慣性聚合 高效追讀感興趣之博客、新聞、科技資訊
閱原文 以慣性聚合開啟

推薦訂閱源

博客园 - 司徒正美
V
V2EX
T
Tailwind CSS Blog
有赞技术团队
有赞技术团队
aimingoo的专栏
aimingoo的专栏
Apple Machine Learning Research
Apple Machine Learning Research
IT之家
IT之家
Blog — PlanetScale
Blog — PlanetScale
A
About on SuperTechFans
月光博客
月光博客
T
The Blog of Author Tim Ferriss
宝玉的分享
宝玉的分享
Martin Fowler
Martin Fowler
博客园 - 聂微东
The GitHub Blog
The GitHub Blog
V
Visual Studio Blog
WordPress大学
WordPress大学
酷 壳 – CoolShell
酷 壳 – CoolShell
Engineering at Meta
Engineering at Meta
GbyAI
GbyAI

DEV Community

Authentication Security Deep Dive: From Brute Force to Salted Hashing (With Java Examples) Why AI Systems Don’t Fail — They Drift Spilling beans for how i learn for exam😁"Reinforcement Learning Cheat Sheet" I Replaced Chrome with Safari for AI Browser Automation. Here's What Broke (and What Finally Worked) How Python Borrows Other People's Work The $40 Architecture: Processing 1 Billion API Requests with 99.99% Uptime Vibe Coding: A Workflow Guide (From Zero to SaaS) Most webhook security guides protect the wrong side. The scary part is delivery. Headless CMS for TanStack Start: Build a Blog with Cosmic EU Age Verification App "Hacked in 2 Minutes" — What Actually Happened Comfy Cloud’s delete function does not actually remove files Running AI Models on GPU Cloud Servers: A Beginner Guide Event-driven media intelligence with AWS Step Functions and Bedrock I scored 500 AI prompts across 8 quality dimensions — here's what broke How to Call Google Gemini API from Next.js (Free Tier, No Backend Needed) The Portal Protocol: Reclaiming Human Connection in the Age of AI How to Fix Your Team's Scattered Knowledge Problem With a Self-Hosted Forum Intro to tc Cloud Functors: A Graph-First Mental Model for the Modern Cloud Designing Multi-Tenant Backends With Both Ownership and Team Access I Built a Neumorphic CSS Library with 77+ Components — Here's What I Learned PostgreSQL Performance Optimization: Why Connection Pooling Is Critical at Scale Cómo construí un SaaS multi-rubro para gestionar expensas en Argentina con FastAPI + Vue 3 🚀 I Built an Ethical Hacking Scanner Tool – Open Source Project I Replaced /usage and /context in Claude Code With a Single Statusline A Pythonic Way to Handle Emails (IMAP/SMTP) with Auto-Discovery and AI-Ready Design I Collected 8.9 Million Polymarket Price Points — Here's What I Found About How Markets Really Move EcoTrack AI — Carbon Footprint Tracker & Dashboard Everyone's Using AI. No One Agrees How. 5 self-hosted ebook managers worth trying in 2026 Building Your First AI Agent with LangChain: From Chatbot to Autonomous Assistant Common SOC 2 Failures (Real World) Stop Vibe-Checking Your AI App: A Practical Guide to Evals How to Use SonarQube and SonarScanner Locally to Level Up Your Code Quality Your Next To-Do App Is Dead — I Replaced Mine with an OpenClaw AI Sign a Nostr event in 60 lines of Python using coincurve — no nostr-sdk, no nbxplorer, no rust toolchain ITGC Audit Explained Like You’re in Big 4 Patch Tuesday abril 2026: Microsoft parcha 163 vulnerabilidades y un zero-day en SharePoint Stop scraping everything: a better way to track competitor price changes Listing on MCPize + the Official MCP Registry while routing payments OUTSIDE the marketplace — how I kept 100% of my x402 revenue Building an AI-Powered Risk Intelligence System Using Serverless Architecture Why We Ripped Function Overloading Out of Our AI Toolchain Testing AI-Generated Code: How to Actually Know If It Works SaaS Churn Is Killing Your Business. Here Is What to Do About It (Without a Support Team) The Speed of AI Is No Longer Linear - And Self-Improving Models Are Why How to Implement RBAC for MCP Tools: A Practical Guide for Engineering Teams From Standard Quote to Persuasive Proposal: AI Automation for Arborists I built a CLI that scaffolds complete multi-tenant SaaS apps Axios CVE-2025–62718: The Silent SSRF Bug That Could Be Hiding in Your Node.js App Right Now The dashboard that ended our friendship Data Pipelines Explained Simply (and How to Build Them with Python)
《php-fpm调优秘笈:五项设定决p99》
Gabriel Anha · 2026-05-24 · via DEV Community

