ta_benchmark

Benchmark comparing different Laravel server adapters under load using k6.

Adapters

Adapter	Port	Workers	Image
PHP-FPM + nginx	8082	12	php:8.4-fpm-alpine + nginx:alpine
Octane Swoole	8084	12	phpswoole/swoole:5.1-php8.3-alpine
TrueAsync FrankenPHP	8083	12	trueasync/php-true-async:latest-frankenphp

Routes tested

Route	Description
GET /hello	Pure JSON response, no DB
GET /test	One DB query (SELECT pg_sleep(0.01), simulates 10ms DB latency)
GET /bench	5 DB queries: SELECT user, SELECT posts, INSERT view, UPDATE counter, SELECT aggregate
GET /debug/connections	Live PostgreSQL connection stats (active, idle, total vs max) — TrueAsync, Swoole only

---

Setup

Requirements

• Docker + Docker Compose
• k6

Clone

git clone https://github.com/YanGusik/ta_benchmark.git
cd ta_benchmark

---

PHP-FPM

cd fpm

# 1. Build the image, then install dependencies before starting
#    (the container crashes on startup if vendor/ is missing)
docker compose build
docker compose run --rm php composer install --ignore-platform-reqs

# 2. Start services
docker compose up -d

# 3. First-time setup
docker compose exec php chmod -R 777 /app/storage /app/bootstrap/cache
docker compose exec php php artisan key:generate
docker compose exec php php artisan migrate
docker compose exec php php artisan db:seed --class=BenchmarkSeeder
curl http://localhost:8082/hello

---

Octane Swoole

cd octane_swoole

# 1. Build the image, then install dependencies before starting
docker compose build
docker compose run --rm app composer install --ignore-platform-reqs

# 2. Start services
docker compose up -d

# 3. First-time setup
docker compose exec app chmod -R 777 /app/storage /app/bootstrap/cache
docker compose exec app php artisan key:generate
docker compose exec app php artisan migrate
docker compose exec app php artisan db:seed --class=BenchmarkSeeder
curl http://localhost:8084/hello

---

TrueAsync FrankenPHP

TrueAsync uses a custom PHP extension for coroutines and requires PHP 8.6 (only available inside the Docker image — not a standard PHP release). The package yangusik/laravel-spawn is pulled from a VCS repository. Use --ignore-platform-reqs so Composer doesn't reject the PHP 8.6 requirement on your local machine.

cd trueasync

# 1. Install dependencies before starting
#    (uses pre-built image — no build step needed)
docker compose run --rm app composer install --ignore-platform-reqs

# 2. Start the server
docker compose up -d

# 3. First-time setup
docker compose exec app chmod -R 777 /app/storage /app/bootstrap/cache
docker compose exec app php artisan key:generate
docker compose exec app php artisan migrate
docker compose exec app php artisan db:seed --class=BenchmarkSeeder
docker compose exec app php artisan vendor:publish --tag=async-config

curl http://localhost:8083/hello

Updating the adapter

When yangusik/laravel-spawn is updated, pull the latest version:

cd trueasync
docker compose run --rm app composer update yangusik/laravel-spawn --ignore-platform-reqs
docker compose restart app

DB connection pool

TrueAsync uses a coroutine-aware PDO pool. The pool size is configured in config/async.php (publish with php artisan vendor:publish --tag=async-config):

'db_pool' => [
    'enabled' => true,
    'min'     => 2,
    'max'     => 10,   // per worker — 12 workers × 10 = 120 max connections total
    'healthcheck_interval' => 30,
],

PostgreSQL max_connections is set to 500 in compose.yml to accommodate the pool. Monitor live connections during load:

curl -s http://localhost:8083/debug/connections | jq .

# Or watch during k6 run
watch -n1 'curl -s http://localhost:8083/debug/connections | jq "{total,max_connections,by_state}"'

---

Running benchmarks

All scripts use the same load: 840 hello req/s + 360 DB req/s = 1200 req/s total.

# PHP-FPM
k6 run k6/fpm.js

# Octane Swoole
k6 run k6/octane_swoole.js

# TrueAsync
k6 run k6/trueasync.js

# /bench endpoint (5 DB queries per request), target adapter via BASE_URL
BASE_URL=http://localhost:8083 k6 run k6/bench.js
BASE_URL=http://localhost:8084 k6 run k6/bench.js
BASE_URL=http://localhost:8082 k6 run k6/bench.js

---

Results

Load: 840 req/s /hello + 360 req/s /test = 1200 req/s total, constant-arrival-rate, 30s duration. All adapters: 12 workers. Environment: WSL2 (Linux 6.6 on Windows).

Metric	PHP-FPM (12w)	Octane Swoole (12w)	TrueAsync (12w)
Target rate	1200 req/s	1200 req/s	1200 req/s
Actual throughput	~200 req/s	~752 req/s	~1118 req/s
Dropped iterations	~28 000	~5 000	20
avg latency	~4 000ms	~880ms	13ms
p(95) latency	~5 000ms	2 320ms	21ms
p(95) < 200ms	✗	✗	✓
Failed requests	0%	0%	0%
DB connections (peak)	—	—	120

---

Architecture comparison

	PHP-FPM	Octane Swoole	TrueAsync
Request model	Process per request	1 process = 1 request at a time	1 worker = N coroutines
DB I/O	Blocking (new conn each req)	Blocking (PDO synchronous)	Non-blocking (coroutine yield)
Memory model	Stateless	Long-lived process	Long-lived process + coroutine context isolation
App bootstrap	Every request	Once per worker	Once per worker

Why TrueAsync wins on DB-bound load: Swoole keeps the app in memory (avoids bootstrap cost) but PDO is still synchronous — a worker blocked on pg_sleep(0.01) cannot accept another request. TrueAsync yields the coroutine on every DB call, so one worker handles hundreds of concurrent DB-bound requests without blocking.

---

Notes

• Each adapter has its own PostgreSQL instance on a separate port to avoid interference
• APP_DEBUG=false in all setups for fair comparison
• OPcache enabled in PHP-FPM
• PostgreSQL max_connections=500 in all setups
• Absolute numbers will be higher on bare metal (benchmarks run on WSL2)