Rejections

TODO update to Java APIs

In the chapter about constructing Routes the ~ operator was introduced, which connects two routes in a way that allows a second route to get a go at a request if the first route “rejected” it. The concept of “rejections” is used by Akka HTTP for maintaining a more functional overall architecture and in order to be able to properly handle all kinds of error scenarios.

When a filtering directive, like the get directive, cannot let the request pass through to its inner route because the filter condition is not satisfied (e.g. because the incoming request is not a GET request) the directive doesn’t immediately complete the request with an error response. Doing so would make it impossible for other routes chained in after the failing filter to get a chance to handle the request. Rather, failing filters “reject” the request in the same way as by explicitly calling requestContext.reject(...).

After having been rejected by a route the request will continue to flow through the routing structure and possibly find another route that can complete it. If there are more rejections all of them will be picked up and collected.

If the request cannot be completed by (a branch of) the route structure an enclosing handleRejections directive can be used to convert a set of rejections into an HttpResponse (which, in most cases, will be an error response). Route.seal internally wraps its argument route with the handleRejections directive in order to “catch” and handle any rejection.

Predefined Rejections

A rejection encapsulates a specific reason why a route was not able to handle a request. It is modeled as an object of type Rejection. Akka HTTP comes with a set of predefined rejections, which are used by the many predefined directives.

Rejections are gathered up over the course of a Route evaluation and finally converted to HttpResponse replies by the handleRejections directive if there was no way for the request to be completed.

The RejectionHandler

The handleRejections directive delegates the actual job of converting a list of rejections to its argument, a RejectionHandler, which is defined like this:

trait RejectionHandler extends (immutable.Seq[Rejection] => Option[Route])

Since a RejectionHandler returns an Option[Route] it can choose whether it would like to handle the current set of rejections or not. If it returns None the rejections will simply continue to flow through the route structure.

The default RejectionHandler applied by the top-level glue code that turns a Route into a Flow or async handler function for the low-level API (via Route.handlerFlow or Route.asyncHandler) will handle all rejections that reach it.

Rejection Cancellation

As you can see from its definition above the RejectionHandler doesn’t handle single rejections but a whole list of them. This is because some route structure produce several “reasons” why a request could not be handled.

Take this route structure for example:

TODO missing sample

For uncompressed POST requests this route structure would initially yield two rejections:

  • a MethodRejection produced by the get directive (which rejected because the request is not a GET request)
  • an UnsupportedRequestEncodingRejection produced by the decodeRequestWith directive (which only accepts gzip-compressed requests here)

In reality the route even generates one more rejection, a TransformationRejection produced by the post directive. It “cancels” all other potentially existing MethodRejections, since they are invalid after the post directive allowed the request to pass (after all, the route structure can deal with POST requests). These types of rejection cancellations are resolved before a RejectionHandler sees the rejection list. So, for the example above the RejectionHandler will be presented with only a single-element rejection list, containing nothing but the UnsupportedRequestEncodingRejection.

Empty Rejections

Since rejections are passed around in a list (or rather immutable Seq) you might ask yourself what the semantics of an empty rejection list are. In fact, empty rejection lists have well defined semantics. They signal that a request was not handled because the respective resource could not be found. Akka HTTP reserves the special status of “empty rejection” to this most common failure a service is likely to produce.

So, for example, if the path directive rejects a request it does so with an empty rejection list. The host directive behaves in the same way.

Customizing Rejection Handling

If you’d like to customize the way certain rejections are handled you’ll have to write a custom RejectionHandler. Here is an example:

TODO missing sample

The easiest way to construct a RejectionHandler is via the RejectionHandler.Builder that Akka HTTP provides. After having created a new Builder instance with RejectionHandler.newBuilder() you can attach handling logic for certain types of rejections through three helper methods:

handle
Handles certain rejections with the given partial function. The partial function simply produces a Route which is run when the rejection is “caught”. This makes the full power of the Routing DSL available for defining rejection handlers and even allows for recursing back into the main route structure if required.
handleAll[T <: Rejection]
Handles all rejections of a certain type at the same time. This is useful for cases where your need access to more than the first rejection of a certain type, e.g. for producing the error message to an unsupported request method.
handleNotFound
As described above “Resource Not Found” is special as it is represented with an empty rejection set. The handleNotFound helper let’s you specify the “recovery route” for this case.

Even though you could handle several different rejection types in a single partial function supplied to handle it is recommended to split these up into distinct handle attachments instead. This way the priority between rejections is properly defined via the order of your handle clauses rather than the (sometimes hard to predict or control) order of rejections in the rejection set.

Once you have defined your custom RejectionHandler you have two options for “activating” it:

  1. Pass it to the seal() method of the Route class
  2. Supply it as an argument to the handleRejections directive

In the first case your handler will be “sealed” (which means that it will receive the default handler as a fallback for all cases your handler doesn’t handle itself) and used for all rejections that are not handled within the route structure itself.

The second case allows you to restrict the applicability of your handler to certain branches of your route structure.