Fox is a zero allocation, lightweight, high performance HTTP request router for Go. The main difference with other routers is that it supports mutation on its routing tree while handling request concurrently. Internally, Fox use a Concurrent Radix Tree that support lock-free reads while allowing concurrent writes. The router tree is optimized for high-concurrency and high performance reads, and low-concurrency write.
Fox supports various use cases, but it is especially designed for applications that require changes at runtime to their routing structure based on user input, configuration changes, or other runtime events.
The current api is not yet stabilize. Breaking changes may occur before v1.0.0 and will be noted on the release note.
Runtime updates: Register, update and remove route handler safely at any time without impact on performance. Fox never block while serving request!
Wildcard pattern: Route can be registered using wildcard parameters. The matched path segment can then be easily retrieved by name. Due to Fox design, wildcard route are cheap and scale really well.
Detect panic: Comes with a ready-to-use, efficient Recovery middleware that gracefully handles panics.
Get the current route: You can easily retrieve the route of the matched request. This actually makes it easier to integrate observability middleware like open telemetry.
Flexible routing: Fox strikes a balance between routing flexibility, performance and clarity by enforcing clear priority rules, ensuring that there are no unintended matches and maintaining high performance even for complex routing pattern.
Redirect trailing slashes: Inspired from httprouter, the router automatically redirects the client, at no extra cost, if another route match with or without a trailing slash.
Automatic OPTIONS replies: Inspired from httprouter, the router has built-in native support for OPTIONS requests.
Of course, you can also register custom NotFound and MethodNotAllowed handlers.
go get -u github.com/tigerwill90/foxpackage main
import (
"github.com/tigerwill90/fox"
"log"
"net/http"
)
type Greeting struct {
Say string
}
func (h *Greeting) Greet(c fox.Context) {
_ = c.String(http.StatusOK, "%s %s\n", h.Say, c.Param("name"))
}
func main() {
f := fox.New(fox.DefaultOptions())
err := f.Handle(http.MethodGet, "/", func(c fox.Context) {
_ = c.String(http.StatusOK, "Welcome\n")
})
if err != nil {
panic(err)
}
h := Greeting{Say: "Hello"}
f.MustHandle(http.MethodGet, "/hello/{name}", h.Greet)
log.Fatalln(http.ListenAndServe(":8080", f))
}Since new route may be added at any given time, Fox, unlike other router, does not panic when a route is malformed or conflicts with another. Instead, it returns the following error values:
ErrRouteExist = errors.New("route already registered")
ErrRouteConflict = errors.New("route conflict")
ErrInvalidRoute = errors.New("invalid route")Conflict error may be unwrapped to retrieve conflicting route.
if errors.Is(err, fox.ErrRouteConflict) {
matched := err.(*fox.RouteConflictError).Matched
for _, route := range matched {
fmt.Println(route)
}
}A route can be defined using placeholder (e.g {name}). The matching segment are recorder into the fox.Params slice accessible
via fox.Context. The Param and Get methods are helpers to retrieve the value using the placeholder name.
Pattern /avengers/{name}
/avengers/ironman match
/avengers/thor match
/avengers/hulk/angry no match
/avengers/ no match
Pattern /users/uuid:{id}
/users/uuid:123 match
/users/uuid no match
Catch-all parameters can be used to match everything at the end of a route. The placeholder start with * followed by a regular
named parameter (e.g. *{name}).
Pattern /src/*{filepath}
/src/ match
/src/conf.txt match
/src/dir/config.txt match
Patter /src/file=*{path}
/src/file= match
/src/file=config.txt match
/src/file=/dir/config.txt match
Routes are prioritized based on specificity, with static segments taking precedence over wildcard segments.
The following rules apply:
- Static segments are always evaluated first.
- A named parameter can only overlap with a catch-all parameter or static segments.
- A catch-all parameter can only overlap with a named parameter or static segments.
- When a named parameter overlaps with a catch-all parameter, the named parameter is evaluated first.
