Fast HTTP implementation for Go.
Currently fasthttp is successfully used in a production serving up to 1M concurrent keep-alive connections doing 100K qps from a single server.
Switching from net/http to fasthttp
HTTP server performance comparison with net/http
In short, fasthttp server is up to 10 times faster than net/http. Below are benchmark results.
GOMAXPROCS=1
net/http:
$ GOMAXPROCS=1 go test -bench=NetHTTPServerGet -benchmem -benchtime=5s
PASS
BenchmarkNetHTTPServerGet1ReqPerConn 300000 21236 ns/op 2404 B/op 30 allocs/op
BenchmarkNetHTTPServerGet2ReqPerConn 500000 14634 ns/op 2371 B/op 24 allocs/op
BenchmarkNetHTTPServerGet10ReqPerConn 1000000 9447 ns/op 2101 B/op 19 allocs/op
BenchmarkNetHTTPServerGet10KReqPerConn 1000000 7939 ns/op 2033 B/op 18 allocs/op
BenchmarkNetHTTPServerGet1ReqPerConn10KClients 300000 30291 ns/op 4589 B/op 31 allocs/op
BenchmarkNetHTTPServerGet2ReqPerConn10KClients 500000 23199 ns/op 3581 B/op 25 allocs/op
BenchmarkNetHTTPServerGet10ReqPerConn10KClients 500000 13270 ns/op 2621 B/op 19 allocs/op
BenchmarkNetHTTPServerGet100ReqPerConn10KClients 500000 11412 ns/op 2119 B/op 18 allocs/op
fasthttp:
$ GOMAXPROCS=1 go test -bench=kServerGet -benchmem -benchtime=5s
PASS
BenchmarkServerGet1ReqPerConn 3000000 2341 ns/op 0 B/op 0 allocs/op
BenchmarkServerGet2ReqPerConn 5000000 1799 ns/op 0 B/op 0 allocs/op
BenchmarkServerGet10ReqPerConn 5000000 1239 ns/op 0 B/op 0 allocs/op
BenchmarkServerGet10KReqPerConn 10000000 1090 ns/op 0 B/op 0 allocs/op
BenchmarkServerGet1ReqPerConn10KClients 3000000 2860 ns/op 4 B/op 0 allocs/op
BenchmarkServerGet2ReqPerConn10KClients 3000000 1992 ns/op 1 B/op 0 allocs/op
BenchmarkServerGet10ReqPerConn10KClients 5000000 1297 ns/op 1 B/op 0 allocs/op
BenchmarkServerGet100ReqPerConn10KClients 10000000 1264 ns/op 9 B/op 0 allocs/op
GOMAXPROCS=4
net/http:
$ GOMAXPROCS=4 go test -bench=NetHTTPServerGet -benchmem -benchtime=5s
PASS
BenchmarkNetHTTPServerGet1ReqPerConn-4 1000000 5545 ns/op 2433 B/op 30 allocs/op
BenchmarkNetHTTPServerGet2ReqPerConn-4 2000000 4147 ns/op 2398 B/op 24 allocs/op
BenchmarkNetHTTPServerGet10ReqPerConn-4 3000000 2628 ns/op 2118 B/op 19 allocs/op
BenchmarkNetHTTPServerGet10KReqPerConn-4 3000000 2304 ns/op 2037 B/op 18 allocs/op
BenchmarkNetHTTPServerGet1ReqPerConn10KClients-4 1000000 7327 ns/op 3561 B/op 30 allocs/op
BenchmarkNetHTTPServerGet2ReqPerConn10KClients-4 1000000 5952 ns/op 3073 B/op 24 allocs/op
BenchmarkNetHTTPServerGet10ReqPerConn10KClients-4 2000000 4345 ns/op 2530 B/op 19 allocs/op
BenchmarkNetHTTPServerGet100ReqPerConn10KClients-4 2000000 3866 ns/op 2132 B/op 18 allocs/op
fasthttp:
$ GOMAXPROCS=4 go test -bench=kServerGet -benchmem -benchtime=5s
PASS
BenchmarkServerGet1ReqPerConn-4 10000000 1053 ns/op 0 B/op 0 allocs/op
BenchmarkServerGet2ReqPerConn-4 10000000 685 ns/op 0 B/op 0 allocs/op
BenchmarkServerGet10ReqPerConn-4 20000000 393 ns/op 0 B/op 0 allocs/op
BenchmarkServerGet10KReqPerConn-4 20000000 338 ns/op 0 B/op 0 allocs/op
BenchmarkServerGet1ReqPerConn10KClients-4 10000000 1033 ns/op 0 B/op 0 allocs/op
BenchmarkServerGet2ReqPerConn10KClients-4 10000000 668 ns/op 0 B/op 0 allocs/op
BenchmarkServerGet10ReqPerConn10KClients-4 20000000 393 ns/op 0 B/op 0 allocs/op
BenchmarkServerGet100ReqPerConn10KClients-4 20000000 384 ns/op 4 B/op 0 allocs/op
In short, fasthttp client is up to 10 times faster than net/http. Below are benchmark results.
