I'm writing a C# Web API server application, and will send JSON to it via a Mirth HTTP Sender destination. This post is about how to handle error conditions. Specifically, there are three scenarios I want to handle:
Sometimes we take the C# application server offline for a short period for system upgrade or maintenance, and Mirth is unable to connect at all. I want Mirth to queue all messages in order, and when the server is available, process them in the order they were received.
The server receives the request, but rejects it due to a problem with the content of the request, e.g., missing a required field. In accordance with REST conventions, the server will return a 400-level HTTP response. This message would be rejected every time it's submitted, so it should not be re-sent; just log the failure and move on to the next message.
The server receives the request, but something goes wrong on the server, and the server returns an HTTP 500 Server Error response. This would be the appropriate response, for example, when something in the server environment has gone wrong. One real-world example was the time the Web API server was running, but somebody rebooted the database server. REST conventions would suggest we continue to resend the message until the transient problem has been resolved.
For #1, initially I had it queue on failure/always, but it appears the response transformer never runs for messages that were queued (at least, the debug statements never showed in the log). I have turned queueing off, and set it to retry every ten seconds for an hour, and that seems to give the desired behavior. Am I on the right track here, or missing something?
For #2 and #3, returning any HTTP 400 or 500 error invokes the 1-hour retries. What I want is to apply the 1-hour retries for the 500 errors, but not the 400 errors. I’ve tried responseStatus = SENT in the response transformer, but the response transformer only runs once, after the hour has expired, and not for each retry.
This seems like a common problem, yet I’m not finding a solution. How are the rest of you handling this?
You're close!
So by default, the response transformer will only run if there's a response payload to transform. For connection problems, or possibly for 4xx/5xx responses that contain no payload, the response transformer won't execute.
However, if you set your response data types (From the Summary -> Set Data Types dialog, or from the Destinations -> Edit Response, Message Templates tab) to Raw, then the response transformer will execute all the time. The reason being that the Raw data type considers even an empty payload to be "transformable".
So turn queuing back on, and set your response data types to Raw. Then in the response transformer, if you look at the Reference tab there's a category for HTTP Sender:
You'll want the "response status line", that's the "HTTP/1.1 200 OK" line of the response that contains the response code. Here's a response transformer script that forces 4xx responses to error:
if (responseStatus == QUEUED) {
var statusLine = $('responseStatusLine');
if (statusLine) {
var parts = statusLine.split(' ');
if (parts.length >= 2) {
var responseCode = parseInt(parts[1], 10);
// Force 4xx responses to error
if (responseCode >= 400 && responseCode < 500) {
responseStatus = ERROR;
responseStatusMessage = statusLine;
}
}
}
}
We have an endpoint which when you post create a new version of resource and returns a 201 and the location of the newly created resource. It determines the new version number based on a comparison of the current version and the version being posted (using a semver like ruleset).
If the version you post is identical to the existing version then no version number is updated. What should we return in this case?
We could just return a 201 even though we have not technically created anything.
I don't want to return a 409 as its not really a conflict, like when you post something with the same id. If you posted the same thing when the existing version was slightly different then you would happily get a 201.
We could just return a 200, but then that would seem weird, and increases the response codes that the users have to deal with
Does the idempotency of the 201 response matter?
Any better suggestions?
How about 303 - See Other? Seems to fit. I draw your attention to this sentence
from the spec at https://www.w3.org/Protocols/rfc2616/rfc2616-sec10.html
This method exists primarily to allow the output of a POST-activated script to redirect the user agent to a selected resource.
That sounds like what you want to do to me. Here's the rest of it.
10.3.4 303 See Other
The response to the request can be found under a different URI and SHOULD be retrieved using a GET method on that resource. This method exists primarily to allow the output of a POST-activated script to redirect the user agent to a selected resource. The new URI is not a substitute reference for the originally requested resource. The 303 response MUST NOT be cached, but the response to the second (redirected) request might be cacheable.
The different URI SHOULD be given by the Location field in the response. Unless the request method was HEAD, the entity of the response SHOULD contain a short hypertext note with a hyperlink to the new URI(s).