For instance, GET /users/{id} and GET /users/{name}/profile cannot coexist, as the {id} and {name} segments
are overlapping. These limitations help to minimize the number of branches that need to be evaluated in order to find
the right match, thereby maintaining high-performance routing.
For example, the followings route are allowed:
GET /*{filepath}
GET /users/{id}
GET /users/{id}/emails
GET /users/{id}/{actions}
POST /users/{name}/emails
Additionally, let's consider an example to illustrate the prioritization:
GET /fs/avengers.txt #1 => match /fs/avengers.txt
GET /fs/{filename} #2 => match /fs/ironman.txt
GET /fs/*{filepath} #3 => match /fs/avengers/ironman.txt
The fox.Context instance is freed once the request handler function returns to optimize resource allocation.
If you need to retain fox.Context or fox.Params beyond the scope of the handler, use the Clone methods.
func Hello(c fox.Context) {
cc := c.Clone()
// cp := c.Params().Clone()
go func() {
time.Sleep(2 * time.Second)
log.Println(cc.Param("name")) // Safe
}()
_ = c.String(http.StatusOK, "Hello %s\n", c.Param("name"))
}Fox implements a Concurrent Radix Tree that supports lock-free reads while allowing concurrent writes, by calculating the changes which would be made to the tree were it mutable, and assembling those changes into a patch, which is then applied to the tree in a single atomic operation.
For example, here we are inserting the new key toast into to the tree which require an existing node to be split:
When traversing the tree during a patch, reading threads will either see the old version or the new version of the (sub-)tree, but both version are consistent view of the tree.
- Routing requests is lock-free (reading thread never block, even while writes are ongoing)
- The router always see a consistent version of the tree while routing request
- Reading threads do not block writing threads (adding, updating or removing a handler can be done concurrently)
- Writing threads block each other but never block reading threads
As such threads that route requests should never encounter latency due to ongoing writes or other concurrent readers.
In this example, the handler for routes/{action} allow to dynamically register, update and remove handler for the
given route and method. Thanks to Fox's design, those actions are perfectly safe and may be executed concurrently.
package main
import (
"encoding/json"
"fmt"
"github.com/tigerwill90/fox"
"log"
"net/http"
"strings"
)
func Action(c fox.Context) {
var data map[string]string
if err := json.NewDecoder(c.Request().Body).Decode(&data); err != nil {
http.Error(c.Writer(), err.Error(), http.StatusBadRequest)
return
}
method := strings.ToUpper(data["method"])
path := data["path"]
text := data["text"]
if path == "" || method == "" {
http.Error(c.Writer(), "missing method or path", http.StatusBadRequest)
return
}
var err error
action := c.Param("action")
switch action {
case "add":
err = c.Fox().Handle(method, path, func(c fox.Context) {
_ = c.String(http.StatusOK, text)
})
case "update":
err = c.Fox().Update(method, path, func(c fox.Context) {
_ = c.String(http.StatusOK, text)
})
case "delete":
err = c.Fox().Remove(method, path)
default:
http.Error(c.Writer(), fmt.Sprintf("action %q is not allowed", action), http.StatusBadRequest)
return
}
if err != nil {
http.Error(c.Writer(), err.Error(), http.StatusConflict)
return
}
_ = c.String(http.StatusOK, "%s route [%s] %s: success\n", action, method, path)
}
func main() {
f := fox.New()
f.MustHandle(http.MethodPost, "/routes/{action}", Action)
log.Fatalln(http.ListenAndServe(":8080", f))
}Fox also enables you to replace the entire tree in a single atomic operation using the Swap methods.
