[performance] simpler throttling logic (#2407)

* reduce complexity of throttling logic to use 1 queue and an atomic int

* use atomic add instead of CAS, add throttling test
This commit is contained in:
kim 2023-12-16 11:53:42 +00:00 committed by GitHub
parent 1312695c46
commit d56a8d095e
No known key found for this signature in database
GPG key ID: 4AEE18F83AFDEB23
4 changed files with 208 additions and 40 deletions

View file

@ -42,6 +42,12 @@ var (
StatusInternalServerErrorJSON = mustJSON(map[string]string{
"status": http.StatusText(http.StatusInternalServerError),
})
ErrorCapacityExceeded = mustJSON(map[string]string{
"error": "server capacity exceeded!",
})
ErrorRateLimitReached = mustJSON(map[string]string{
"error": "rate limit reached!",
})
EmptyJSONObject = mustJSON("{}")
EmptyJSONArray = mustJSON("[]")

View file

@ -29,6 +29,8 @@ import (
"github.com/superseriousbusiness/gotosocial/internal/util"
"github.com/ulule/limiter/v3"
"github.com/ulule/limiter/v3/drivers/store/memory"
apiutil "github.com/superseriousbusiness/gotosocial/internal/api/util"
)
const rateLimitPeriod = 5 * time.Minute
@ -141,10 +143,12 @@ func RateLimit(limit int, exceptions []string) gin.HandlerFunc {
if context.Reached {
// Return JSON error message for
// consistency with other endpoints.
c.AbortWithStatusJSON(
apiutil.Data(c,
http.StatusTooManyRequests,
gin.H{"error": "rate limit reached"},
apiutil.AppJSON,
apiutil.ErrorRateLimitReached,
)
c.Abort()
return
}

View file

@ -29,9 +29,12 @@ import (
"net/http"
"runtime"
"strconv"
"sync/atomic"
"time"
"github.com/gin-gonic/gin"
apiutil "github.com/superseriousbusiness/gotosocial/internal/api/util"
)
// token represents a request that is being processed.
@ -80,55 +83,61 @@ func Throttle(cpuMultiplier int, retryAfter time.Duration) gin.HandlerFunc {
}
var (
limit = runtime.GOMAXPROCS(0) * cpuMultiplier
backlogLimit = limit * cpuMultiplier
backlogChannelSize = limit + backlogLimit
tokens = make(chan token, limit)
backlogTokens = make(chan token, backlogChannelSize)
retryAfterStr = strconv.FormatUint(uint64(retryAfter/time.Second), 10)
limit = runtime.GOMAXPROCS(0) * cpuMultiplier
queueLimit = limit * cpuMultiplier
tokens = make(chan token, limit)
requestCount = atomic.Int64{}
retryAfterStr = strconv.FormatUint(uint64(retryAfter/time.Second), 10)
)
// prefill token channels
// prefill token channel
for i := 0; i < limit; i++ {
tokens <- token{}
}
for i := 0; i < backlogChannelSize; i++ {
backlogTokens <- token{}
}
return func(c *gin.Context) {
// inside this select, the caller tries to get a backlog token
select {
case <-c.Request.Context().Done():
// request context has been canceled already
// Always decrement request counter.
defer func() { requestCount.Add(-1) }()
// Increment request count.
n := requestCount.Add(1)
// Check whether the request
// count is over queue limit.
if n > int64(queueLimit) {
c.Header("Retry-After", retryAfterStr)
apiutil.Data(c,
http.StatusTooManyRequests,
apiutil.AppJSON,
apiutil.ErrorCapacityExceeded,
)
c.Abort()
return
case btok := <-backlogTokens:
}
// Sit and wait in the
// queue for free token.
select {
case <-c.Request.Context().Done():
// request context has
// been canceled already.
return
case tok := <-tokens:
// caller has successfully
// received a token, allowing
// request to be processed.
defer func() {
// when we're finished, return the backlog token to the bucket
backlogTokens <- btok
// when we're finished, return
// this token to the bucket.
tokens <- tok
}()
// inside *this* select, the caller has a backlog token,
// and they're waiting for their turn to be processed
select {
case <-c.Request.Context().Done():
// the request context has been canceled already
return
case tok := <-tokens:
// the caller gets a token, so their request can now be processed
defer func() {
// whatever happens to the request, put the
// token back in the bucket when we're finished
tokens <- tok
}()
c.Next() // <- finally process the caller's request
}
default:
// we don't have space in the backlog queue
c.Header("Retry-After", retryAfterStr)
c.JSON(http.StatusTooManyRequests, gin.H{"error": "server capacity exceeded"})
c.Abort()
// Process
// request!
c.Next()
}
}
}