汝之Laravel应用,负重则缓。CPU恒居30%,内存无碍。数据库未觉疲。然p99于周朝之晨仍骤升至8秒。

尔启Grafana。尔察询。尔察缓存之击率。尔增Redis节点。无有变改。

其弊非尔之码。乃php-fpm。详言之,www.conf中五行,载有值,乃为Raspberry Pi所择。

六十秒之池模型

php-fpm者,运行工作进程之池也。网请求至Nginx,Nginx转而经FastCGI至池,一工者取之,运行汝之PHP码,复响应而还,乃待次请求。

其要义在:工者几何,何时而生,何时而复。

汝择三者之一于www.conf

; /etc/php/8.4/fpm/pool.d/www.conf
pm = dynamic

Enter fullscreen mode 退出全屏模式

  • pm = static:固定工作者数量,常存不灭。内存可预知。无生成延迟。流量低谷时耗损RAM。
  • pm = dynamic:设最小最大范围,依需求伸缩。多发行版默认此法。边界处有生成延迟。
  • pm = ondemand:请求至则生工作者,闲置超时则灭。内存最低。冷请求增5-50毫秒。

为专一用途之API服务器,应真通流,static几为正道。骤发之际,生成本为用户所受最劣迟滞。后当复论此。

设一:pm.max_children(众团队多误之)

此乃上限之数。言此箱中可同时运行几PHP之进程。若设之过低,则请求积压;若设之过高,则内核将杀汝之工作者及Redis之副车。

诸发行版之文皆略此公式:

pm.max_children = (total_RAM - reserved_for_OS_and_sidecars) / avg_worker_memory

入全屏模式 出全屏模式

汝需测量而非臆度之二值:

  1. 工作者内存之平均:久居之工,经手数事,其实用之 RSS 也,非初启之值。工者渐长,随自载之器载类,缓存渐热,Laravel 濡养容器而增。须待实热之后,乃可量之。
# settled RSS per worker, in MB
ps -ylC php-fpm8.4 --sort:rss \
  | awk 'NR>1 { sum+=$8; n++ } END { printf "avg=%.1f MB\n", sum/n/1024 }'

Enter fullscreen mode Exit fullscreen mode

素之Laravel十二工者,常五十至八十兆。二百途之巨构,兼Telescope与数Filament之资,则止百二十至百八十兆。工司繁象之变,可瞬息逾四百兆。

  1. 预存之RAM:其操作系统本身,Nginx,Redis若在本地,OPcache共享内存(默认128MB),若有APM代理。于4GB之机,通常需预留700MB至1GB,待php-fpm分得一杯羹。

:例证。c5.large EC2:4GB内存,2个vCPU,Ubuntu,Nginx+php-fpm+本地Redis。

  • :总计:4096MB
  • :预留(操作系统+Nginx+Redis+OPcache+缓冲):约900MB
  • 可备 php-fpm:约三千二百MB
  • 测得平均工作进程RSS:百四十MB
  • pm.max_children = 3200 / 140 = ~22
pm = static
pm.max_children = 22

入全屏模式 出全屏模式

此箱常见之形:有人寄之pm.max_children = 50 盖因某Stack Overflow之答,载于二〇一四年云。交通至,其二十三工推之入换页,迟滞陡升,而值班工程师增工,以为可助。非也。反甚。

CPU之侧验。 若汝之平均请求,系于CPU而短(不及五十毫秒),则增工无益。vCPU * 2-4,彼辈徒然鞭挞调度之器。于c5.large,若受制于CPU,则工作者最多四至八人。择其较小者,RAM抑或CPU是也。

