Lagoon php-8.x-cli images currently do not shut down gracefully in Kubernetes. Instead on deletion they ignore the graceful stop signal and are forcefully killed after 30 seconds. This means that it takes far longer than necessary to delete a cli pod, which has several impacts such as increasing Lagoon build rollout time.

The graceful shutdown signal is ignored because the Lagoon php-8.x-cli images are built on top of the upstream php-fpm image (via Lagoon's php-8.x-fpm). PHP's FPM uses SIGQUIT as a special signal to mean "graceful shutdown", so the upstream php-fpm image is configured with STOPSIGNAL SIGQUIT. Container runtimes generally respect the STOPSIGNAL metadata, so Kubernetes uses SIGQUIT too. Since we build the Lagoon php-8.x-cli on Lagoon's php-8.x-fpm, which is built on upstream php-fpm, the Lagoon php-8.x-cli image inherits STOPSIGNAL SIGQUIT. Unfortunately the docker-sleep script used in the CLI pod ignores the non-standard SIGQUIT.

You can see the scenario described above in the image metadata. Note that upstream php-cli doesn't specify a STOPSIGNAL, so falls back to the default SIGTERM:

From what I can see there are a couple of options to fix this:

1. Stop building Lagoon php-8.x-cli images on upstream php-fpm, and instead build them on upstream php-cli which uses the default STOPSIGNAL SIGTERM.
2. Add STOPSIGNAL SIGTERM to Lagoon's php-8.x-cli Dockerfiles.

Here's the ~98% pod termination speedup observed when adding STOPSIGNAL SIGTERM to a Lagoon php-8.x-cli image (option 2):