dykyi-roman PHP-FPM Configuration Optimization
Provides production-ready PHP-FPM tuning for Docker containers based on workload type and available resources.
Process Manager Modes
┌─────────────────────────────────────────────────────────────────┐
│ PHP-FPM PROCESS MANAGERS │
├─────────────────────────────────────────────────────────────────┤
│ STATIC: [W][W][W][W][W][W][W][W] Fixed pool │
│ Best for: dedicated containers, predictable load │
│ DYNAMIC: [W][W][W][W][ ][ ][ ][ ] Scales up/down │
│ Best for: variable load, shared resources │
│ ONDEMAND: [ ][ ][ ][ ][ ][ ][ ][ ] Starts on request │
│ Best for: low-traffic, dev environments │
└─────────────────────────────────────────────────────────────────┘
Memory Calculation Formula
max_children = (available_memory - system_overhead) / avg_worker_memory
Where:
available_memory = container_memory_limit
system_overhead = ~50MB (OS + PHP-FPM master + nginx)
avg_worker_memory = 30-60MB (depends on application)
| Container Memory | System Overhead | Worker Memory | max_children |
|---|---|---|---|
| 256MB | 50MB | 40MB | 5 |
| 512MB | 50MB | 40MB | 11 |
| 1024MB | 50MB | 40MB | 24 |
| 2048MB | 50MB | 50MB | 39 |
Measure actual worker memory:
ps aux | grep "php-fpm: pool" | awk '{sum+=$6; n++} END {print sum/n/1024 " MB"}'
Configuration by Workload
API Workload (Many Fast Requests)
[www]
pm = static
pm.max_children = 20
request_terminate_timeout = 30s
request_slowlog_timeout = 5s
pm.max_requests = 1000
slowlog = /proc/self/fd/2
Worker Workload (Few Long Requests)
[www]
pm = dynamic
pm.max_children = 8
pm.start_servers = 2
pm.min_spare_servers = 1
pm.max_spare_servers = 4
request_terminate_timeout = 300s
request_slowlog_timeout = 30s
pm.max_requests = 200
Mixed Workload
[www]
pm = dynamic
pm.max_children = 15
pm.start_servers = 5
pm.min_spare_servers = 3
pm.max_spare_servers = 8
request_terminate_timeout = 60s
request_slowlog_timeout = 10s
pm.max_requests = 500
Dynamic Mode Formulas
pm.start_servers = (min_spare + max_spare) / 2
pm.min_spare_servers = max_children * 0.25
pm.max_spare_servers = max_children * 0.75
pm.max_requests by Application Type
| Application Type | Recommended Value |
|---|---|
| Stateless API | 1000-5000 |
| Framework with ORM | 500-1000 |
| Legacy application | 100-500 |
| Memory-intensive processing | 50-200 |
Health Check and Monitoring
pm.status_path = /status
ping.path = /ping
ping.response = pong
HEALTHCHECK --interval=30s --timeout=5s --start-period=10s --retries=3 \
CMD php-fpm-healthcheck || exit 1
Key status metrics to monitor:
- active processes: current load
- listen queue: requests waiting (should be 0)
- max children reached: need more workers
- slow requests: performance issues
Docker-Specific Settings
; Logging to stdout/stderr
error_log = /proc/self/fd/2
access.log = /proc/self/fd/2
; Must be "no" in containers: workers inherit vars from docker run -e / compose environment
clear_env = no
catch_workers_output = yes
decorate_workers_output = no
Before/After Comparison
| Metric | Default | Optimized (static) | Improvement |
|---|---|---|---|
| max_children | 5 | 20 | +300% capacity |
| Idle memory (512MB) | 200MB wasted | ~50MB overhead | -75% waste |
| Request latency (p99) | 500ms | 150ms | -70% |
| Throughput (req/s) | 50 | 200 | +300% |
| Memory leak recovery | Never | Every 1000 req | Predictable |
Complete Production Configuration
[global]
error_log = /proc/self/fd/2
log_level = warning
daemonize = no
[www]
user = www-data
group = www-data
listen = 0.0.0.0:9000
pm = static
pm.max_children = 20
pm.max_requests = 1000
request_terminate_timeout = 30s
request_slowlog_timeout = 5s
slowlog = /proc/self/fd/2
pm.status_path = /status
ping.path = /ping
ping.response = pong
clear_env = no
catch_workers_output = yes
decorate_workers_output = no
access.log = /proc/self/fd/2
access.format = "%R - %u %t \"%m %r\" %s %{mili}dms %{mega}MMB %C%%"
Generation Instructions
- Determine workload type: API, Worker, or Mixed
- Calculate max_children: Based on container memory and worker memory
- Select PM mode: Static for production, dynamic for variable load
- Configure timeouts: Based on expected request duration
- Set max_requests: Based on application memory behavior
- Enable monitoring: Status page, health check, slow log