502s from one pod: PHP-FPM, CLI OPcache and Kubernetes under load
Under heavy load some Laravel pods on Kubernetes started answering 502. Two separate causes, one memory leak by design and one crashing worker, and what fixed each.
$ grep -l "#cloud" posts/
4 posts
Under heavy load some Laravel pods on Kubernetes started answering 502. Two separate causes, one memory leak by design and one crashing worker, and what fixed each.
The application knows when a request is an attack. The edge knows how to block an IP. A status code nobody else uses is the cheapest way to connect the two.
A multi-stage Dockerfile that compiles PHP 8.5 from the official tarball into a slim runtime image, and a GitHub workflow that builds it for amd64 and arm64 without recompiling PHP on every push.
A Rust WebSocket server that speaks the Pusher protocol, scales over Redis and runs happily next to php artisan dev. Why I'm switching production to it, and why Reverb still has a place.
loading…start typing to search