View file

@ -0,0 +1,149 @@
// GoToSocial
// Copyright (C) GoToSocial Authors admin@gotosocial.org
// SPDX-License-Identifier: AGPL-3.0-or-later
//
// This program is free software: you can redistribute it and/or modify
// it under the terms of the GNU Affero General Public License as published by
// the Free Software Foundation, either version 3 of the License, or
// (at your option) any later version.
//
// This program is distributed in the hope that it will be useful,
// but WITHOUT ANY WARRANTY; without even the implied warranty of
// MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
// GNU Affero General Public License for more details.
//
// You should have received a copy of the GNU Affero General Public License
// along with this program. If not, see <http://www.gnu.org/licenses/>.
/*
The code in this file is adapted from MIT-licensed code in github.com/go-chi/chi. Thanks chi (thi)!
See: https://github.com/go-chi/chi/blob/e6baba61759b26ddf7b14d1e02d1da81a4d76c08/middleware/throttle.go
And: https://github.com/sponsors/pkieltyka
*/
package middleware_test
import (
"context"
"net/http"
"net/http/httptest"
"runtime"
"strconv"
"testing"
"time"
"github.com/gin-gonic/gin"
"github.com/superseriousbusiness/gotosocial/internal/middleware"
)
func TestThrottlingMiddleware(t *testing.T) {
testThrottlingMiddleware(t, 2, time.Second*10)
testThrottlingMiddleware(t, 4, time.Second*15)
testThrottlingMiddleware(t, 8, time.Second*30)
}
func testThrottlingMiddleware(t *testing.T, cpuMulti int, retryAfter time.Duration) {
// Calculate expected request limit + queue.
limit := runtime.GOMAXPROCS(0) * cpuMulti
queueLimit := limit * cpuMulti
// Calculate expected retry-after header string.
retryAfterStr := strconv.FormatUint(uint64(retryAfter/time.Second), 10)
// Gin test http engine
// (used for ctx init).
e := gin.New()
// Add middleware to the gin engine handler stack.
middleware := middleware.Throttle(cpuMulti, retryAfter)
e.Use(middleware)
// Set the blocking gin handler.
handler := blockingHandler()
e.Handle("GET", "/", handler)
var cncls []func()
for i := 0; i < queueLimit+limit; i++ {
// Prepare a gin test context.
r := httptest.NewRequest("GET", "/", nil)
rw := httptest.NewRecorder()
// Wrap request with new cancel context.
ctx, cncl := context.WithCancel(r.Context())
r = r.WithContext(ctx)
// Pass req through
// engine handler.
go e.ServeHTTP(rw, r)
time.Sleep(time.Millisecond)
// Get http result.
res := rw.Result()
if i < queueLimit {
// Check status == 200 (default, i.e not set).
if res.StatusCode != http.StatusOK {
t.Fatalf("status code was set (%d) with queueLimit=%d and request=%d", res.StatusCode, queueLimit, i)
}
// Add cancel to func slice.
cncls = append(cncls, cncl)
} else {
// Check the returned status code is expected.
if res.StatusCode != http.StatusTooManyRequests {
t.Fatalf("did not return status 429 (%d) with queueLimit=%d and request=%d", res.StatusCode, queueLimit, i)
}
// Check the returned retry-after header is set.
if res.Header.Get("Retry-After") != retryAfterStr {
t.Fatalf("did not return retry-after %s with queueLimit=%d and request=%d", retryAfterStr, queueLimit, i)
}
// Cancel on return.
defer cncl()
}
}
// Cancel all blocked reqs.
for _, cncl := range cncls {
cncl()
}
time.Sleep(time.Second)
// Check a bunchh more requests
// can now make it through after
// previous requests were released!
for i := 0; i < limit; i++ {
// Prepare a gin test context.
r := httptest.NewRequest("GET", "/", nil)
rw := httptest.NewRecorder()
// Pass req through
// engine handler.
go e.ServeHTTP(rw, r)
time.Sleep(time.Millisecond)
// Get http result.
res := rw.Result()
// Check status == 200 (default, i.e not set).
if res.StatusCode != http.StatusOK {
t.Fatalf("status code was set (%d) with queueLimit=%d and request=%d", res.StatusCode, queueLimit, i)
}
}
}
func blockingHandler() gin.HandlerFunc {
return func(ctx *gin.Context) {
<-ctx.Done()
ctx.Status(201) // specifically not 200
}
}