A toolkit to create code-first HTTP Reverse Proxies hosted in ASP.NET Core as middleware. This allows focused code-first proxies that can be embedded in existing ASP.NET Core applications or deployed as a standalone server. Deployable anywhere ASP.NET Core is deployable such as Windows, Linux, Containers and Serverless (with caveats).
Having built proxies many times before, I felt it is time to make a package. Forked
from ASP.NET labs, it has been heavily modified with a different
API, to facilitate a wider variety of proxying scenarios (i.e. routing based on
a JWT claim) and interception of the proxy requests / responses for
customization of headers and (optionally) request / response bodies. It also
uses HttpClientFactory
internally that will mitigate against DNS caching
issues making it suitable for microservice / container environments.
- 1. Quick Start
- 2. Core Features
- 3. Recipes
- 4. Making upstream servers reverse proxy friendly
- 5. Performance considerations
- 6. Note about serverless
- 7. Comparison with Ocelot
- 8. How to build
- 9. Contributing / Feedback / Questions
- 10. Articles, blogs and other external links
ProxyKit is a NetStandard2.0
package. Install into your ASP.NET Core project:
dotnet add package ProxyKit
In your Startup
, add the proxy service:
public void ConfigureServices(IServiceCollection services)
{
...
services.AddProxy();
...
}
Forward HTTP requests to upstream-server:5001
:
public void Configure(IApplicationBuilder app)
{
app.RunProxy(context => context
.ForwardTo("http://upstream-server:5001/")
.AddXForwardedHeaders()
.Send());
}
What is happening here?
context.ForwardTo(upstreamHost)
is an extension method onHttpContext
that creates and initializes anHttpRequestMessage
with the original request headers copied over, yielding aForwardContext
.AddXForwardedHeaders
addsX-Forwarded-For
,X-Forwarded-Host
,X-Forwarded-Proto
andX-Forwarded-PathBase
headers to the upstream request.Send
Sends the forward request to the upstream server and returns anHttpResponseMessage
.- The proxy middleware then takes the response and applies it to
HttpContext.Response
.
Note: RunProxy
is terminal - anything added to the pipeline after
RunProxy
will never be executed.
Forward WebSocket requests to upstream-server:5002
:
public void Configure(IApplicationBuilder app)
{
app.UseWebSockets();
app.UseWebSocketProxy(
context => new Uri("ws://upstream-host:80/"),
options => options.AddXForwardedHeaders());
}
What is happening here?
app.UseWebSockets()
must first be added otherwise websocket requests will never be handled by ProxyKit.- The first parameter must return the URI of the upstream host with a scheme
of
ws://
. - The second parameter
options
allows you to do some customisation of the initial upstream requests such as adding some headers.
One can modify the upstream request headers prior to sending them to suit customisation needs. ProxyKit doesn't add, remove, nor modify any headers by default; one must opt in any behaviours explicitly.
In this example we will add a X-Correlation-ID
header if the incoming request does not bear one:
public const string XCorrelationId = "X-Correlation-ID";
public void Configure(IApplicationBuilder app)
{
app.RunProxy(context =>
{
var forwardContext = context.ForwardTo("http://upstream-server:5001/");
if (!forwardContext.UpstreamRequest.Headers.Contains(XCorrelationId))
{
forwardContext.UpstreamRequest.Headers.Add(XCorrelationId, Guid.NewGuid().ToString());
}
return forwardContext.Send();
});
}
This can be encapsulated as an extension method:
public static class CorrelationIdExtensions
{
public const string XCorrelationId = "X-Correlation-ID";
public static ForwardContext ApplyCorrelationId(this ForwardContext forwardContext)
{
if (!forwardContext.UpstreamRequest.Headers.Contains(XCorrelationId))
{
forwardContext.UpstreamRequest.Headers.Add(XCorrelationId, Guid.NewGuid().ToString());
}
return forwardContext;
}
}
... making the proxy code a little nicer to read:
public void Configure(IApplicationBuilder app)
{
app.RunProxy(context => context
.ForwardTo("http://upstream-server:5001/")
.ApplyCorrelationId()
.Send());
}
The response from an upstream server can be modified before it is sent to the client. In this example we are removing a header:
public void Configure(IApplicationBuilder app)
{
app.RunProxy(async context =>
{
var response = await context
.ForwardTo("http://localhost:5001/")
.Send();
response.Headers.Remove("MachineID");
return response;
});
}
X-Forward-*
headers from the incoming request to the upstream
request by default. Copying them requires opting in; see 2.3.3 Copying
X-Forwarded headers below.