Note that router's options apply automatically on the new tree.
package main
import (
"fox-by-example/db"
"github.com/tigerwill90/fox"
"html/template"
"log"
"net/http"
"strings"
"time"
)
type HtmlRenderer struct {
Template template.HTML
}
func (h *HtmlRenderer) Render(c fox.Context) {
log.Printf("matched handler path: %s", c.Path())
_ = c.Stream(
http.StatusOK,
fox.MIMETextHTMLCharsetUTF8,
strings.NewReader(string(h.Template)),
)
}
func main() {
f := fox.New()
go Reload(f)
log.Fatalln(http.ListenAndServe(":8080", f))
}
func Reload(r *fox.Router) {
for ; true; <-time.Tick(10 * time.Second) {
routes := db.GetRoutes()
tree := r.NewTree()
for _, rte := range routes {
h := HtmlRenderer{Template: rte.Template}
if err := tree.Handle(rte.Method, rte.Path, h.Render); err != nil {
log.Printf("error reloading route: %s\n", err)
continue
}
}
// Swap the currently in-use routing tree with the new provided.
r.Swap(tree)
log.Println("route reloaded")
}
}In certain situations, it's necessary to maintain a consistent view of the tree while performing updates.
The Tree API allow you to take control of the internal sync.Mutex to prevent concurrent writes from
other threads. Remember that all write operation should be run serially.
In the following example, the Upsert function needs to perform a lookup on the tree to check if a handler
is already registered for the provided method and path. By locking the Tree, this operation ensures
atomicity, as it prevents other threads from modifying the tree between the lookup and the write operation.
Note that all read operation on the tree remain lock-free.
func Upsert(t *fox.Tree, method, path string, handler fox.HandlerFunc) error {
t.Lock()
defer t.Unlock()
if t.Has(method, path) {
return t.Update(method, path, handler)
}
return t.Handle(method, path, handler)
}When working with the Tree API, it's important to keep some considerations in mind. Each instance has its
own sync.Mutex that can be used to serialize writes. However, unlike the router API, the lower-level Tree API
does not automatically lock the tree when writing to it. Therefore, it is the user's responsibility to ensure
all writes are executed serially.
Moreover, since the router tree may be swapped at any given time, you MUST always copy the pointer locally to
avoid inadvertently causing a deadlock by locking/unlocking the wrong Tree.
// Good
t := r.Tree()
t.Lock()
defer t.Unlock()
// Dramatically bad, may cause deadlock
r.Tree().Lock()
defer r.Tree().Unlock()
// Dramatically bad, may cause deadlock
func handle(c fox.Context) {
c.Fox().Tree().Lock()
defer c.Fox().Tree().Unlock()
}Note that fox.Context carries a local copy of the Tree that is being used to serve the handler, thereby eliminating
the risk of deadlock when using the Tree within the context.
// Ok
func handle(c fox.Context) {
c.Tree().Lock()
defer c.Tree().Unlock()
}Fox itself implements the http.Handler interface which make easy to chain any compatible middleware before the router. Moreover, the router
provides convenient fox.WrapF and fox.WrapH adapter to be use with http.Handler.
The route parameters can be accessed by the wrapped handler through the context.Context when the adapters fox.WrapF and fox.WrapH are used.
Wrapping an http.Handler
articles := http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
params := fox.ParamsFromContext(r.Context())
_, _ = fmt.Fprintf(w, "Article id: %s\n", params.Get("id"))
})
f := fox.New(fox.DefaultOptions())
f.MustHandle(http.MethodGet, "/articles/{id}", fox.WrapH(httpRateLimiter.RateLimit(articles)))Middlewares can be registered globally using the fox.WithMiddleware option. The example below demonstrates how
to create and apply automatically a simple logging middleware to all routes (including 404, 405, etc...).
package main
import (
"github.com/tigerwill90/fox"
"log"
"net/http"
"time"
)
func Logger(next fox.HandlerFunc) fox.HandlerFunc {
return func(c fox.Context) {
start := time.Now()
next(c)
log.Printf("route: %s, latency: %s, status: %d, size: %d",
c.Path(),
time.Since(start),
c.Writer().Status(),
c.Writer().Size(),
)
}
}
func main() {
f := fox.New(fox.WithMiddleware(Logger))
f.MustHandle(http.MethodGet, "/", func(c fox.Context) {
resp, err := http.Get("https://api.coindesk.com/v1/bpi/currentprice.json")
if err != nil {
http.Error(c.Writer(), http.StatusText(http.StatusInternalServerError), http.StatusInternalServerError)
return
}
defer resp.Body.Close()
_ = c.Stream(http.StatusOK, fox.MIMEApplicationJSON, resp.Body)
})
log.Fatalln(http.ListenAndServe(":8080", f))
}Additionally, fox.WithMiddlewareFor option provide a more fine-grained control over where a middleware is applied, such as
only for 404 or 405 handlers. Possible scopes include fox.RouteHandlers (regular routes), fox.NoRouteHandler, fox.NoMethodHandler,
fox.RedirectHandler, fox.OptionsHandler and any combination of these.