GOMAXPROCS=1
net/http:
$ GOMAXPROCS=1 go test -bench='HTTPClient(Do|GetEndToEnd)' -benchmem -benchtime=5s
PASS
BenchmarkNetHTTPClientDoFastServer 500000 17535 ns/op 2624 B/op 38 allocs/op
BenchmarkNetHTTPClientGetEndToEnd 200000 56593 ns/op 5012 B/op 59 allocs/op
fasthttp:
$ GOMAXPROCS=1 go test -bench='kClient(Do|GetEndToEnd)' -benchmem -benchtime=5s
PASS
BenchmarkClientDoFastServer 5000000 1420 ns/op 0 B/op 0 allocs/op
BenchmarkClientGetEndToEnd 500000 17912 ns/op 0 B/op 0 allocs/op
GOMAXPROCS=4
net/http:
$ GOMAXPROCS=4 go test -bench='HTTPClient(Do|GetEndToEnd)' -benchmem -benchtime=5s
PASS
BenchmarkNetHTTPClientDoFastServer-4 1000000 5795 ns/op 2626 B/op 38 allocs/op
BenchmarkNetHTTPClientGetEndToEnd-4 500000 19304 ns/op 5953 B/op 62 allocs/op
fasthttp:
$ GOMAXPROCS=4 go test -bench='kClient(Do|GetEndToEnd)' -benchmem -benchtime=5s
PASS
BenchmarkClientDoFastServer-4 20000000 443 ns/op 0 B/op 0 allocs/op
BenchmarkClientGetEndToEnd-4 1000000 5954 ns/op 0 B/op 0 allocs/op
Unfortunately, fasthttp doesn't provide API identical to net/http. See the FAQ for details.
Important points:
- Fasthttp works with RequestHandler functions instead of objects implementing Handler interface. Fortunately, it is easy to pass bound struct methods to fasthttp:
type MyHandler struct {
foobar string
}
// request handler in net/http style, i.e. method bound to MyHandler struct.
func (h *MyHandler) HandleFastHTTP(ctx *fasthttp.RequestCtx) {
// notice that we may access MyHandler properties here - see h.foobar.
fmt.Fprintf(ctx, "Hello, world! Requested path is %q. Foobar is %q",
ctx.Path(), h.foobar)
}
// request handler in fasthttp style, i.e. just plain function.
func fastHTTPHandler(ctx *fasthttp.RequestCtx) {
fmt.Fprintf(ctx, "Hi there! RequestURI is %q", ctx.RequestURI())
}
// pass bound struct method to fasthttp
myHandler := &MyHandler{
foobar: "foobar",
}
fasthttp.ListenAndServe(":8080", myHandler.HandleFastHTTP)
// pass plain function to fasthttp
fasthttp.ListenAndServe(":8081", fastHTTPHandler)
- The RequestHandler accepts only one argument - RequestCtx. It contains all the functionality required for http request processing and response writing. Below is an example of a simple request handler conversion from net/http to fasthttp.
