results found.
Alien: Victory for the mighty Irken Empire!
Return the harshest, unfriendliest result possible in the event of a malformed request (one that returns more data than your metering allows is malformed). I suggest returning a 4** error code. Then, also provide paging parameters, so that users may request pages. oData has this feature, for instance. Do not truncate the data silently, under any circumstances.
Consulting with customers is a bad idea. They are going to tell you to do whatever possible to minimize errors, which is a bad engineering approach. This is your decision, take it by the horns and do the right thing.
An example of a paginated api is oData:
http://www.odata.org/documentation/odata-version-2-0/uri-conventions/
Best Answer
The whole of the Internet is built on conventions. We call them RFCs. While nobody will come and arrest you if you violate an RFC, you do run the risk that your service will not interoperate with the rest of the world. And if that happens, you run the risk of your startup not getting any customers, your business getting bad press, your stockholders revolting, your getting laid off permanently, etc.
HTTP status codes have their own IANA registry, each one traceable back to the RFC (or in one case, I-D) that defined it.
In the particular case of Twitter's strange 420 status code versus the standard 429 status code defined in RFC 6585, the most likely explanation is that the latter was only recently defined; the RFC dates to April 2012. We see that Twitter only uses 420 in the previous deprecated version 1 of its API; the current API version 1.1 actually uses the 429 status code. So it's clear that Twitter needed a status code for this and defined their own; once a standard one was available they switched to it.
Best practice, of course, is to stick as closely to the standards as possible. When you read RFCs, you will almost always find words like "MUST" and "SHOULD"; these have specific meanings when you are building your application, which you can find in RFC 2119.