Planning a maintenance window honestly

Starting point

Maintenance by the host had been announced for the night before 27 October 2026.

How I prepared

Before the maintenance I recorded the state of the server. For the day of the maintenance I had decided not to publish anything and not to start the next planned change. In the window I changed nothing, to the best of my knowledge; I checked containers, the Git history and logins with a terminal.

What happened

The kernel had already been installed automatically on 14 October. But it only ran after the restart that night. Who triggered the restart is not proven. What is derived is that it was the host.

How long the pages were gone

About 20 minutes. The figure comes from the server’s journal and the start times of the containers, and is derived from them. I did not measure the pages from the outside.

What I checked afterwards

Containers, pages, certificates and backups. The check was completed on 28 October 2026.

Limits

The downtime was not measured from the outside. Who triggered the restart is not proven. Whether the counter lost a click around the restart cannot be proven. I claim nothing more.