gin-limit

Stay under the limit with this handy Gin middleware.

Purpose

By default, http.ListenAndServe (which gin.Run wraps) will serve an unbounded number of requests. Limiting the number of simultaneous connections can sometimes greatly speed things up under load. Inspired by x/net/netutil.LimitListener.

Example

package main

import (
  "github.com/aviddiviner/gin-limit"
  "github.com/gin-gonic/gin"
)

func main() {
  r := gin.Default()
  r.Use(limit.MaxAllowed(20))
  // ...
  r.Run(":8080")
}

Lies, damned lies and statistics

Everyone loves synthetic benchmarks, so have some numbers from my 2015 Macbook (on a fast rendering page; single sqlite query, basic templates).

% wrk -t12 -c400 -d20s http://localhost:4560/
Running 20s test @ http://localhost:4560/
  12 threads and 400 connections
  Thread Stats   Avg      Stdev     Max   +/- Stdev
    Latency   848.63ms  525.39ms   1.64s    62.81%
    Req/Sec    45.43     61.85   360.00     90.50%
  8908 requests in 20.10s, 21.91MB read
  Socket errors: connect 0, read 219, write 0, timeout 0
Requests/sec:    443.19
Transfer/sec:      1.09MB

Now 10x faster with limit.MaxAllowed(3) (although that would be higher in the real world). Hooray!

% wrk -t12 -c400 -d20s http://localhost:4560/
Running 20s test @ http://localhost:4560/
  12 threads and 400 connections
  Thread Stats   Avg      Stdev     Max   +/- Stdev
    Latency    94.40ms   32.65ms 656.72ms   86.44%
    Req/Sec   351.61     84.32   666.00     79.32%
  84181 requests in 20.09s, 207.05MB read
  Socket errors: connect 0, read 165, write 0, timeout 0
Requests/sec:   4189.75
Transfer/sec:     10.30MB