Many applications will need to know what their "outside" host / URL is in order
to generate correct values. This is achieved using X-Forwarded-*
and
Forwarded
headers. ProxyKit supports applying X-Forward-*
headers out of the
box (applying Forwarded
headers support is on backlog). At the time of writing,
Forwarded
is not supported
in ASP.NET Core.
To add X-Forwarded-*
headers to the request to the upstream server:
public void Configure(IApplicationBuilder app)
{
app.RunProxy(context => context
.ForwardTo("http://upstream-server:5001/")
.AddXForwardedHeaders()
.Send());
}
This will add X-Forwarded-For
, X-Forwarded-Host
and X-Forwarded-Proto
headers to the upstream request using values from HttpContext
. If the proxy
middleware is hosted on a path and a PathBase
exists on the request, then an
X-Forwarded-PathBase
is also added.
Chaining proxies is a common pattern in more complex setups. In this case, if
the proxy is an "internal" proxy, you will want to copy the "X-Forwarded-*"
headers from previous proxy. To do so, use CopyXForwardedHeaders()
:
public void Configure(IApplicationBuilder app)
{
app.RunProxy(context => context
.ForwardTo("http://upstream-server:5001/")
.CopyXForwardedHeaders()
.Send());
}
You may optionally also add the "internal" proxy details to the X-Forwarded-*
header values by combining CopyXForwardedHeaders()
and
AddXForwardedHeaders()
(note the order is important):
public void Configure(IApplicationBuilder app)
{
app.RunProxy(context => context
.ForwardTo("http://upstream-server:5001/")
.CopyXForwardedHeaders()
.AddXForwardedHeaders()
.Send());
}
When adding the Proxy to your application's service collection, there is an
opportunity to configure the internal HttpClient. As
HttpClientFactory
is used, its builder is exposed for you to configure:
services.AddProxy(httpClientBuilder => /* configure http client builder */);
Below are two examples of what you might want to do:
-
Configure the HTTP Client's timeout to 5 seconds:
services.AddProxy(httpClientBuilder => httpClientBuilder.ConfigureHttpClient = client => client.Timeout = TimeSpan.FromSeconds(5));
-
Configure the primary
HttpMessageHandler
. This is typically used in testing to inject a test handler (see Testing below).services.AddProxy(httpClientBuilder => httpClientBuilder.ConfigurePrimaryHttpMessageHandler = () => _testMessageHandler);
When HttpClient
throws, the following logic applies:
- When upstream server is not reachable, then
503 ServiceUnavailable
is returned. - When upstream server is slow and client timeouts, then
504 GatewayTimeout
is returned.
Not all exception scenarios and variations are caught, which may result in a
InternalServerError
being returned to your clients. Please create an issue if
a scenario is missing.
As ProxyKit is a standard ASP.NET Core middleware, it can be tested using the
standard in-memory TestServer
mechanism.
Often you will want to test ProxyKit with your application and perhaps test the behaviour of your application when load balanced with two or more instances as indicated below.
+----------+
|"Outside" |
|HttpClient|
+-----+----+
|
|
|
+-----------+---------+
+-------------------->RoutingMessageHandler|
| +-----------+---------+
| |
| |
| +--------------------+-------------------------+
| | | |
+---+-----------v----+ +--------v---------+ +---------v--------+
|Proxy TestServer | |Host1 TestServer | |Host2 TestServer |
|with Routing Handler| |HttpMessageHandler| |HttpMessageHandler|
+--------------------+ +------------------+ +------------------+
RoutingMessageHandler
is an HttpMessageHandler
that will route requests
to specific hosts based on the origin it is configured with. For ProxyKit
to forward requests (in memory) to the upstream hosts, it needs to be configured
to use the RoutingMessageHandler
as its primary HttpMessageHandler
.
Full example can been viewed in Recipe 6.
Load balancing is a mechanism to decide which upstream server to forward the request to. Out of the box, ProxyKit currently supports one type of load balancing - Weighted Round Robin. Other types are planned.
Round Robin simply distributes requests as they arrive to the next host in a distribution list. With optional weighting, more requests are sent to the host with the greater weight.
public void Configure(IApplicationBuilder app)
{
var roundRobin = new RoundRobin
{
new UpstreamHost("http://localhost:5001/", weight: 1),
new UpstreamHost("http://localhost:5002/", weight: 2)
};
app.RunProxy(
async context =>
{
var host = roundRobin.Next();
return await context
.ForwardTo(host)
.Send();
});
}
New in version 2.1.0
Instead of specifying a delegate, it is possible to use a typed handler. The reason you may want to do this is when you want to better leverage dependency injection.
