Linux servers for Laravel applications in production

I audited and reconfigured the production Ubuntu servers behind the team's applications: Apache with mpm_event and HTTP/2, PHP-FPM, OPcache, InnoDB sized to the RAM, queues under Supervisor, and hardening.

Ubuntu

Apache

PHP-FPM

OPcache

MySQL/MariaDB

Supervisor

Hardening

My role

Auditing and configuring production servers; moving blocking tasks to queues.

Evidence

What I did

Audited production Ubuntu servers with 8 GB of RAM.
Apache from mpm_prefork to mpm_event, with HTTP/2.
PHP split out of Apache with PHP-FPM and capped worker pools.
OPcache, InnoDB buffer pool, and swappiness tuned to the available memory.
Blocking tasks moved to Laravel queues, with workers under Supervisor.
Apache and PHP hardening, and protection of sensitive files.

Still to be measured

Latency, memory, and swap before and after the changes.

Next step

A performance baseline and monitoring.
Turn the configuration into code (Ansible or Terraform) and containers.

The idea

On an 8 GB server, Apache, PHP, the database, and the workers all share the same machine. The work was about dividing memory on purpose: every component gets a limit, so a traffic spike becomes a short wait instead of an outage.

What I changed

Apache mpm_event + PHP-FPM instead of prefork with embedded PHP: less memory per connection and an explicit cap on PHP processes. It also enabled HTTP/2.
Database sized to the machine: the InnoDB buffer pool gets what’s left after PHP, not “all the RAM”.
Low vm.swappiness: the server favors RAM over swap.
OPcache: PHP stops recompiling scripts on every request.
Queues with Supervisor: what used to make users wait moved to the background, and Supervisor keeps the workers alive.
Hardening: Apache and PHP don’t advertise their versions, there are defensive headers, and files like .env aren’t reachable.

What I don’t claim

I didn’t measure performance before the changes, so I don’t publish improvement percentages. The next step is a baseline and monitoring, and turning this configuration into reproducible infrastructure: that’s exactly the path toward DevOps I’m on.