LNMP Server Architecture Troubleshooting and Optimization: Nginx and PHP‑FPM Tuning

The article begins with an overview of common enterprise web server stacks (LNMP/LAMP) and notes the industry trend of replacing Apache with Nginx for its superior performance and built‑in high‑availability features.

A specific incident is described: backend users experienced slow responses and occasional 502 errors. Initial checks of network connectivity proved insufficient, leading to a deeper investigation of the LNMP stack.

Step 1 – Nginx diagnostics: The error log was examined with $ cat /usr/local/nginx/logs/error.log | grep error , which showed no issues. Access logs were then inspected using $ cat /var/log/access_nging.log | grep error , revealing multiple 502 entries.

Step 2 – Nginx file‑descriptor limit: The default worker_rlimit_nofile of 4096 was increased to 51200 in vim /usr/local/nginx/conf/nginx.conf and the worker connections were set to the same value. Nginx was reloaded with service nginx reload .

Step 3 – System limits: The OS limit was checked via ulimit –n (still 4096). The file vim /etc/security/limits.conf was edited to set: soft nofile 65535 hard nofile 65535 soft nproc 65535 hard nproc 65535 These values exceed the Nginx limits, ensuring the kernel permits the higher descriptor count.

Step 4 – FastCGI timeout: FastCGI parameters (connect, send, read) were increased to 300 seconds in nginx.conf , then Nginx was reloaded again.

After these changes, the site became reachable, though still slightly slow, prompting a focus on PHP‑FPM.

Step 5 – PHP‑FPM log inspection: The PHP log was checked with tail -n 100 /usr/local/php/var/log/php-fpm.log , revealing a warning: WARNING: [pool www] seems busy (you may need to increase pm.start_servers, or pm.min/max_spare_servers) .

Step 6 – PHP‑FPM parameter tuning: The article explains the meaning of dynamic vs. static modes and the relevant parameters (pm.max_children, pm.start_servers, pm.min_spare_servers, pm.max_spare_servers). For an 8 GB server, a static mode with a max of 125‑150 children was chosen, and PHP‑FPM was reloaded via service php-fpm relod . The number of PHP processes was verified with netstat -anpo | grep php-fpm | wc -l , returning 128 .

Step 7 – Result verification: After reloading, page load times improved markedly, the warning disappeared from the logs, and backend access stabilized.

Step 8 – Load testing: Performance was measured using ApacheBench: ab -n 1000000 -c 10000 (one PHP file). The article explains the meaning of the -n (total requests) and -c (concurrent users) options and encourages iterative testing to locate bottlenecks.

Overall, the case demonstrates a systematic approach to diagnosing and optimizing an LNMP server stack, covering both Nginx and PHP‑FPM configuration, OS limits, and load‑testing techniques.