routing

Documentation

Overview

Package routing implements matching of http requests to a continuously updatable set of skipper routes.

Request Evaluation

1. The path in the http request is used to find one or more matching route definitions in a lookup tree.

2. The rest of the request attributes is matched against the non-path conditions of the routes found in the lookup tree, from the most to the least strict one. The result is the first route where every condition is met.

(The regular expression conditions for the path, 'PathRegexp', are applied only in step 2.)

The matching conditions and the built-in filters that use regular expressions, use the go stdlib regexp, which uses re2:

https://github.com/google/re2/wiki/Syntax

Matching Conditions

The following types of conditions are supported in the route definitions.

- Path: the route definitions may contain a single path condition, optionally with wildcards, used for looking up routes in the lookup tree.

- PathSubtree: similar to Path, but used to match full subtrees including the path of the definition.

- PathRegexp: regular expressions to match the path.

- Host: regular expressions that the host header in the request must match.

- Method: the HTTP method that the request must match.

- Header: a header key and exact value that must be present in the request. Note that Header("Key", "Value") is equivalent to HeaderRegexp("Key", "^Value$").

- HeaderRegexp: a header key and a regular expression, where the key must be present in the request and one of the associated values must match the expression.

Wildcards

Path matching supports two kinds of wildcards:

- simple wildcard: e.g. /some/:wildcard/path. Simple wildcards are matching a single name in the request path.

- freeform wildcard: e.g. /some/path/*wildcard. Freeform wildcards are matching any number of names at the end of the request path.

In case of PathSubtree, simple wildcards behave similar to Path, while freeform wildcards only set the name of the path parameter containing the path in the subtree. If no free wildcard is used in the PathSubtree predicate, the name of this parameter will be "*". This makes the PathSubtree("/foo") predicate equivalent to having routes with Path("/foo"), Path("/foo/") and Path("/foo/**") predicates.

Custom Predicates

It is possible to define custom route matching rules in the form of custom predicates. Custom predicates need to implement the PredicateSpec interface, that serves as a 'factory' and is used in the routing package during constructing the routing tree. When a route containing a custom predicate is matched based on the path tree, the predicate receives the request object, and it returns true or false meaning that the request is a match or not.

Data Clients

Routing definitions are not directly passed to the routing instance, but they are loaded from clients that implement the DataClient interface. The router initially loads the complete set of the routes from each client, merges the different sets based on the route id, and converts them into their runtime representation, with initialized filters based on the filter specifications in the filter registry.

During operation, the router regularly polls the data clients for updates, and, if an update is received, generates a new lookup tree. In case of communication failure during polling, it reloads the whole set of routes from the failing client.

The active set of routes from the last successful update are used until the next successful update happens.

Currently, the routes with the same id coming from different sources are merged in an nondeterministic way, but this behavior may change in the future.

For a full description of the route definitions, see the documentation of the skipper/eskip package.

Example

package main

import (
	"fmt"
	"github.com/zalando/skipper/filters/builtin"
	"github.com/zalando/skipper/routing"
	"github.com/zalando/skipper/routing/testdataclient"
	"log"
	"net/http"
	"time"
)

func main() {
	// create a data client with a predefined route:
	dataClient, err := testdataclient.NewDoc(
		`Path("/some/path/to/:id") -> requestHeader("X-From", "skipper") -> "https://www.example.org"`)
	if err != nil {
		log.Fatal(err)
	}

	// create a router:
	r := routing.New(routing.Options{
		FilterRegistry:  builtin.MakeRegistry(),
		MatchingOptions: routing.IgnoreTrailingSlash,
		DataClients:     []routing.DataClient{dataClient}})

	// let the route data get propagated in the background:
	time.Sleep(36 * time.Millisecond)

	// create a request:
	req, err := http.NewRequest("GET", "https://www.example.com/some/path/to/Hello,+world!", nil)
	if err != nil {
		log.Fatal(err)
	}

	// match the request with the router:
	route, params := r.Route(req)
	if route == nil {
		log.Fatal("failed to route")
	}

	// verify the matched route and the path params:
	fmt.Println(route.Backend)
	fmt.Println(params["id"])

}

Output:

https://www.example.org
Hello, world!

Index

• Constants
• type DataClient
• type MatchingOptions
• type Options
• type PostProcessor
• type Predicate
• type PredicateSpec
• type Route
• type RouteFilter
• type RouteLookup
  • func (rl *RouteLookup) Do(req *http.Request) (*Route, map[string]string)
• type Routing
  • func New(o Options) *Routing
  • func (r *Routing) Close()
  • func (r *Routing) Get() *RouteLookup
  • func (r *Routing) Route(req *http.Request) (*Route, map[string]string)
  • func (r *Routing) ServeHTTP(w http.ResponseWriter, req *http.Request)

Examples

• Package

Constants

const (
	// PathName represents the name of builtin path predicate.
	// (See more details about the Path and PathSubtree predicates
	// at https://godoc.org/github.com/zalando/skipper/eskip)
	PathName = "Path"

	// PathSubtreeName represents the name of the builtin path subtree predicate.
	// (See more details about the Path and PathSubtree predicates
	// at https://godoc.org/github.com/zalando/skipper/eskip)
	PathSubtreeName = "PathSubtree"
)