f := fox.New(
fox.WithMethodNotAllowed(true),
fox.WithMiddlewareFor(fox.RouteHandlers, fox.Recovery(fox.DefaultHandleRecovery), Logger),
fox.WithMiddlewareFor(fox.NoRouteHandler|fox.NoMethodHandler, SpecialLogger),
)- tigerwill90/otelfox: Distributed tracing with OpenTelemetry
- tigerwill90/foxdump: Body dump middleware for capturing requests and responses payload.
- tigerwill90/foxtimeout:
http.TimeoutHandlermiddleware optimized for Fox.
The WithAutoOptions setting or the WithOptionsHandler registration enable automatic responses to OPTIONS requests.
This feature can be particularly useful in the context of Cross-Origin Resource Sharing (CORS).
An OPTIONS request is a type of HTTP request that is used to determine the communication options available for a given resource or API endpoint. These requests are commonly used as "preflight" requests in CORS to check if the CORS protocol is understood and to get permission to access data based on origin.
When automatic OPTIONS responses is enabled, the router will automatically respond to preflight OPTIONS requests and set the Allow
header with the appropriate value. To customize how OPTIONS requests are handled (e.g. adding CORS headers), you may register a middleware for the
fox.OptionsHandler scope or override the default handler.
f := fox.New(
fox.WithOptionsHandler(func(c fox.Context) {
if c.Header("Access-Control-Request-Method") != "" {
// Setting CORS headers.
c.SetHeader("Access-Control-Allow-Methods", c.Writer().Header().Get("Allow"))
c.SetHeader("Access-Control-Allow-Origin", "*")
}
// Respond with a 204 status code.
c.Writer().WriteHeader(http.StatusNoContent)
}),
)The primary goal of Fox is to be a lightweight, high performance router which allow routes modification at runtime. The following benchmarks attempt to compare Fox to various popular alternatives, including both fully-featured web frameworks and lightweight request routers. These benchmarks are based on the julienschmidt/go-http-routing-benchmark repository.
Please note that these benchmarks should not be taken too seriously, as the comparison may not be entirely fair due to the differences in feature sets offered by each framework. Performance should be evaluated in the context of your specific use case and requirements. While Fox aims to excel in performance, it's important to consider the trade-offs and functionality provided by different web frameworks and routers when making your selection.
GOOS: Linux
GOARCH: amd64
GO: 1.20
CPU: Intel(R) Core(TM) i9-9900K CPU @ 3.60GHz
It is just a collection of random static paths inspired by the structure of the Go directory. It might not be a realistic URL-structure.
GOMAXPROCS: 1
BenchmarkHttpRouter_StaticAll 161659 7570 ns/op 0 B/op 0 allocs/op
BenchmarkHttpTreeMux_StaticAll 132446 8836 ns/op 0 B/op 0 allocs/op
BenchmarkFox_StaticAll 102577 11348 ns/op 0 B/op 0 allocs/op
BenchmarkStdMux_StaticAll 91304 13382 ns/op 0 B/op 0 allocs/op
BenchmarkGin_StaticAll 78224 15433 ns/op 0 B/op 0 allocs/op
BenchmarkEcho_StaticAll 77923 15739 ns/op 0 B/op 0 allocs/op
BenchmarkBeego_StaticAll 10000 101094 ns/op 55264 B/op 471 allocs/op
BenchmarkGorillaMux_StaticAll 2283 525683 ns/op 113041 B/op 1099 allocs/op
BenchmarkMartini_StaticAll 1330 936928 ns/op 129210 B/op 2031 allocs/op
BenchmarkTraffic_StaticAll 1064 1140959 ns/op 753611 B/op 14601 allocs/op
BenchmarkPat_StaticAll 967 1230424 ns/op 602832 B/op 12559 allocs/op
In this benchmark, Fox performs as well as Gin and Echo which are both Radix Tree based routers. An interesting fact is
that HttpTreeMux also support adding route while serving request concurrently.
