What should the first screen of a Proxmox server tell you?

Open the overview of a Proxmox server and you get counts. So many machines, so much memory used, so much storage consumed. All of it is true, and after the third week you stop reading it, because counts do not change in a way that means anything.

A first screen earns its place by answering a different question: is there anything here that needs me today? That question has a yes or no answer, and a screen that cannot produce one is decoration.

The difference between a state and a change

Ninety percent memory usage is a state. It might be perfectly normal on a host that has been at ninety percent for a year, and it might be an emergency on a host that sat at forty percent yesterday.

The number is identical in both cases. What differs is the change, and the change is what a human should be shown. A screen that reports only states forces you to remember yesterday, and you will not remember yesterday.

This is why "everything green" is a weaker statement than it sounds. Green usually means "no threshold was crossed". It rarely means "nothing moved".

Four questions worth putting on the first screen

Did anything restart that I did not restart? An unplanned reboot is the single most informative event on a server, and it is often invisible after the fact because everything comes back up and looks normal.

Is a kernel installed but not running? After a kernel update the machine keeps running the old kernel until it reboots. The package list says up to date. The running system says otherwise. This gap can last months and nobody notices, because nothing in the interface contradicts itself.

When was the newest backup taken? Not whether backup jobs exist, and not whether the last run reported success. The age of the newest usable backup, in days. That number is the one that matters when something goes wrong, and it is the one that quietly grows when a job has stopped.

Which machines that should be running are not? A machine that was stopped on purpose and a machine that stopped by itself look identical in a list. Only the first screen can hold the intent.

What does not belong there

Anything you cannot act on. A disk temperature of 41 degrees is not information, it is anxiety. If reading it never changes what you do, it costs attention and returns nothing.

Anything that is always the same colour. A check that has been green for six months has trained you to skip it. It is not protecting you any more, it is only taking up the space where something useful could be.

Every metric that exists. The value of an overview is what it leaves out. A screen with forty numbers is a screen with no priority, and a screen with no priority is read the way a wall is read.

The hardest case: not knowing is not the same as fine

If a check cannot read its data and goes silent, the screen stays green and you read that as health. In reality nothing is known.

This is the most dangerous failure an overview can have, because it fails in the direction of comfort. The right behaviour is not to say "nothing wrong" but to say "I could not read this". A watchman who tells you what he could not see is more useful than one who says nothing at all.

Why a picture beats a list here

There is one thing a list cannot show and a map can: what is connected to what. A list can tell you a disk is nearly full. It cannot tell you that eight machines are downstream of it.

That relationship is exactly what turns a number into a decision. Ninety percent on a disk holding one test machine is a note. Ninety percent on a disk holding everything is tonight's work. Same number, different picture.

What Atlas does

The Atlas overview is built around the four questions above rather than around a metric list. Unplanned reboots, a kernel installed but not booted, the age of the newest backup, and critical machines that are stopped all sit on the first screen, because each one has a yes or no answer and each one can be acted on.

Reachability is drawn rather than listed: the screen shows the server, the machines and the resources beneath them as a connected picture, so a disk filling up carries its consequences with it instead of arriving as an isolated number.

When a check cannot read its source, the screen says so instead of staying quiet. Silence is read as health by every person who has ever looked at a dashboard, and an overview that knows this is worth more than one with more numbers on it.