Skip to content

Development

Local Setup

Build Skipper Binary

Clone repository and compile with Go.

git clone https://github.com/zalando/skipper.git
cd skipper
make skipper

binary will be ./bin/skipper

Run Skipper as Proxy with 2 backends

As a small example, we show how you can run one proxy skipper and 2 backend skippers.

Start the proxy that listens on port 9999 and serves all requests with a single route, that proxies to two backends using the round robin algorithm:

./bin/skipper -inline-routes='r1: * -> <roundRobin, "http://127.0.0.1:9001", "http://127.0.0.1:9002">' --address :9999

Start two backends, with similar routes, one responds with “1” and the other with “2” in the HTTP response body:

./bin/skipper -inline-routes='r1: * -> inlineContent("1") -> <shunt>' --address :9001 &
./bin/skipper -inline-routes='r1: * -> inlineContent("2") -> <shunt>' --address :9002

Test the proxy with curl as a client:

curl -s http://localhost:9999/foo
1
curl -s http://localhost:9999/foo
2
curl -s http://localhost:9999/foo
1
curl -s http://localhost:9999/foo
2

Debugging Skipper

It can be helpful to run Skipper in a debug session locally that enables one to inspect variables and do other debugging activities in order to analyze filter and token states.

For Visual Studion Code users, a simple setup could be to create following launch configuration that compiles Skipper, runs it in a Delve debug session, and then opens the default web browser creating the request. By setting a breakpoint, you can inspect the state of the filter or application. This setup is especially useful when inspecting oauth flows and tokens as it allows stepping through the states.

Example `.vscode/launch.json` file
{
    "version": "0.2.0",
    "configurations": [
        {
            "name": "Launch Package",
            "type": "go",
            "request": "launch",
            "mode": "debug",
            "program": "${workspaceFolder}/cmd/skipper/main.go",
            "args": [
                "-application-log-level=debug",
                "-address=:9999",
                "-inline-routes=PathSubtree(\"/\") -> inlineContent(\"Hello World\") -> <shunt>",
               // example OIDC setup, using https://developer.microsoft.com/en-us/microsoft-365/dev-program
               //  "-oidc-secrets-file=${workspaceFolder}/.vscode/launch.json",
               //  "-inline-routes=* -> oauthOidcAnyClaims(\"https://login.microsoftonline.com/<tenant Id>/v2.0\",\"<application id>\",\"<client secret>\",\"http://localhost:9999/authcallback\", \"profile\", \"\", \"\", \"x-auth-email:claims.email x-groups:claims.groups\") -> inlineContent(\"restricted access\") -> <shunt>",
            ],
            "serverReadyAction": {
                "pattern": "route settings applied",
                "uriFormat": "http://localhost:9999",
                "action": "openExternally"
            }
        }
    ]
}

Docs

We have user documentation and developer documentation separated. In docs/ you find the user documentation in mkdocs format and rendered at https://opensource.zalando.com/skipper which is updated automatically with each docs/ change merged to master branch. Developer documentation for skipper as library users godoc format is used and rendered at https://godoc.org/github.com/zalando/skipper.

User documentation

To see rendered documentation locally run mkdocs serve and navigate to http://127.0.0.1:8000.

Filters

Filters allow to change arbitrary HTTP data in the Request or Response. If you need to read and write the http.Body, please make sure you discuss the use case before creating a pull request.

A filter consists of at least two types a filters.Spec and a filters.Filter. Spec consists of everything that is needed and known before a user will instantiate a filter.

A spec will be created in the bootstrap procedure of a skipper process. A spec has to satisfy the filters.Spec interface Name() string and CreateFilter([]interface{}) (filters.Filter, error).

The actual filter implementation has to satisfy the filter.Filter interface Request(filters.FilterContext) and Response(filters.FilterContext).

The simplest filter possible is, if filters.Spec and filters.Filter are the same type:

type myFilter struct{}

func NewMyFilter() *myFilter {
    return &myFilter{}
}

func (spec *myFilter) Name() string { return "myFilter" }

func (spec *myFilter) CreateFilter(config []interface{}) (filters.Filter, error) {
     return NewMyFilter(), nil
}

func (f *myFilter) Request(ctx filters.FilterContext) {
     // change data in ctx.Request() for example
}

func (f *myFilter) Response(ctx filters.FilterContext) {
     // change data in ctx.Response() for example
}

Find a detailed example at how to develop a filter.

