Fewer servers, same Drupal: what worker mode needs

Room Drei-Hügel Time Slot Fri 4:00pm to 4:45pm (11/27/26)

Drupal rebuilds everything on every request: kernel, container, configuration. That costs CPU and RAM, and whoever pays for the infrastructure pays it on every page view. Pages that wait on external services, such as AI APIs, hold a whole PHP process while they wait.

Persistent workers such as FrankenPHP boot once and then serve many requests. Since Drupal 11.4 boots through symfony/runtime, the way there has become shorter. Two things still stand in the way:

  • state that carries over from one request into the next

  • fibers that nobody owns and that outlive their caller

The session shows:

  • where the infrastructure costs of a Drupal site arise, and which of them worker mode lowers

  • what a site needs to run in a worker, and how to tell whether it is ready

  • why the way back to classic operation must always stay open

At the end we look for two or three sites and agencies for a test phase. Payment comes only from the measured saving, for a limited time; whatever is saved after that belongs to the operator. And we want to hear what holds you back.

Learning objectives

  • understand what worker mode saves and which risks it brings: fiber and state leaks

  • whether your site or a client site is a candidate for worker mode

  • trace the infrastructure costs of a Drupal site to their causes: bootstrap per request, waiting processes, database load

Speaker
Merlin A. Rutz