Note: Many pre-HTTP/1.1 user agents do not understand the 303
status. When interoperability with such clients is a concern, the
302 status code may be used instead, since most user agents react
to a 302 response as described here for 303
I am a bit puzzled by the other answers as some get it almost right. So, let's clear up things a bit. If all requests are indeed performed with the POST method, in the context of ReSTfulness, they are supposed to modify state on the target server. Otherwise, the meaning of POST is a bit relaxed as you can see in RFC 7231, sec. 4.3.3.
Since the intent of the requests is to create a new version of a resource, they have failed if a version with the given presentation already exists. This would disqualify any 2xx-class response codes. From section 6.3:
The 2xx (Successful) class of status code indicates that the client's request was successfully received, understood, and accepted.
If you absolutely wanted to, you could go for 202/Accepted, which "is intentionally noncommittal." This is a bit of a stretch, though, as this status code is intended for queued processing. I would advise against it.
The 204/No Content code suggested by others is a bit of a poor choice. It were absolutely correct if you POSTed to the resource you were updating, though.
As the result is neither informational (1xx) nor a fault by the server (5xx). Let us have a look at the 3xx class first. From section 6.4:
The 3xx (Redirection) class of status code indicates that further action needs to be taken by the user agent in order to fulfill the request.
One of the most prominent one here would be 304/Not Modified. While sounding like a perfect fit, this code is unfortunately not applicable here. It can only be returned in response to conditional GET or HEAD requests.
302/Found may sound like the next best fit. However, this code is intended for temporary redirects, which is in all likelyhood not what you want.
As has been suggested here, 303/See Other is indeed a good choice:
The 303 (See Other) status code indicates that the server is redirecting the user agent to a different resource [...] which is intended to provide an indirect response to the original request.
[...]
This status code is applicable to any HTTP method. It is primarily used to allow the output of a POST action to redirect the user agent to a selected resource
All other 3xx codes are dealing with various forms of redirects that hardly relate to the situation here.
A final look, 4xx-class of status codes. From RFC 7231, sec. 6.5:
The 4xx (Client Error) class of status code indicates that the client seems to have erred. Except when responding to a HEAD request, the server SHOULD send a representation containing an explanation of the error situation, and whether it is a temporary or permanent condition. These status codes are applicable to any request method.
Very few of these are actually deeling with the request body. Two of those who do would stand out here: One is 400/Bad Request, which is by design overly broad. It is - if you will - a catch-all solution. However, this would imply that the request body is malformed (as in syntactically incorrect) in some way, which is probably not the case.
More interesting is 409/Conflict. From the RFC (emphasis mine):
The 409 (Conflict) status code indicates that the request could not be completed due to a conflict with the current state of the target resource.
The wording of the definition places this code close to the PUT method, but is not exclusive. To reiterate the definition of the 4xx codes:
These status codes are applicable to any request method.
422/Unprocessable Entity is a contender, but it implies a semantic error, which really isn't the case here.
Ultimately (drumroll) the final piece of the puzzle could be found in section 4.3.3:
If the result of processing a POST would be equivalent to a representation of an existing resource, an origin server MAY redirect the user agent to that resource by sending a 303 (See Other) response with the existing resource's identifier in the Location field.
Note the "MAY." So you can really choose between 303 and 409. I feel 409 were the better fit, as clearly an earlier request introduced a state that is incompatible with the current one. OTOH, 303 may be the politer way to go and is closer to the standard. Either way, as a consumer of your API, I would really like to know if my requests failed. And be it by not having any effect whatsoever.
If nothing has been created by the operation, 201 is not suitable for that:
6.3.2. 201 Created
The 201 (Created) status code indicates that the request has been fulfilled and has resulted in one or more new resources being created. [...]
See below some options you could consider if the operation succeeds:
6.3.1. 200 OK
The 200 (OK) status code indicates that the request has succeeded. The payload sent in a 200 response depends on the request method. For the methods defined by this specification, the intended meaning of the payload can be summarized as:
[...]
POST: a representation of the status of, or results obtained from, the action;
[...]
Aside from responses to CONNECT, a 200 response always has a payload, though an origin server MAY generate a payload body of zero length. If no payload is desired, an origin server ought to send 204 (No Content) instead. [...]
