Winning at PHP in Production in 2025

The basics high level things you should be doing in 2025 to run a PHP application in production to avoid outages, 503s and carelessly leave performance on the table.

Talk was presented at Shopware Unconference 2025 in Cologne.

Transcript

1. Benjamin Eberlei Winning @ PHP in Production 2025 High-Performance, -Availability Benjamin Eberlei
2. Why care? Isn’t PHP getting faster every version?
4. Benchmarks show no gain during 8.x series
5. Benchmarks show no gain during 8.x series
6. Lets start with php.ini settings
10. Lets start with php.ini settings
12. More php.ini settings • display_errors=off • log_errors=on • error_log=/var/log/php/php_error.log • Looking for max time exceeded, max memory limit reached errors
13. Canonical Application Log Lines • https:/ /stripe.com/blog/canonical-log-l ines • key=value pairs in a text log • “Nginx/Apache log for Application”
15. PHP-FPM • Prefer pool “static” • Calculate max_children based on available memory • https:/ /tideways.com/tools/fpm-configu ration-calculator
17. CPU, I/O or Memory bound? • Watch Scheduler Statistics • https:/ /docs.kernel.org/scheduler/sche d-stats.html • If difference is 0 after request, no scheduling took place.
19. • fpm_get_status() • pm.status_path=/fpm-status
20. Multiple PHP-FPM Pools
23. Opcache Three critical settings to adjust:
24. Opcache Monitoring • opcache_get_status
25. Opcache Preloading • opcache.preload • Moves autoloading and include time “off-request”
26. JIT 🤷
27. Sessions • Default Filesystem handler does not scale beyond one application server • Be aware of Session locking
30. Sessions • Default Filesystem handler does not scale beyond one application server • Locking of sessions
31. Composer Optimize Autoload
32. File Storage • Avoid Network File System (NFS) for the whole application • only to share assets/files if you must
33. Timeouts Reduce the connect timeouts of everything, defaults break your neck • default_socket_timeout ✅ • Redis default 5 seconds • cURL indefinitely waiting ◦ A
34. Health Checks https:/ /github.com/spatie/laravel-health https:/ /github.com/laminas/laminas-diag nostics Are services + processes up?
35. Background Jobs • Use Symfony Messenger • Supervisord • Canonical log lines (again!) • Health Checks & Restart automatically
37. High Availability PHP • Load Balancing in front of app servers • Web servers only run php-fpm • Independent worker and cron servers • Databases and caches have their servers • each system redundant or clustered if you want to avoid single point of failure
38. InnoDB Buffer Pool Size is often too small
39. THANKS E-Mail: Benjamin@Tideways.com YouTube: https:/ /youtube.com/@TidewaysPHP Mastodon: https:/ /phpc.social/@beberlei