Filters with cleanup

Sometimes your filter needs to cleanup resources on shutdown. In Go functions that do this have often the name Close(). There is the filters.FilterCloser interface that if you comply with it, the routing.Route will make sure your filters are closed in case of routing.Routing was closed.

type myFilter struct{}

func NewMyFilter() *myFilter {
    return &myFilter{}
}

func (spec *myFilter) Name() string { return "myFilter" }

func (spec *myFilter) CreateFilter(config []interface{}) (filters.Filter, error) {
     return NewMyFilter(), nil
}

func (f *myFilter) Request(ctx filters.FilterContext) {
     // change data in ctx.Request() for example
}

func (f *myFilter) Response(ctx filters.FilterContext) {
     // change data in ctx.Response() for example
}

func (f *myFilter) Close() error {
     // cleanup your filter
}

Filters with error handling

Sometimes you want to have a filter that wants to get called Response() even if the proxy will not send a response from the backend, for example you want to count error status codes, like the admissionControl filter. In this case you need to comply with the following proxy interface:

// errorHandlerFilter is an opt-in for filters to get called
// Response(ctx) in case of errors.
type errorHandlerFilter interface {
    // HandleErrorResponse returns true in case a filter wants to get called
    HandleErrorResponse() bool
}

Example:

type myFilter struct{}

func NewMyFilter() *myFilter {
    return &myFilter{}
}

func (spec *myFilter) Name() string { return "myFilter" }

func (spec *myFilter) CreateFilter(config []interface{}) (filters.Filter, error) {
     return NewMyFilter(), nil
}

func (f *myFilter) Request(ctx filters.FilterContext) {
     // change data in ctx.Request() for example
}

func (f *myFilter) Response(ctx filters.FilterContext) {
     // change data in ctx.Response() for example
}

func (f *myFilter) HandleErrorResponse() bool() {
     return true
}

Predicates

Predicates allow to match a condition, that can be based on arbitrary HTTP data in the Request. There are also predicates, that use a chance Traffic() or the current local time, for example After(), to match a request and do not use the HTTP data at all.

A predicate consists of at least two types routing.Predicate and routing.PredicateSpec, which are both interfaces.

A spec will be created in the bootstrap procedure of a skipper process. A spec has to satisfy the routing.PredicateSpec interface Name() string and Create([]interface{}) (routing.Predicate, error).

The actual predicate implementation has to satisfy the routing.Predicate interface Match(*http.Request) bool and returns true if the predicate matches the request. If false is returned, the routing table will be searched for another route that might match the given request.

The simplest possible predicate implementation is, if routing.PredicateSpec and routing.Predicate are the same type:

type myPredicate struct{}

func NewMyPredicate() routing.PredicateSpec {
    return &myPredicate{}
}

func (spec *myPredicate) Name() string { return "myPredicate" }

func (spec *myPredicate) Create(config []interface{}) (routing.Predicate, error) {
     return NewMyPredicate(), nil
}

func (f *myPredicate) Match(r *http.Request) bool {
     // match data in *http.Request for example
     return true
}

Predicates are quite similar to implement as Filters, so for a more complete example, find an example how to develop a filter.

Dataclients

Dataclients are the way how to integrate new route sources. Dataclients pull information from a source and create routes for skipper’s routing table.

You have to implement routing.DataClient, which is an interface that defines function signatures LoadAll() ([]*eskip.Route, error) and LoadUpdate() ([]*eskip.Route, []string, error).

The LoadUpdate() method can be implemented either in a way that returns immediately, or blocks until there is a change. The routing package will regularly call the LoadUpdate() method with a small delay between the calls.

A complete example is the routestring implementation, which fits in less than 50 lines of code.

Opentracing

Your custom Opentracing implementations need to satisfy the opentracing.Tracer interface from https://github.com/opentracing/opentracing-go and need to be loaded as a plugin, which might change in the future. Please check the tracing package and ask for further guidance in our community channels.

Core

Non trivial changes, proposals and enhancements to the core of skipper should be discussed first in a Github issue, such that we can think about how this fits best in the project and how to achieve the most useful result. Feel also free to reach out to our community channels and discuss there your idea.

Every change in core has to have tests included and should be a non breaking change. We planned since a longer time a breaking change, but we should coordinate to make it as good as possible for all skipper as library users. Most often a breaking change can be postponed to the future and a feature independently added and the old feature might be deprecated to delete it later. Use of deprecated features should be shown in logs with a log.Warning.