AlliBalliBaba
commented
1 week ago What session driver / database driver are you using? How many CPU cores does your server have? Are you setting APP_ENV=production? Are you logging something?
50 requests per second feels kinda low to be honest,
on my 20 Core i9 laptop in WSL I can get ~8.000 requests per second with a single Laravel Eloquent query, local mysql and cookie session driver.
It's still interesting though that you are seeing such a difference for big queries. I wonder what the bottleneck here could be, since at that point you're mainly benchmarking your database and PHP/eloquent. I'll try and see if I can reproduce this.
What happened?
I’ve been experimenting with Laravel Octane + FrankenPHP. Initially, the benchmarks were quite promising, but after real-world usage, I discovered a few bottlenecks. Here’s my experience.
I have two very similar Dockerfiles: one is based on the php:8.3.13-apache-bookworm image, and the other on dunglas/frankenphp:php8.13.3-bookworm. Both use the same
php.iniconfiguration. The FrankenPHP container is started with:I ran benchmarks on an API for an ERP system that involves numerous database queries, some of which return large datasets.
Response on Cached Requests
The first test was on requests that use cache (using spatie/laravel-responsecache). All tests were conducted with the command:
The results were similar when testing in the browser.
PHP:
FrankenPHP:
The results were excellent, showing around a 150% improvement with FrankenPHP.
Response on Requests Returning Simple Data
For this test, I used requests that return simple data, such as enums or lookup tables. Typically, these involve only a single query.
PHP:
FrankenPHP:
Again, the results were impressive, with FrankenPHP showing about a 122% improvement.
Requests with Large Datasets
Next, I tested requests that return a dataset of 1,000 rows using Eloquent models with 15+ relations. I confirmed there were no N+1 query issues, and the database queries themselves took around a few hundred milliseconds (up to 500ms). However, the remaining time was spent in PHP runtime. After some xdebug tracing, I found that a significant amount of time was spent on model hydration.
PHP:
FrankenPHP:
In this scenario, FrankenPHP was about 20% slower.
I’ve spent many hours trying to eliminate various factors without much success. I tried adjusting
php.inisettings, aggressive OPCache and JIT conf, testing on both a local machine (MacBook + Docker) and a server environment (Ryzen + Debian + Kubernetes).Does anyone have any ideas? Has anyone experienced something similar?
Build Type
Docker (Debian Bookworm)
Worker Mode
Yes
Operating System
GNU/Linux
CPU Architecture
x86_64
PHP configuration
Relevant log output
No response