惯性聚合 高效追踪和阅读你感兴趣的博客、新闻、科技资讯
阅读原文 在惯性聚合中打开

推荐订阅源

aimingoo的专栏
aimingoo的专栏
WordPress大学
WordPress大学
阮一峰的网络日志
阮一峰的网络日志
博客园 - 司徒正美
月光博客
月光博客
宝玉的分享
宝玉的分享
Recent Announcements
Recent Announcements
小众软件
小众软件
H
Hackread – Cybersecurity News, Data Breaches, AI and More
美团技术团队
博客园 - 三生石上(FineUI控件)
A
About on SuperTechFans
J
Java Code Geeks
云风的 BLOG
云风的 BLOG
罗磊的独立博客
大猫的无限游戏
大猫的无限游戏
IT之家
IT之家
Vercel News
Vercel News
量子位
Martin Fowler
Martin Fowler
OSCHINA 社区最新新闻
OSCHINA 社区最新新闻
V
Visual Studio Blog
腾讯CDC
有赞技术团队
有赞技术团队

Jiajun的技术笔记

你好,2026! TiDB 源码阅读(六):TiDB Coprocessor 源码解析 性能优化的核心思想 TiDB 源码阅读(五):索引 TiDB 源码阅读(四):AST、逻辑计划、物理计划 CockroachDB Serverless Architecture podman 无故退出 Cursor Control-L (CTRL-L) Keyboard Shortcuts in Terminal Replace docker with podman Using xmonad with xfce4 A RC script for freebsd frpc 自己动手写一个k8s controller AI 会取代你的(编程)岗位吗? 自建DERP服务器提升Tailscale连接速度(使用Nginx转发) 自动升级Docker容器 再读《程序员修炼之道-从小工到专家》 让浏览器下载文件 再读《软件随想录》/《黑客与画家》/《软技能》 HTTP 压力测试中的 Coordinated Omission 2的补码 编程语言中的 context 是什么? flutter macOS 构建出错 Flatpak 使用小记 Golang CAS 操作是怎么实现的 PostgreSQL 当MQ来使用 Clash 结合 工作VPN 的网络设计 使用 PostgreSQL 搭建 JuiceFS PostgreSQL 配置优化和日志分析 有GitHub Copilot?那就可以搭建你的ChatGPT4服务 窗口函数的使用(以PG为例)
Memory leak in net/http
Jiajun Huang · 2018-11-24 · via Jiajun的技术笔记

English is not my mother languange, help me to improve this if you’d like, thanks!

What happened?

Recently we’ve get stuck with a problem that we’re serving file download service in Go, but the service is killed by server because of OOM, so we decide to dig out the reason.

we are: @jiajunhuang, @lailin

Code in server is like this:

package main

import (
	"io/ioutil"
	"log"
	"net/http"
	"runtime"
	"time"

	_ "net/http/pprof"
)

func main() {
	go func() { // here, this goroutine is for debug
		for {
			println("gonna gc")
			runtime.GC()
			time.Sleep(time.Second * 30)
		}
	}()

	http.Handle("/download", http.HandlerFunc(func(w http.ResponseWriter, req *http.Request) {
		data, err := ioutil.ReadFile("/Users/jiajun/Images/ubuntu-18.04.1-live-server-amd64.iso")
		if err != nil {
			log.Panicf("failed to read file: %s", err)
		}
		w.Header().Set("Connection", "close")
		w.Write(data)
	}))

	log.Fatal(http.ListenAndServe(":8080", nil))
}

and client code:

package main

import (
	"log"
	"net/http"
)

func main() {
	client := &http.Client{}

	req, err := http.NewRequest("GET", "http://127.0.0.1:8080/download", nil)
	if err != nil {
		log.Panicf("error: %s", err)
	}

	req.Header.Set("Content-Type", "application/x-www-form-urlencoded")
	req.Header.Set("Connection", "close")
	resp, err := client.Do(req)
	defer resp.Body.Close()

	println("file received")
	select {} // client will block here
}

run the server and client, we then use go tool pprof to inspect the memory usage:

$ go run server.go &
$ go run client.go &
$ go tool pprof http://localhost:8080/debug/pprof/heap

and we got this:

memory leak

although we call runtime.GC periodically, the memory is still hold by Golang runtime, until we close the client:

memory leak2

Reason

Although we’ve set Connection: close header, but as RFC2616 says:

HTTP/1.1 defines the “close” connection option for the sender to signal that the connection will be closed after completion of the response.

it does not speficy who is responsible for close the connection, so Golang wish the client to close the connection.

func (w *response) Write(data []byte) (n int, err error) {
	return w.write(len(data), data, "")
}

func (w *response) write(lenData int, dataB []byte, dataS string) (n int, err error) {
	if w.conn.hijacked() {
		if lenData > 0 {
			caller := relevantCaller()
			w.conn.server.logf("http: response.Write on hijacked connection from %s (%s:%d)", caller.Function, path.Base(caller.File), caller.Line)
		}
		return 0, ErrHijacked
	}
	if !w.wroteHeader {
		w.WriteHeader(StatusOK)
	}
	if lenData == 0 {
		return 0, nil
	}
	if !w.bodyAllowed() {
		return 0, ErrBodyNotAllowed
	}

	w.written += int64(lenData) // ignoring errors, for errorKludge
	if w.contentLength != -1 && w.written > w.contentLength {
		return 0, ErrContentLength
	}
	if dataB != nil {
		return w.w.Write(dataB)
	} else {
		return w.w.WriteString(dataS)
	}
}

and the writer use a byte slice to hold the data:

// Write writes the contents of p into the buffer.
// It returns the number of bytes written.
// If nn < len(p), it also returns an error explaining
// why the write is short.
func (b *Writer) Write(p []byte) (nn int, err error) {
	for len(p) > b.Available() && b.err == nil {
		var n int
		if b.Buffered() == 0 {
			// Large write, empty buffer.
			// Write directly from p to avoid copy.
			n, b.err = b.wr.Write(p)
		} else {
			n = copy(b.buf[b.n:], p)
			b.n += n
			b.Flush()
		}
		nn += n
		p = p[n:]
	}
	if b.err != nil {
		return nn, b.err
	}
	n := copy(b.buf[b.n:], p)
	b.n += n
	nn += n
	return nn, nil
}

How to solve it

  • Write files in chunk, instead of read large files into memory