// net/http request handler
requestHandler := func(w http.ResponseWriter, r *http.Request) {
switch r.URL.Path {
case "/foo":
fooHandler(w, r)
case "/bar":
barHandler(w, r)
default:
http.Error(w, "Unsupported path", http.StatusNotFound)
}
}
// the corresponding fasthttp request handler
requestHandler := func(ctx *fasthttp.RequestCtx) {
switch string(ctx.Path()) {
case "/foo":
fooHandler(ctx)
case "/bar":
barHandler(ctx)
default:
ctx.Error("Unsupported path", fasthttp.StatusNotFound)
}
}
- Fasthttp allows setting response headers and writing response body in arbitray order. There is no 'headers first, then body' restriction like in net/http. The following code is valid for fasthttp:
requestHandler := func(ctx *fasthttp.RequestCtx) {
// set some headers and status code first
ctx.SetContentType("foo/bar")
ctx.SetStatusCode(fasthttp.StatusOK)
// then write the first part of body
fmt.Fprintf(ctx, "this is the first part of body\n")
// then set more headers
ctx.Response.Header.Set("Foo-Bar", "baz")
// then write more body
fmt.Fprintf(ctx, "this is the second part of body\n")
// then override already written body
ctx.SetBody([]byte("this is completely new body contents"))
// then update status code
ctx.SetStatusCode(fasthttp.StatusNotFound)
// basically, anything may be updated many times before
// returning from RequestHandler.
//
// Unlike net/http fasthttp doesn't put response to the wire until
// returning from RequestHandler.
}
- Fasthttp doesn't provide ServeMux, since I believe third-party request routers like httprouter must be used instead. Net/http code with simple ServeMux is trivially converted to fasthttp code:
// net/http code
m := &http.ServeMux{}
m.HandleFunc("/foo", fooHandlerFunc)
m.HandleFunc("/bar", barHandlerFunc)
m.Handle("/baz", bazHandler)
http.ListenAndServe(":80", m)
// the corresponding fasthttp code
m := func(ctx *fasthttp.RequestCtx) {
switch string(ctx.Path()) {
case "/foo":
fooHandlerFunc(ctx)
case "/bar":
barHandlerFunc(ctx)
case "/baz":
bazHandler.HandlerFunc(ctx)
default:
ctx.Error("not found", fasthttp.StatusNotFound)
}
}
fastttp.ListenAndServe(":80", m)
-
net/http -> fasthttp conversion table:
- All the pseudocode below assumes w, r and ctx have these types:
var ( w http.ResponseWriter r *http.Request ctx *fasthttp.RequestCtx )
- r.Body -> ctx.PostBody()
- r.URL.Path -> ctx.Path()
- r.URL -> ctx.URI()
- r.Method -> ctx.Method()
- r.Header -> ctx.Request.Header
- r.Header.Get() -> ctx.Request.Header.Peek()
- r.Host -> ctx.Host()
- r.Form -> ctx.QueryArgs() + ctx.PostArgs()
- r.PostForm -> ctx.PostArgs()
- r.FormValue() -> ctx.QueryArgs().Peek()
- r.MultipartForm -> ctx.MultipartForm()
- r.RemoteAddr -> ctx.RemoteAddr()
- r.RequestURI -> ctx.RequestURI()
- r.TLS -> ctx.IsTLS()
- r.Cookie() -> ctx.Request.Header.Cookie()
- r.Referer() -> ctx.Referer()
- r.UserAgent() -> ctx.Request.Header.UserAgent()
- w.Header() -> ctx.Response.Header
- w.Header().Set() -> ctx.Response.Header.Set()
- w.Header().Set("Content-Type") -> ctx.SetContentType()
- w.Header().Set("Set-Cookie") -> ctx.Response.Header.SetCookie()
- w.Write() -> ctx.Write(), ctx.SetBody(), ctx.SetBodyStream()
- w.WriteHeader() -> ctx.SetStatusCode()
- w.(http.Hijacker).Hijack() -> ctx.Hijack()
- http.Error() -> ctx.Error()
-
VERY IMPORTANT! Fasthttp disallows holding references to RequestCtx or to its' members after returning from RequestHandler. Otherwise data races are unevitable. Carefully inspect all the net/http request handlers converted to fasthttp whether they retain references to RequestCtx or to its' members after returning. RequestCtx provides the following band aids for this case:
- Wrap RequestHandler into TimeoutHandler.