Typed handlers must implement IProxyHandler
that has a single method with same
signature as HandleProxyRequest
. In this example our typed handler has a
dependency on an imaginary service to lookup hosts:
public class MyTypedHandler : IProxyHandler
{
private IUpstreamHostLookup _upstreamHostLookup;
public MyTypeHandler(IUpstreamHostLookup upstreamHostLookup)
{
_upstreamHostLookup = upstreamHostLookup;
}
public Task<HttpResponseMessage> HandleProxyRequest(HttpContext context)
{
var upstreamHost = _upstreamHostLookup.Find(context);
return context
.ForwardTo(upstreamHost)
.AddXForwardedHeaders()
.Send();
}
}
We then need to register our typed handler service:
public void ConfigureServices(IServiceCollection services)
{
...
services.AddSingleton<MyTypedHandler>();
...
}
When adding the proxy to the pipeline, use the generic form:
public void ConfigureServices(IServiceCollection services)
{
...
appInner.RunProxy<MyTypedHandler>());
...
}
Recipes have moved to own repo.
Applications that are deployed behind a reverse proxy typically need to be
somewhat aware of that so they can generate correct URLs and paths when
responding to a browser. That is, they look at X-Forward-*
/ Forwarded
headers and use their values.
In ASP.NET Core, this means using the ForwardedHeaders
middleware in your
application. Please refer to the documentation
for correct usage (and note the security advisory!).
Note: the Forwarded Headers middleware does not support
X-Forwarded-PathBase
. This means if you proxy http://example.com/foo/
to
http://upstream-host/
the /foo/
part is lost and absolute URLs cannot be
generated unless you configure your application's PathBase
directly.
Related issues and discussions:
To support PathBase dynamically in your application with X-Forwarded-PathBase
,
examine the header early in your pipeline and set the PathBase
accordingly:
var options = new ForwardedHeadersOptions
{
...
};
app.UseForwardedHeaders(options);
app.Use((context, next) =>
{
if (context.Request.Headers.TryGetValue("X-Forwarded-PathBase", out var pathBases))
{
context.Request.PathBase = pathBases.First();
}
return next();
});
Alternatively you can use ProxyKit's UseXForwardedHeaders
extension that
performs the same as the above (including calling UseForwardedHeaders
):
var options = new ForwardedHeadersOptions
{
...
};
app.UseXForwardedHeaders(options);
According to TechEmpower's Web Framework Benchmarks, ASP.NET Core is up there with the fastest for plain text. As ProxyKit simply captures headers and async copies request and response body streams, it will be fast enough for most scenarios.
If absolute raw throughput is a concern for you, then consider nginx or alternatives. For me being able to create flexible proxies using C# is a reasonable tradeoff for the (small) performance cost. Note that what your specific proxy (and its specific configuration) does will impact performance so you should measure for yourself in your context.
On Windows, ProxyKit is ~3x faster than nginx. However, nginx has clearly documented that it has known performance issues on Windows. Since one wouldn't be running production nginx on Windows, this comparison is academic.
Memory wise, ProxyKit maintained a steady ~20MB of RAM after processing millions of requests for simple forwarding. Again, it depends on what your proxy does so you should analyse and measure yourself.
Whilst it is possible to run full ASP.NET Core web application in AWS Lambda and Azure Functions it should be noted that Serverless systems are message based and not stream based. Incoming and outgoing HTTP request messages will be buffered and potentially encoded as Base64 if binary (so larger). This means ProxyKit should only be used for API (json) proxying in production on Serverless. (Though proxying other payloads is fine for dev / exploration / quick'n'dirty purposes.)
Ocelot is an API Gateway that also runs on ASP.NET Core. A key difference between API Gateways and general Reverse Proxies is that the former tend to be message based whereas a reverse proxy is stream based. That is, an API Gateway will typically buffer every request and response message to be able to perform transformations. This is fine for an API Gateway but not suitable for a general reverse proxy performance wise nor for responses that are chunked-encoded. See Not Supported Ocelot docs.
Combining ProxyKit with Ocelot would give some nice options for a variety of scenarios.
Requirements: .NET Core SDK 2.2.100 or later.
On Windows:
.\build.cmd
On Linux:
./build.sh
Any ideas for features, bugs or questions, please create an issue. Pull requests gratefully accepted but please create an issue for discussion first.
I can be reached on twitter at @randompunter
logo is distribute by ChangHoon Baek from the Noun Project.