Types

type DataClient

type DataClient interface {
	LoadAll() ([]*eskip.Route, error)
	LoadUpdate() ([]*eskip.Route, []string, error)
}

DataClient instances provide data sources for route definitions.

type MatchingOptions

type MatchingOptions uint

MatchingOptions controls route matching.

const (
	// MatchingOptionsNone indicates that all options are default.
	MatchingOptionsNone MatchingOptions = 0

	// IgnoreTrailingSlash indicates that trailing slashes in paths are ignored.
	IgnoreTrailingSlash MatchingOptions = 1 << iota
)

type Options

type Options struct {

	// Registry containing the available filter
	// specifications that are used during processing
	// the filter chains in the route definitions.
	FilterRegistry filters.Registry

	// Matching options are flags that control the
	// route matching.
	MatchingOptions MatchingOptions

	// The timeout between requests to the data
	// clients for route definition updates.
	PollTimeout time.Duration

	// The set of different data clients where the
	// route definitions are read from.
	DataClients []DataClient

	// Specifications of custom, user defined predicates.
	Predicates []PredicateSpec

	// Performance tuning option.
	//
	// When zero, the newly constructed routing
	// tree will take effect on the next routing
	// query after every update from the data
	// clients. In case of higher values, the
	// routing queries have priority over the
	// update channel, but the next routing tree
	// takes effect only a few requests later.
	//
	// (Currently disabled and used with hard wired
	// 0, until the performance benefit is verified
	// by benchmarks.)
	UpdateBuffer int

	// Set a custom logger if necessary.
	Log logging.Logger

	// SuppressLogs indicates whether to log only a summary of the route changes.
	SuppressLogs bool

	// PostProcessrs contains custom route post-processors.
	PostProcessors []PostProcessor
}

Options for initialization for routing.

type PostProcessor

type PostProcessor interface {
	Do([]*Route) []*Route
}

PostProcessor is an interface for custom post-processors applying changes to the routes after they were created from their data representation and before they were passed to the proxy.

This feature is experimental.

type Predicate

type Predicate interface {

	// Returns true if the request matches the predicate.
	Match(*http.Request) bool
}

Predicate instances are used as custom user defined route matching predicates.

type PredicateSpec

type PredicateSpec interface {

	// Name of the predicate as used in the route definitions.
	Name() string

	// Creates a predicate instance with concrete arguments.
	Create([]interface{}) (Predicate, error)
}

PredicateSpec instances are used to create custom predicates (of type Predicate) with concrete arguments during the construction of the routing tree.

type Route

type Route struct {

	// Fields from the static route definition.
	eskip.Route

	// The backend scheme and host.
	Scheme, Host string

	// The preprocessed custom predicate instances.
	Predicates []Predicate

	// The preprocessed filter instances.
	Filters []*RouteFilter

	// Next is forming a linked to the next route of a
	// loadbalanced group of routes. This is nil if the route is
	// the last in the linked list or there is only one route. To
	// find the Next in case of the last route of the list, you
	// have to use the Head.
	Next *Route

	// Head is the pointer to the head of linked list that forms
	// the loadbalancer group of Route. Every Route will point to
	// the same Route for being Head.
	Head *Route

	// Me is a pointer to self, to workaround Go type missmatch
	// check, because eskip.Route != routing.Route
	Me *Route

	// Group is equal for all routes, members, forming a loadbalancer pool.
	Group string

	// IsLoadBalanced tells the proxy that the current route
	// is a member of a load balanced group.
	IsLoadBalanced bool
	// contains filtered or unexported fields
}

Route object with preprocessed filter instances.

type RouteFilter

type RouteFilter struct {
	filters.Filter
	Name  string
	Index int
}

RouteFilter contains extensions to generic filter interface, serving mainly logging/monitoring purpose.

type RouteLookup

type RouteLookup struct {
	// contains filtered or unexported fields
}

RouteLookup captures a single generation of the lookup tree, allowing multiple lookups to the same version of the lookup tree.

Experimental feature. Using this solution potentially can cause large memory consumption in extreme cases, typically when: the total number routes is large, the backend responses to a subset of these routes is slow, and there's a rapid burst of consecutive updates to the routing table. This situation is considered an edge case, but until a protection against is found, the feature is experimental and its exported interface may change.

func (*RouteLookup) Do

func (rl *RouteLookup) Do(req *http.Request) (*Route, map[string]string)

Do executes the lookup against the captured routing table. Equivalent to Routing.Route().

type Routing

type Routing struct {
	// contains filtered or unexported fields
}

Routing ('router') instance providing live updatable request matching.

func New

func New(o Options) *Routing

New initializes a routing instance, and starts listening for route definition updates.

func (*Routing) Close

func (r *Routing) Close()

Close closes routing, stops receiving routes.

func (*Routing) Get

func (r *Routing) Get() *RouteLookup

Get returns a captured generation of the lookup table. This feature is experimental. See the description of the RouteLookup type.

func (*Routing) Route

func (r *Routing) Route(req *http.Request) (*Route, map[string]string)

Route matches a request in the current routing tree.

If the request matches a route, returns the route and a map of parameters constructed from the wildcard parameters in the path condition if any. If there is no match, it returns nil.

func (*Routing) ServeHTTP

func (r *Routing) ServeHTTP(w http.ResponseWriter, req *http.Request)

ServeHTTP renders the list of current routes.