- Call TimeoutError before returning from RequestHandler if there are references to RequestCtx or to its' members. See the example for more details.
Use brilliant tool - race detector - for detecting and eliminating data races in your program. If you detected data race related to fasthttp in your program, then there is high probability you forgot calling TimeoutError before returning from RequestHandler.
- Blind switching from net/http to fasthttp won't give you performance boost. While fasthttp is optimized for speed, its' performance may be easily saturated by slow RequestHandler. So profile and optimize your code after switching to fasthttp.
- Use reuseport listener.
- Run a separate server instance per CPU core with GOMAXPROCS=1.
- Pin each server instance to a separate CPU core using taskset.
- Ensure the interrupts of multiqueue network card are evenly distributed between CPU cores. See this article for details.
- Do not allocate objects and
[]byte
buffers - just reuse them as much as possible. Fasthttp API design encourages this. - sync.Pool is your best friend.
- Profile your program
in production.
go tool pprof --alloc_objects your-program mem.pprof
usually gives better insights for optimization opportunities thango tool pprof your-program cpu.pprof
. - Write tests and benchmarks for hot paths.
- Avoid conversion between
[]byte
andstring
, since this may result in memory allocation+copy. Fasthttp API provides functions for both[]byte
andstring
- use these functions instead of converting manually between[]byte
andstring
. - Verify your tests and production code under race detector on a regular basis.
The following tricks are used by fasthttp. Use them in your code too.
- Standard Go functions accept nil buffers
var (
// both buffers are uninitialized
dst []byte
src []byte
)
dst = append(dst, src...) // this is legal code
copy(dst, src) // this is legal code
(string(src) == "") // is true
(len(src) == 0) // is true
So throw away nil checks for []byte
buffers from you code. For example,
srcLen := 0
if src != nil {
srcLen = len(src)
}
becomes
srcLen := len(src)
- String may be appended to
[]byte
buffer withappend
dst = append(dst, "foobar"...)
- All fasthttp functions accept nil
[]byte
buffer
statusCode, body, err := fasthttp.Get(nil, "http://google.com/")
uintBuf := fasthttp.AppendUint(nil, 1234)
-
Why creating yet another http package instead of optimizing net/http?
Because net/http API limits many optimization opportunities. For example:
- net/http Request object lifetime isn't limited by request handler execution time. So the server must create new request object per each request instead of reusing existing objects like fasthttp do.
- net/http headers are stored in a
map[string][]string
. So the server must parse all the headers, convert them from[]byte
tostring
and put them into the map before calling user-provided request handler. This all requires unnesessary memory allocations avoided by fasthttp. - net/http client API requires creating new response object per each request.
-
Why fasthttp API is incompatible with net/http?
Because net/http API limits many optimization opportunities. See the answer above for more details. Also certain net/http API parts are suboptimal for use:
-
Why fasthttp doesn't support HTTP/2.0 and WebSockets?
There are plans for adding HTTP/2.0 and WebSockets support in the future. In the mean time, third parties may use RequestCtx.Hijack for implementing these goodies.
-
Are there known net/http advantages comparing to fasthttp?
Yes:
- net/http supports HTTP/2.0 starting from go1.6.
- net/http API is stable, while fasthttp API constantly evolves.
- net/http handles more HTTP corner cases.
- net/http should contain less bugs, since it is used and tested by much wider audience.
- Many existing web frameworks and request routers are built on top of net/http.
- net/http works on Go older than 1.5.
-
Which GO versions are supported by fasthttp?
Go1.5+. Older versions won't be supported, since their standard package miss useful functions.
-
Please provide real benchmark data and sever information
See this issue.
-
Are there plans to add request routing to fasthttp?
There are no plans to add request routing into fasthttp. I believe request routing must be implemented in a separate package(s) like httprouter. See this issue for more info.
-
I detected data race in fasthttp!
Cool! File a bug. But before doing this check the following in your code:
- Make sure there are no references to RequestCtx or to its' members after returning from RequestHandler.
- Make sure you call TimeoutError before returning from RequestHandler if there are references to RequestCtx or to its' members, which may be accessed by other goroutines.