Production Tuning

Running PHP in a long-lived process unlocks massive performance gains, but requires adjusting your server configuration to ensure system resources are fully optimized.

Operating System Configuration

Open Files Limit (nofile)

Each TCP socket, database connection, and file handle consumes a file descriptor. The default limit (often 1024) causes "too many open files" errors under load.

Edit /etc/security/limits.conf:

* soft nofile 65535
* hard nofile 65535

Apply immediately to your session:

ulimit -n 65535

For systemd services, add to the [Service] block:

LimitNOFILE=65535

TCP Backlog

For high-concurrency deployments, increase the kernel connection queue:

sysctl -w net.core.somaxconn=65535
sysctl -w net.ipv4.tcp_max_syn_backlog=65535

Persist in /etc/sysctl.conf:

net.core.somaxconn=65535
net.ipv4.tcp_max_syn_backlog=65535

---

PHP OPCache Optimization

In production, disable file modification checks entirely to eliminate disk syscalls on every request.

Edit php.ini:

opcache.enable=1
opcache.enable_cli=1
opcache.memory_consumption=256
opcache.interned_strings_buffer=16
opcache.max_accelerated_files=20000
opcache.validate_timestamps=0   ; Never check file modification time
opcache.save_comments=1
opcache.jit=tracing
opcache.jit_buffer_size=128M

Cache config and routes:

php nexphant cache:config
composer dump-autoload -o

---

Memory Limit

Nexphant workers are long-lived. Set a generous but bounded memory limit:

memory_limit=512M

Monitor actual usage via /api/metrics. Raise if workers are hitting the limit; lower if workers are idle and memory is scarce.

---

Garbage Collection (GC)

GC ticks in a persistent runtime can introduce latency spikes during high-throughput peaks.

Nexphant's Adaptive GC defers cycle collection to quiet traffic periods automatically. No configuration needed for most apps.

For high-allocation background jobs, trigger manually inside idle callbacks:

gc_collect_cycles();

Disable refcount-based GC entirely only if you have confirmed zero circular references and are comfortable managing memory manually:

zend.enable_gc=0

---

DB Connection Pooling

Traditional PHP opens and closes connections per request. Nexphant maintains persistent pools:

use Nexphant\Database\DB;

DB::configurePool([
    'min_connections' => 10,
    'max_connections' => 100,
    'idle_timeout'    => 30.0, // seconds
]);

Sizing rule:

DB max_connections ≥ nodes × workers × pool_max_connections

• Set idle_timeout to release unused connections during low traffic.
• Monitor pool exhaustion via ConnectionPool::stats().

---

Worker Count

Start with workers = CPU core count. For I/O-heavy workloads (many DB queries, external API calls), try CPU cores × 2.

$app->workers((int) env('APP_WORKERS', 4));

Tune based on pressure metrics:

curl http://localhost:8080/api/metrics | grep pressure

• pressure > 0.8 consistently → add workers or increase max_accept_per_tick
• pressure < 0.2 consistently → reduce workers to free OS resources

---

Runtime Mode

$app->runtimeMode('throughput'); // 'throughput' | 'balanced' | 'latency'

Mode	Use Case
throughput	High-volume APIs, batch endpoints
balanced	General-purpose (default)
latency	Real-time, WebSocket, SSE

---

Event Loop Backend

$app->eventLoop('ev'); // 'auto' | 'ev' | 'event' | 'native'

Install ext-ev for best performance:

pecl install ev

auto selects the best available backend automatically.

---

Checklist

• [ ] opcache.validate_timestamps=0
• [ ] opcache.jit=tracing
• [ ] ulimit -n 65535 (or LimitNOFILE in systemd)
• [ ] composer dump-oa-claude-sonnet-4.6load -o
• [ ] php nexphant cache:config
• [ ] ext-ev installed
• [ ] DB max_connections sized for all pools
• [ ] supervisor(true) enabled

Next Steps

• Deployment
• Benchmarks
• Configuration