Rate Limiting

Usage guide for Kitex Default and Custom rate limiting.

Rate limiting is an imperative technique to protect server, which prevents server from overloaded by sudden traffic increase from a client.

Kitex supports the user-defined QPS limiter and connections limiter, and provides default implementation.

Note

  • When server.WithLimit and server.WithQPSLimiter or server.WithLimit and server.WithConnectionLimiter are used at the same time, only the latter will take effect.
  • To save request deserialization overhead and improve performance, in non-multiplexing scenarios, Kitex’s default QPS limiter takes effect at the OnRead hook, while in multiplexing or user-defined QPS limiter scenarios, the current limiter takes effect at the OnMessage hook. This is to ensure that the user-defined QPS limiter can obtain the basic information of the request such as rpc method.
  • Rate limiting only works on Thrift or Kitexpb protocols, but not for gRPC protocols. gRPC can use flow control to limit traffic at transport layer. Kitex provides WithGRPCInitialWindowSize and WithGRPCInitialConnWindowSize to set the flow control window size of stream and connection respectively. For details, see gRPC official documentation

Use default rate limiter

code example

import "github.com/cloudwego/kitex/pkg/limit"

func main() {
	svr := xxxservice.NewServer(handler, server.WithLimit(&limit.Option{MaxConnections: 10000, MaxQPS: 1000}))
	svr.Run()
}

Parameter description：

  • MaxConnections: max connections

  • MaxQPS: max QPS (Queries Per Second)

  • UpdateControl: provide the ability to modify the rate limit threshold dynamically, for example:

import "github.com/cloudwego/kitex/pkg/limit"

// define your limiter updater to update limit threshold
type MyLimiterUpdater struct {
	updater limit.Updater
}

func (lu *MyLimiterUpdater) YourChange() {
	// your logic: set new option as needed
	newOpt := &limit.Option{
		MaxConnections: 20000,
		MaxQPS:         2000,
	}
	// update limit config
	isUpdated := lu.updater.UpdateLimit(newOpt)
	// your logic
}

func (lu *MyLimiterUpdater) UpdateControl(u limit.Updater) {
	lu.updater = u
}

//--- init server ---
var lu  = MyLimiterUpdater{}
svr := xxxservice.NewServer(handler, server.WithLimit(&limit.Option{MaxConnections: 10000, MaxQPS: 1000, UpdateControl: lu.UpdateControl}))

Implementation

The default ConcurrencyLimiter and RateLimiter are used respectively to limit max connection and max QPS.

  • ConcurrencyLimiter：a simple counter；
  • RateLimiter：token bucket algorithm is used here.

Monitoring

The default limiters define the LimitReporter interface, which is used by rate limiting status monitoring, e.g. connection overloaded, QPS overloaded, etc.

Users may implement this interface and inject this implementation by WithLimitReporter if required.

// LimitReporter is the interface define to report(metric or print log) when limit happen
type LimitReporter interface {
    ConnOverloadReport()
    QPSOverloadReport()
}

Use user-defined rate limiter

import (
    "context"
    "time"

    "github.com/cloudwego/kitex/pkg/limiter"
    "github.com/cloudwego/kitex/pkg/rpcinfo"
    "github.com/cloudwego/kitex/server"
)

type qpsLimiter struct{}

func (l *qpsLimiter) Acquire(ctx context.Context) bool {
    ri := rpcinfo.GetRPCInfo(ctx)
    md := ri.From().Method()
    return acquire(md) // return true to allow this request
}

func (l *qpsLimiter) Status(ctx context.Context) (max, current int, interval time.Duration) {
    // max: the maximum number of requests allowed in the interval;
    // current: the remaining number of requests allowed in the interval;
    return
}

type connectionLimiter struct{}

func (l *connectionLimiter) Acquire(ctx context.Context) bool {
    ri := rpcinfo.GetRPCInfo(ctx)
    addr := ri.From().Address()
    return acquire(addr) // return true to allow this connection
}

func (l *connectionLimiter) Release(ctx context.Context) {
    ri := rpcinfo.GetRPCInfo(ctx)
    addr := ri.From().Address()
    return release(addr) // release occupied resource by the connection, only called after the release is successful.
}

func (l *connectionLimiter) Status(ctx context.Context) (limit, occupied int) {
    // limit: the maximum number of connections allowed.
    // occupied: the number of existing connections.
    return
}

func main() {
    myQPSLimiter := &qpsLimiter{}
    myConnectionLimiter := &connectionLimiter{}
    svr := xxxservice.NewServer(handler, server.WithQPSLimiter(myQPSLimiter), server.WithConnectionLimiter(myConnectionLimiter))
    svr.Run()
}

Last modified September 18, 2024 : Update typo in hertz docs (#1138) (12492e4)