二之设:pm.max_requests(内存漏失之保险)

PHP无C之漏失,然php-fpm 工者实自增。乐府自载,缓存之器,静置之盒,己之码中偶成之静。五十万请之後,初起八十兆之工,或至二百二十兆而不复。

pm.max_requests 乃于N请之後更工。其去也清,php-fpm生新易之。

pm.max_requests = 500

入全幅之境 出全幅之境

值之宜者,高则回收之费不可见,低则漏泄之患不须忧。凡应用之属,多在五百至千之数。库中之码,若分配之甚厉(如生成PDF、图像之处理、具缓存层之HTTP客户端),则宜取其下限。纤微之API工作者,可运行于二千以上。

然有险处:pm.max_requests = 0(部分发行版之默认设置)者,谓工作者永不复用。若汝设为0而未尝显设,则内存增容,汝未知也。宜设之。纵2000亦胜于0

设三:request_terminate_timeout(实发之超时)

此乃胜出之超时。

PHP有max_execution_timephp.iniNginx有fastcgi_read_timeout在其服务器区块中。php-fpm有request_terminate_timeoutwww.conf有三时之限,皆可绝同请,而最短者胜。

request_terminate_timeout = 30s

入全景模式 出全屏模式

何故失序:慢查询耗六十秒。fastcgi_read_timeout乃六十年耳。max_execution_time 者三十秒而居 sleep() 之中,或为阻塞之套接字调用,故不能发(PHP惟于指令边界检计时器)。请求使工作者悬停六十秒。复行二十二次,则全池皆陷于同一破败之源。

request_terminate_timeout 乃由 php-fpm 之主以信号施之,故无论工作者何为,皆能发。此乃惟一可确实释出陷顿工作者之超时。

悉设三。存之。request_terminate_timeout略高於max_execution_time故PHP得机以掷之Maximum execution time exceeded且于主核轰击工核之前,录栈踪于日志。

; www.conf
request_terminate_timeout = 35s

入全屏模式 出全屏模式

; php.ini
max_execution_time = 30

入全景模式 出全屏模式

# Nginx server block
fastcgi_read_timeout = 40s;

入全景模式 退出全屏模式

其序:PHP投掷于卅秒(汝得栈踪),php-fpm毙于卅五秒(工者释),Nginx闭连于四十分(请归五百四)。各层皆前一层之失所倚。

设四:pm.process_idle_timeout(唯于需时相关)

若在pm = dynamicpm = static,可略此节。乃虚设耳。

pm = ondemand设,工者应首请而生,无所为则pm.process_idle_timeout秒而亡。

pm = ondemand
pm.max_children = 22
pm.process_idle_timeout = 60s
pm.max_requests = 500

入全屏模式 出全屏模式

所陷:时过短焉。process_idle_timeout,其扰生灭,乃php-fpm之费事至极。六十息之限,于稀疏之务,诚宜以释内存于静时。下及十息,几无当者。

,其时实用ondemand 一箱之中设五十池,多者沉寂。有闲役者应突涌之时序。有试场之境,内存重于迟滞。

永勿用 ondemand 任何面向用户且具实际流量的应用。工作者(自动加载器、操作缓存水合、若非使用Octane则需框架初始化)的诞生成本为50至200毫秒。此乃PHP代码运行单行前所耗尽之全部延迟预算。static 若以更严密的pm.max_children为代价,则此权衡更优。

第五设置:listen.backlog(那无人触碰的SYN队列旋钮)

此乃众人所未知,而决交通高峰时君之行止者也.

listen.backlog此乃套接字接受队列之巨细。当诸役者皆忙,而Nginx欲移他请时,此连接坐于队列中,待役者得闲.

listen = /run/php/php8.4-fpm.sock
listen.backlog = 4096

入全屏模式 出全屏模式

諸多分發系統之預設值為511。於能容22並行工作者處理50毫秒請求之機器,滿載突發時可將511排隊連接約1.2秒內消弭。其後,新連接即遭拒絕,以connect() failed (11: Resource temporarily unavailable)

