proxy

Overview ¶

Package proxy implements an HTTP reverse proxy based on continuously updated skipper routing rules.

The proxy matches each incoming request to the lookup tree for the first matching route, and handles it accordingly to the rules defined in it. This typically means augmenting the request with the filters and forwarding it to the route endpoint. It may also mean to handle the request internally if it is a 'shunt' route or to continue routing further in case of 'loopback'.

Proxy Mechanism ¶

1. route matching:

The incoming request is matched to the current routing tree, implemented in skipper/routing. The result may be a route, which will be used for forwarding or handling the request, or nil, in which case the proxy responds with a configured http status code (defaults to 404).

2. upstream request augmentation:

In case of a matched route, the request handling method of all filters in the route will be executed in the order they are defined. The filters share a context object, that provides the in-memory representation of the incoming request, the outgoing response writer, the path parameters derived from the actual request path (see skipper/routing) and a free-form state bag. The filters may modify the request or pass data to each other using the state bag.

Filters can break the filter chain, serving their own response object. This will prevent the request from reaching the route endpoint. The filters that are defined in the route after the one that broke the chain will never handle the request.

3.a upstream request:

The incoming and augmented request is mapped to an outgoing request and executed, addressing the endpoint defined by the current route.

If a filter chain was broken by some filter this step is skipped.

3.b shunt:

In case the route is a 'shunt', an empty response is created with default 404 status.

3.c loopback:

In case of the `loopback` route no upstream request is made. Routing will happen again with an augmented request. This can happen several times in a row until maxLoopbacks limit is reached.

3.d dynamic:

In case of a `dynamic` route, the final target must be defined in a filter.

4. downstream response augmentation:

The response handling method of all the filters processed in step 2 will be executed in reverse order. The filter context is the same instance as the one in step 2. It will include the response object from step 3, or the one provided by the filter that broke the chain.

If the route is a shunt route, one of the filters needs to handle the request latest in this phase. It should set the status and response headers and write the response body, if any, to the writer in the filter context.

5. response:

In case none of the filters handled the request, the response properties, including the status and the headers, are mapped to the outgoing response writer, and the response body is streamed to it, with continuous flushing.

Routing Rules ¶

The route matching is implemented in the skipper/routing package. The routing rules are not static, but they can be continuously updated by new definitions originated in one or more data sources.

The only exceptions are the priority routes, that have not originated from the external data sources, and are tested against the requests before the general routing tree.

Handling the Host header ¶

The default behavior regarding the 'Host' header of the proxy requests is that the proxy ignores the value set in the incoming request. This can be changed setting the `OptionsProxyPreserveHost` init flag.

In either case, it is possible to override this behavior individually for each route with the `preserveHost` filter. If the filter argument is "true", the request host is used, if it is "false", the backend host is used, regardless of the global setting.

If a filter sets the 'Host' header to a value other than the value in the incoming request or the backend host, this custom value will be used instead of the global setting or the route specific override.

To control the value of the outgoing 'Host' header, the `OutgoingHost()` and `SetOutgoingHost()` methods of the `FilterContext` need to be used instead of the `Request.Header` map.

Circuit Breakers ¶

When configured, skipper can use circuit breakers for the backend requests. It asks the registry for a matching circuit breaker for every request, and if there is any, checks if it is closed before sending out the backend request. On connection errors and responses with a status code higher than 499, it reports a failure to the current breaker, otherwise it reports a success.

For details, see: https://godoc.org/github.com/zalando/skipper/circuit.

Proxy Example ¶

The below example demonstrates creating a routing proxy as a standard http.Handler interface:

// create a target backend server. It will return the value of the 'X-Echo' request header
// as the response body:
targetServer := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
	w.Write([]byte(r.Header.Get("X-Echo")))
}))

defer targetServer.Close()

// create a filter registry, and register the custom filter:
filterRegistry := builtin.MakeRegistry()
filterRegistry.Register(&setEchoHeader{})

// create a data client with a predefined route, referencing the filter and a path condition
// containing a wildcard called 'echo':
routeDoc := fmt.Sprintf(`Path("/return/:echo") -> setEchoHeader() -> "%s"`, targetServer.URL)
dataClient, err := testdataclient.NewDoc(routeDoc)
if err != nil {
	log.Fatal(err)
}

// create routing object:
rt := routing.New(routing.Options{
	FilterRegistry: filterRegistry,
	DataClients:    []routing.DataClient{dataClient}})
defer rt.Close()

// create a proxy instance, and start an http server:
proxy := proxy.New(rt, proxy.OptionsNone)
defer proxy.Close()

router := httptest.NewServer(proxy)
defer router.Close()

// make a request to the proxy:
rsp, err := http.Get(fmt.Sprintf("%s/return/Hello,+world!", router.URL))
if err != nil {
	log.Fatal(err)
}

defer rsp.Body.Close()

// print out the response:
if _, err := io.Copy(os.Stdout, rsp.Body); err != nil {
	log.Fatal(err)
}

// Output:
// Hello, world!