6.3.5. 204 No Content
The 204 (No Content) status code indicates that the server has successfully fulfilled the request and that there is no additional content to send in the response payload body. [...]
If the operation fails:
6.5.1. 400 Bad Request
The 400 (Bad Request) status code indicates that the server cannot will not process the request due to something that is perceived to be a client error (e.g., malformed request syntax, invalid request message framing, or deceptive request routing).
11.2. 422 Unprocessable Entity
The 422 (Unprocessable Entity) status code means the server understands the content type of the request entity (hence a 415 (Unsupported Media Type) status code is inappropriate), and the syntax of the request entity is correct (thus a 400 (Bad Request) status code is inappropriate) but was unable to process the contained instructions. For example, this error condition may occur if an XML request body contains well-formed (i.e., syntactically correct), but semantically erroneous, XML instructions.
A 201 Created should be used whenever you creating a new resource without doubt.
As defined in HTTP Method Definitions RFC, either200 Ok or 204 No Contentis an appropriate response if the operation does not create a new resource depending on the response body content.
The action performed by the POST method might not result in a resource
that can be identified by a URI. In this case, either 200 (OK) or 204
(No Content) is the appropriate response status, depending on whether
or not the response includes an entity that describes the result.
If a resource has been created on the origin server, the response
SHOULD be 201 (Created) and contain an entity which describes the
status of the request and refers to the new resource, and a Location
header (see section 14.30).
Now, coming back to your original question about what to use when the operation is successful and there is nothing to return, you should use 204 No Content. This status code is specifically meant for scenarios where the requested operation is successfully completed but there is no additional relevant information that the server can provide.
The 204 (No Content) status code indicates that the server has successfully fulfilled the request and that there is no additional content to send in the response payload body.
I don't think that for this case the idempotency is an issue, because the state of the system is not the same as it was in the inicial request, because the entity now exists, so you can respond with a different code.
200 should be fine, but it is a little weird as you said.
I have never use this, but I read that for some case you should redirect with a 302, to make a get for other resource, in this case I think this apply, return a 302 and make a get pointing for the old semver, assuming that you have a get endpoint for this entity.
If the POST'd resource has the same version number, but different data, then a 409 would be fitting. But if the data is the same as what's already stored, then a 409 might not be required. Through process of elimination I would say 200 OK would be fine.
We could just return a 200, but then that would seem weird, and increases the response codes that the users have to deal with
If this is a costly concern, consider eliminating the 201, not the 200. The fact of the matter is that for any decently complex service there may be at some point a situation where a 20X (where X is not 0) is applicable. So does that mean we code with each of the 20X responses in mind and spend time checking if our services has situations where 20X is preferred over 200? I suspect not. So unless there is a specific reason to respond with a 20X, for example to deal with a specific use case, then just use 200 and reduce the amount of coding and documenting required. I suspect for most scenarios, the calling client does not care.
Ultimately, the correct answer probably depends on whatever client is consuming your API. If you are building the client too, you could do whatever you prefer... (Just don't get too crazy.)
Assuming you are writing the API and client:
My opinion/suggestion is...
If there IS a new version number: The 201 HTTP status code would fit will.
If there is NOT a new version number: The 200 or 204 HTTP status code would fit well.
If there is no benefit to the client knowing the version number has changed or is the same: Send the 200 HTTP status code.
If you don't control the client consuming your API: Obviously defer to what they expect.
You may also wish review all of the HTTP Status Codes in the HTTP RFC spec. The above status codes also link directly to the relevant section.
201 : when new version is created
202 : when existing version is updated
204 : when request is accepted but no processing is done
by def, 204 = No Content
The server has fulfilled the request but does not need to return an
entity-body, and might want to return updated metainformation. The
response MAY include new or updated metainformation in the form of
entity-headers, which if present SHOULD be associated with the
requested variant.
If the client is a user agent, it SHOULD NOT change its document view
from that which caused the request to be sent. This response is
primarily intended to allow input for actions to take place without
causing a change to the user agent's active document view, although
any new or updated metainformation SHOULD be applied to the document
currently in the user agent's active view.
The 204 response MUST NOT include a message-body, and thus is always
terminated by the first empty line after the header fields.
So its slight tangential to your needs but I think its the best fit.