However, it takes a slightly different approach, by using an optional RWMutex that may not scale as well as Fox under heavy load. The next
test compare HttpTreeMux with and without the *SafeAddRouteFlag (concurrent reads and writes) and Fox in parallel benchmark.
GOMAXPROCS: 16
Route: all
BenchmarkFox_StaticAll-16 99322 11369 ns/op 0 B/op 0 allocs/op
BenchmarkFox_StaticAllParallel-16 831354 1422 ns/op 0 B/op 0 allocs/op
BenchmarkHttpTreeMux_StaticAll-16 135560 8861 ns/op 0 B/op 0 allocs/op
BenchmarkHttpTreeMux_StaticAllParallel-16* 172714 6916 ns/op 0 B/op 0 allocs/op
As you can see, this benchmark highlight the cost of using higher synchronisation primitive like RWMutex to be able to register new route while handling requests.
The following benchmarks measure the cost of some very basic operations.
In the first benchmark, only a single route, containing a parameter, is loaded into the routers. Then a request for a URL matching this pattern is made and the router has to call the respective registered handler function. End.
GOMAXPROCS: 1
BenchmarkFox_Param 33024534 36.61 ns/op 0 B/op 0 allocs/op
BenchmarkEcho_Param 31472508 38.71 ns/op 0 B/op 0 allocs/op
BenchmarkGin_Param 25826832 52.88 ns/op 0 B/op 0 allocs/op
BenchmarkHttpRouter_Param 21230490 60.83 ns/op 32 B/op 1 allocs/op
BenchmarkHttpTreeMux_Param 3960292 280.4 ns/op 352 B/op 3 allocs/op
BenchmarkBeego_Param 2247776 518.9 ns/op 352 B/op 3 allocs/op
BenchmarkPat_Param 1603902 676.6 ns/op 512 B/op 10 allocs/op
BenchmarkGorillaMux_Param 1000000 1011 ns/op 1024 B/op 8 allocs/op
BenchmarkTraffic_Param 648986 1686 ns/op 1848 B/op 21 allocs/op
BenchmarkMartini_Param 485839 2446 ns/op 1096 B/op 12 allocs/op
Same as before, but now with multiple parameters, all in the same single route. The intention is to see how the routers scale with the number of parameters.
GOMAXPROCS: 1
BenchmarkFox_Param5 16608495 72.84 ns/op 0 B/op 0 allocs/op
BenchmarkGin_Param5 13098740 92.22 ns/op 0 B/op 0 allocs/op
BenchmarkEcho_Param5 12025460 96.33 ns/op 0 B/op 0 allocs/op
BenchmarkHttpRouter_Param5 8233530 148.1 ns/op 160 B/op 1 allocs/op
BenchmarkHttpTreeMux_Param5 1986019 616.9 ns/op 576 B/op 6 allocs/op
BenchmarkBeego_Param5 1836229 655.3 ns/op 352 B/op 3 allocs/op
BenchmarkGorillaMux_Param5 757936 1572 ns/op 1088 B/op 8 allocs/op
BenchmarkPat_Param5 645847 1724 ns/op 800 B/op 24 allocs/op
BenchmarkTraffic_Param5 424431 2729 ns/op 2200 B/op 27 allocs/op
BenchmarkMartini_Param5 424806 2772 ns/op 1256 B/op 13 allocs/op
BenchmarkGin_Param20 4636416 244.6 ns/op 0 B/op 0 allocs/op
BenchmarkFox_Param20 4667533 250.7 ns/op 0 B/op 0 allocs/op
BenchmarkEcho_Param20 4352486 277.1 ns/op 0 B/op 0 allocs/op
BenchmarkHttpRouter_Param20 2618958 455.2 ns/op 640 B/op 1 allocs/op
BenchmarkBeego_Param20 847029 1688 ns/op 352 B/op 3 allocs/op
BenchmarkHttpTreeMux_Param20 369500 2972 ns/op 3195 B/op 10 allocs/op
BenchmarkGorillaMux_Param20 318134 3561 ns/op 3195 B/op 10 allocs/op
