Seems in the k8s world there are sufficient enough race conditions in shutting down pods and removing those from endpoint slices in time. Thus people started to do all kind of workarounds like adding a statically linked sleep binary to otherwise "distroless" and rather empty OCI images to just run a sleep command on shutdown before really shutting down. Or even base64 encoding the sleep binary and shipping it via configMap. Or whatever else. Eventually the situation was so severe that upstream decided to implement a sleep feature in the deployment resource directly.

In short it looks like this:

apiVersion: apps/v1
kind: Deployment
metadata:
  name: foo
spec:
  template:
    spec:
      lifecycle:
        preStop:
          sleep:
            seconds: 10

Maybe highlighting that "feature" helps some more people to get rid of their own preStop sleep commands and make some deployments a tiny bit simpler.