Container Role Configuration to Optimize Resource Utilization and Reduce Memory Usage

[okaufmann]

Hi guys,

I am utilizing the serversideup/php image to orchestrate a Laravel application across three distinct containers:

• Webapp Container: Hosts nginx and PHP-FPM, exposing a designated port for application accessibility.
• Queue Container: Dedicated to running Horizon.
• Scheduler Container: Executes the php artisan schedule:work command.

This architecture facilitates individual scaling of each component on Kubernetes.

However, I've observed that nginx and PHP-FPM are initialized in every container, which significantly increases the Memory usage, especially when managing numerous Laravel instances.

To mitigate this, I've resorted to creating two unique images:

1. A webapp image derived from serversideup/php:8.2-fpm-nginx.
2. A shared image for the queue and scheduler containers based on serversideup/php:8.2-cli.

This workaround, albeit functional, introduces a complex initial build configuration and heightens the potential for failure due to the need to manage multiple images.

In the past, I've employed a container role concept. Despite the image having the capability to host the webapp, assigning a role like queue via environment variable prevented the initialization of nginx and PHP-FPM, thereby optimizing Memory utilization.

I am inquiring if a similar container role feature could be envisaged for future releases, to streamline resource utilization, reduce Memory usage, and simplify image management.

[jaydrogers]

In our use case, we do the same thing. We use fpm-nginx multiple times (even for things like Queue and Scheduler). In our case this works fine. An app with a similar structure runs on a server running 2GB of RAM and 1 CPU.

Optimizing servers is very unique to every application. In the case if you do not want to use the method above, you will need to use an FPM image only and ensure that your server is only processing PHP requests.

This will require extra work where you'll need to ensure you have a load balancer and/or HTTP server outside of your PHP Containers and you'll need to make sure the static assets are stored in an accessible location for all containers.

Hope this helps!

[okaufmann]

I think I have expressed myself inaccurately. My problem is that for the queue the corresponding container also starts NGINX and FPM (with s3) which eats memory. But I want to start only the queue without starting NGINX and FPM. To save memory. We run a large amount of Laravel applications and it is important that the individual applications only use as much memory as necessary.

[jaydrogers]

Check the PHP_PM_* variables here: https://serversideup.net/open-source/docker-php/docs/reference/environment-variable-specification

You might be able to really scale these numbers down. The biggest one for memory savings for me was setting PHP_PM_CONTROL to ondemand, which saved a ton of memory (which is the default for the image)

[okaufmann]

PHP_PM_CONTROL is already set to ondemand by default:

I started my base image and serversideup/php:8.2-cli with the command sleep 60 to compare the memory usage.

Seems like nginx+fpm consumes about 20Mb more than the php-cli container. Exactly this 20Mb I would like to save.

[jaydrogers]

Your other option may be overriding the entrypoints so it doesn't kick off the S6 processor supervision of both services? https://docs.docker.com/compose/compose-file/compose-file-v3/#entrypoint

[okaufmann]

I've been searching so much for configuration values and messing around with the s6 that I completely forgot about the simple stuff.... a bit embarrassing 🤦‍♂️ Thanks for the tip! That has reduced the memory consumption.