[Bug]: Nginx 502 error

[JacobDelcroix]

Platform

macOS

Operating system version

macOS

System architecture

ARM64 (M1, M2, etc)

Herd Version

1.25.0

PHP Version

No response

Bug description

---

Bug: PHP-FPM silently crashes on certain requests while PHP CLI works fine

Environment

• Herd version: 1.25.0
• macOS: Darwin 25.2.0 (Apple Silicon)
• PHP: 8.5.2 (also tested with 8.4, same issue)
• Framework: Laravel 12 + Filament v5

Description

PHP-FPM crashes silently on certain HTTP requests (e.g., /admin/teams in Filament panel), while the same requests work
perfectly with:

• php artisan serve
• curl (sometimes)
• PHP CLI / Tinker

The crash is so brutal that even register_shutdown_function() is not called, suggesting a segfault or the process being
killed.

Nginx error log

upstream prematurely closed connection while reading response header from upstream,
client: 127.0.0.1, server: basket-assistant.test, request: "GET /admin/teams HTTP/2.0",
upstream: "fastcgi://unix:/Users/xxx/Library/Application Support/Herd/herd.sock:"

Debug findings

I added detailed logging in public/index.php and a custom middleware. The request flow stops abruptly:

[16:55:45] Step 4b: Kernel resolved (mem:10MB)
[16:55:45] MW: path=admin/teams ua=Mozilla/5.0... (mem:18MB)
[16:55:45] MW: cookies=XSRF-TOKEN,coach-basket-session (mem:18MB)
[16:55:45] MW: calling next... (mem:18MB)

<-- No more logs, process killed before shutdown handler

What I tried (none worked)

• ✅ Disabling OPcache via .user.ini: opcache.enable = 0
• ✅ Suppressing E_DEPRECATED: error_reporting = E_ALL & ~E_DEPRECATED
• ✅ Switching PHP version (8.4, 8.5)
• ✅ Restarting PHP/Nginx: herd restart php && herd restart nginx
• ✅ Clearing all caches: php artisan optimize:clear

What works

• php artisan serve → ✅ Works perfectly
• php -S 127.0.0.1:8888 -t public → ✅ Works perfectly
• Docker with same codebase → ✅ Works perfectly

Reproduction

This happens on multiple Laravel projects, not just one. The crash occurs specifically when:

1. User is authenticated (has session cookies)
2. Request goes through Herd's PHP-FPM
3. Request involves complex page rendering (Filament resources, not simple routes)

Simple pages like /admin (dashboard) work fine, but resource listing pages crash.

Suspected cause

The issue might be related to:

• Herd's custom PHP extension (herd-85-arm64.so)
• PHP-FPM worker configuration
• Interaction between HTTP/2, sessions, and Filament's rendering

Steps to reproduce

No response

Relevant log output