Nancy perform AfterRequest action after OnError - nancy

Question

I want to add some request timings to the response headers of a request to a Nancy module. I've added some before/after request handlers into the RequestStartup and added the headers no problem (example below) and all was good. I've also added an OnError handler to the ApplicationStartup, to catch errors and return a nice formatted Json response.
pipelines.BeforeRequest += ctx =>
{
ctx.Items.Add("X-RequestStart", DateTime.Now.ToString());
return null;
};
pipelines.AfterRequest += ctx =>
{
var now = DateTime.Now;
try
{
//Not got around to forcing the culture on the datetime parsing yet...
var startTime = DateTime.Parse(ctx.Items["X-RequestStart"].ToString());
ctx.Response.Headers.Add("X-RequestStart", startTime.ToString(CultureInfo.InvariantCulture));
ctx.Response.Headers.Add("X-RequestComplete", now.ToString(CultureInfo.InvariantCulture));
ctx.Response.Headers.Add("X-RequestDuration", (now - startTime).ToString());
}
catch (Exception)
{
ctx.Response.Headers.Add("X-RequestComplete", now.ToString(CultureInfo.InvariantCulture));
}
};
pipelines.OnError += (ctx, exception) =>
{
return ErrorResponse.FromException(exception);
};
What I am noticing however, is that when I have an error thrown, the AfterRequest action is not performed - thus I have no timing headers in to the error response. I've tried moving the before/after request handling to the application startup, but this has no effect either.
The question is in two parts really, firstly, is it possible to get the framework to perform an AfterRequest action after the OnError action has been performed, or is the pipeline set up in a way to prevent this, and secondly, should these before/after request actions be part of the RequestStartup or ApplicationStartup? ApplicationStartup seemed sensible for error handling, whereas RequestStartup seemed sensible for interacting with the response headers as it should be on a per request basis, but I'm not sure if there is a convention for this, or if my assumptions were incorrect.

Answer 1

Unfortunately, this does not seem possible in NancyFx. I took a detailed look at the source code, specifically DefaultRequestDispatcher. Dispatch catches any exceptions thrown during route processing, calls ResolveErrorResult, then gets the response from the negotiator. There does not appear to be an extensibility point to modify responses generated in this way.
In my opinion, this is an oversight that the developers should consider addressing.
public async Task<Response> Dispatch(NancyContext context, CancellationToken cancellationToken)
{
// TODO - May need to make this run off context rather than response .. seems a bit icky currently
var resolveResult = this.Resolve(context);
context.Parameters = resolveResult.Parameters;
context.ResolvedRoute = resolveResult.Route;
try
{
context.Response = await ExecuteRoutePreReq(context, cancellationToken, resolveResult.Before)
.ConfigureAwait(false);
if(context.Response == null)
{
context.Response = await this.routeInvoker.Invoke(resolveResult.Route, cancellationToken,
resolveResult.Parameters, context)
.ConfigureAwait(false);
if (context.Request.Method.Equals("HEAD", StringComparison.OrdinalIgnoreCase))
{
context.Response = new HeadResponse(context.Response);
}
}
await this.ExecutePost(context, cancellationToken, resolveResult.After, resolveResult.OnError)
.ConfigureAwait(false);
}
catch(Exception ex)
{
context.Response = this.ResolveErrorResult(context, resolveResult.OnError, ex);
if (context.Response == null)
{
throw;
}
}
return context.Response;
}
private Response ResolveErrorResult(NancyContext context, Func<NancyContext, Exception, dynamic> resolveResultOnError, Exception exception)
{
if (resolveResultOnError != null)
{
var flattenedException = exception.FlattenInnerExceptions();
var result = resolveResultOnError.Invoke(context, flattenedException);
if (result != null)
{
return this.negotiator.NegotiateResponse(result, context);
}
}
return null;
}