BenchmarkMartini_Param20 223070 5117 ns/op 3619 B/op 15 allocs/op
BenchmarkPat_Param20 157380 7442 ns/op 4094 B/op 73 allocs/op
BenchmarkTraffic_Param20 119677 9864 ns/op 7847 B/op 47 allocs/op
Now let's see how expensive it is to access a parameter. The handler function reads the value (by the name of the parameter, e.g. with a map
lookup; depends on the router) and writes it to /dev/null
GOMAXPROCS: 1
BenchmarkFox_ParamWrite 16707409 72.53 ns/op 0 B/op 0 allocs/op
BenchmarkHttpRouter_ParamWrite 16478174 73.30 ns/op 32 B/op 1 allocs/op
BenchmarkGin_ParamWrite 15828385 75.73 ns/op 0 B/op 0 allocs/op
BenchmarkEcho_ParamWrite 13187766 95.18 ns/op 8 B/op 1 allocs/op
BenchmarkHttpTreeMux_ParamWrite 4132832 279.9 ns/op 352 B/op 3 allocs/op
BenchmarkBeego_ParamWrite 2172572 554.3 ns/op 360 B/op 4 allocs/op
BenchmarkPat_ParamWrite 1200334 996.8 ns/op 936 B/op 14 allocs/op
BenchmarkGorillaMux_ParamWrite 1000000 1005 ns/op 1024 B/op 8 allocs/op
BenchmarkMartini_ParamWrite 454255 2667 ns/op 1168 B/op 16 allocs/op
BenchmarkTraffic_ParamWrite 511766 2021 ns/op 2272 B/op 25 allocs/op
In those micro benchmarks, we can see that Fox scale really well, even with long wildcard routes. Like Gin, this router reuse the
data structure (e.g. fox.Context slice) containing the matching parameters in order to remove completely heap allocation.
Finally, this benchmark execute a request for each GitHub API route (203 routes).
GOMAXPROCS: 1
BenchmarkFox_GithubAll 63984 18555 ns/op 0 B/op 0 allocs/op
BenchmarkEcho_GithubAll 49312 23353 ns/op 0 B/op 0 allocs/op
BenchmarkGin_GithubAll 48422 24926 ns/op 0 B/op 0 allocs/op
BenchmarkHttpRouter_GithubAll 45706 26818 ns/op 14240 B/op 171 allocs/op
BenchmarkHttpTreeMux_GithubAll 14731 80133 ns/op 67648 B/op 691 allocs/op
BenchmarkBeego_GithubAll 7692 137926 ns/op 72929 B/op 625 allocs/op
BenchmarkTraffic_GithubAll 636 1916586 ns/op 845114 B/op 14634 allocs/op
BenchmarkMartini_GithubAll 530 2205947 ns/op 238546 B/op 2813 allocs/op
BenchmarkGorillaMux_GithubAll 529 2246380 ns/op 203844 B/op 1620 allocs/op
BenchmarkPat_GithubAll 424 2899405 ns/op 1843501 B/op 29064 allocs/op
- Update route syntax @v0.6.0
- Route overlapping @v0.7.0
- Route overlapping (catch-all and params) @v0.10.0
- Improving performance and polishing
This project aims to provide a lightweight, high performance and easy to use http router. It purposely has a limited set of features and exposes a relatively low-level api. The intention behind these choices is that it can serve as a building block for implementing your own "batteries included" frameworks. Feature requests and PRs along these lines are welcome.
- npgall/concurrent-trees: Fox design is largely inspired from Niall Gallagher's Concurrent Trees design.
- julienschmidt/httprouter: some feature that implements Fox are inspired from Julien Schmidt's router. Most notably, this package uses the optimized httprouter.Cleanpath function.
- The router API is influenced by popular routers such as gin and echo.