I am looking into the Swift Vapor framework.
I am trying to create a controller class that maps data obtained on an SSL link to a third party system (an Asterisk PBX server..) into a response body that is sent over some time down to the client.
So I need to send received text lines (obtained separately on the SSL connection) as they get in, without waiting for a 'complete response' to be constructed.
Seeing this example:
return Response(status: .ok) { chunker in
for name in ["joe\n", "pam\n", "cheryl\n"] {
sleep(1)
try chunker.send(name)
}
try chunker.close()
}
I thought it might be the way to go.
But what I see connecting to the Vapor server is that the REST call waits for the loop to complete, before the three lines are received as result.
How can I obtain to have try chunker.send(name) send it's characters back the client without first waiting for the loop to complete?
In the real code the controller method can potentially keep an HTTP connection to the client open for a long time, sending Asterisk activity data to the client as soon as it is obtained. So each .send(name) should actually pass immediately data to the client, not waiting for the final .close() call.
Adding a try chunker.flush() did not produce any better result..
HTTP requests aren't really designed to work like that. Different browsers and clients will function differently depending on their implementations.
For instance, if you connect with telnet to the chunker example you pasted, you will see the data is sent every second. But Safari on the other hand will wait for the entire response before displaying.
If you want to send chunked data like this reliably, you should use a protocol like WebSockets that is designed for it.
I have a REST API where the GET request can take 10-20 seconds. So I usually return a 202 code with a location like http://fakeserver/pending/blah where the client can check the status of this request. pending/blah returns a 200 code with "Status: pending" if the request is still pending, and a 303 code when it's done, with a final location for the result: http://fakeserver/finished/blah .
But what if the server crashes during the request processing? Should pending/blah return a 303 code, and then finished/blah returns a 404? How can I alert the client that the resource may be available at a location, but I'm not sure? Assume the requests are persistent, so that when the server reboots, it continues processing the request.
First of all I'll make the state of processed resource an internal field of this resource. This way you can avoid using strange endpoints like: /finished/blah/ or /pending/blah/ and instead of it introduce a single endpoint /resources/blah/ which will among other fields return the state it's currently in.
After changing architecture to the endpoint mentioned above if you ask for blah and server has crashed you can:
return 200 with pending status - client doesn't have necessarily to know about the crash
return 404, simple not found with and extra message that server has crashed.
return 500 and inform the client explicitly what the problem is.
Other useful codes may be also 409 or 503. Returning any 3XX is not a good idea IMO since no redirection applies here. Personally I'd go for 200 or 500(3).
When a client invokes my REST-ful service, it needs to know if the response came back was 'from me' or rather a diagnosis from the containing web server that something awful happened.
One theory is that, if my code is called, it should always return an HTTP OK(=200), and any errors I've got to return should be just represented in the data I return. After all, it's my code that gets the response, not the naked browser.
Somewhat self-evidently, if I'm using REST to generate HTML read directly by a browser, I absolutely must return an error code if there's an error. In the case I care about, it's always Javascript or Java that is interpreting the entrails of the response.
Another possibility is that there is some family of HTTP status codes that I could return with a high confidence that it/they would never be generated by a problem in the surrounding container. Is this the case?
I use the following:
GET
200 OK
400 Bad Request (when input criteria not correct)
POST
202 Accepted (returned by authorization method)
401 Unauthorized (also returned by authorization)
201 Created (when creating a new resource; I also set the location header)
400 Bad Request (when data for creating new entity is invalid or transaction rollback)
PUT
Same as POST
201 Ok
400 Bad Request
DELETE
200 OK
404 Not Found (same as GET)
I would not know how to avoid that some container returns codes like 404.
4xx codes are meant to handle client errors along with possibly some entity that describes the problem in detail (and thus would mean a combination of both of your mentioned approaches). Since REST relies on HTTP and the according semantics of status as well as methods, always returning 200 in any possible case is a violation of this principle in my opinion.
If you for instance have a request such as http://foo.com/bar/123 which represents a bar ressource with id=123 and you return 200 with some content, the client has no chance to figure out if this was the intended response or some sort of error that occured. Therefore one should try to map error conditions to status codes as discussed in REST: Mapping application errors to HTTP Status codes for example.