汝已見此於Nginx之記錄。似若php-fpm崩潰。實非。乃排隊溢出也。

listen.backlog增至4096或8192:

listen.backlog = 8192

入全景模式 出全屏模式

核之顶,噬之。 listen.backlog默然限于此与核之小者。net.core.somaxconn于旧时Linux之系统somaxconn默认为128。汝可设之。listen.backlog = 65535www.conf而实际队列之大小,乃128也。

核值验之。

sysctl net.core.somaxconn

入全景模式 出全景模式

若其低于汝之php-fpm设定,则举之

# /etc/sysctl.d/99-php-fpm.conf
net.core.somaxconn = 8192
net.ipv4.tcp_max_syn_backlog = 8192

入全景模式 出全景模式

sudo sysctl -p /etc/sysctl.d/99-php-fpm.conf

入全景模式 出全景模式

此更易之后,当重启php-fpm,使监听者得接新之待办。重载不足,必复造此套接字。

sudo systemctl restart php8.4-fpm

入全景模式 出全屏模式

谛听生产中队列之深浅以ss

# Recv-Q on the LISTEN socket = current queue depth
ss -lntp | grep php-fpm

入全景模式 出全屏模式

倘若有Recv-Q 之时,众工未克相随。积压之务,暂缓其时,非解通量之策。其解也,需增众工,速其码,或用Octane。

真实调优之程,于c5.large,Laravel 12

之设:c5.large,二CPU,四GB RAM,Laravel 12之应用,Telescope已启,MySQL别置一箱,Redis在本地,Nginx加php-fpm 8.4。

之先:

pm = dynamic
pm.max_children = 50
pm.start_servers = 10
pm.min_spare_servers = 5
pm.max_spare_servers = 20
pm.max_requests = 0
request_terminate_timeout = 0
listen.backlog = 511

入全景模式 退出全屏模式

k6,200 RPS,5分钟载入测试。p99达4800毫秒。运行期间,稳态RAM占92%,OOM-killer触发两次。

后:

pm = static
pm.max_children = 20
pm.max_requests = 800
request_terminate_timeout = 35s
listen = /run/php/php8.4-fpm.sock
listen.backlog = 8192

进入全屏模式 退出全屏模式

# /etc/sysctl.d/99-php-fpm.conf
net.core.somaxconn = 8192

进入全屏模式 退出全屏模式

同负荷测试。九九分之百于三百八十毫秒。内存稳定在七八成。无内存溢出。吞吐量提升,盖因箱体已停分页。

其训:pm.max_children = 50于四吉字节之箱,非"扩展",实乃于交换区中积压请求数。减至二十则诸事皆速。

何时当转用Octane乎

辛烷(Swoole或FrankenPHP后端)使Laravel应用在请求间驻留于内存。自动加载不复行,服务容器不重建,中间件栈保持完整。那30-80毫秒的框架启动成本归零矣。

若汝之php-fpm工者,六十毫秒耗于框架之启动,四十毫秒用于实请求之劳,则Octane使汝于同此硬件上,得增通量二点五倍。此非调优之进,乃架构之变也。

Octane需汝之码慎之又慎。静态状态存于请求间,容器单例恒为单例跨请求,凡假定为新进程者,今皆谬误。高流量应用,此工实值为之。CMS若每秒十请求,php-fpm调适得宜,已足矣。

五设若此,可至八分之能,未改架构之先。Octane得余二分。先调适,后决断。

汝之今状何如?pm.max_children吾何故择此?请于注解中列汝之数字(RAM、平均工作者RSS、汝所定之值)。好奇众人之默认设置相去几何。


若此有益

php-fpm之调适,得箱体之宜,然箱体实乃应用规模渐缓之微末。其余之份(服务界限、框架耦合、决团队能否自调代码之层级),乃解耦PHP之所指。此乃架构层级,一旦逾越框架之常,代码库所求之阶也。

Decoupled PHP — Clean and Hexagonal Architecture for Applications That Outlive the Framework

有Kindle、平装及精装版。英、德、日语版已出,葡、西语版将即至。