The fleet as one picture, even when it is large

The fleet diagram (Servers → Fleet diagram) draws what your machines do for each other: which pool they schedule from, where their backups go, which nameserver takes over from them and which gateway their mail runs through. It is not a measurement of a machine — everything you see is a relation the panel has recorded.

On three servers that is a picture you read at a glance. On two hundred it is not, and this page is about what the screen does then.

What is on the screen

Four numbers at the top: how many machines you may see, how many are draining, how many connections run between them and how many moves are in progress. Below that is the drawing: pools on the left, machines in the middle, what they use outside themselves on the right. And under the drawing the same information as a table — not a fallback for the drawing but its other half: the detail of a connection lives there, and it is also what a screen reader and a phone get.

Clicking a block takes you to that machine's placement page.

Every block is a link, and it has a name. That sounds obvious and is not: to a browser a drawing is normally one picture with no content in it. Not here — the machine name you can read inside the block is also the name the browser announces the link by, so you can tab from block to block and hear where you are. Measured in round 5: all forty-four links in the drawing are announced by their own name. The table below it is therefore not the substitute answer for anyone not using a mouse, but the place where a connection's details live.

The text inside the drawing uses the same type sizes as the rest of the panel. That is a choice rather than a coincidence: a drawing with sizes of its own falls out of step the moment you enlarge the browser.

A fleet larger than one drawing

A drawing with two hundred machines stacked in it is no longer a drawing: it becomes thousands of pixels long and nobody reads it. So the panel draws at most seventy-five of them, in name order, and says so:

This fleet is too large for one drawing: 75 of the 200 machines are in it, in name order. Pick a pool above to reach the rest.

Two things to know:

• The count at the top is your fleet's, not the drawing's. If it says 200 and you see 75, that is correct: the number counts the machines you may see, and the drawing shows part of them.
• The table below follows the same selection. Pick a pool and the drawing, the table and the numbers all follow that choice.

Reaching the rest of your fleet

Above the numbers is a list with Whole fleet and each of your pools. Pick a pool and the screen draws only what is in it. That is usually the question you have anyway: "what does eu-west look like" is a question, "the first seventy-five machines alphabetically" is not.

If a pool does fit in one drawing, the notice disappears by itself — you only see it when something is genuinely out of view.

To look at one machine rather than a group, go through Servers to that machine and open its Placement tab. It shows what that machine carries and what it depends on, with no ceiling.

Who may reach a database server

On the Placement tab of a server with the database role there is a block Who may reach this database server: the list the server keeps itself, beside what the panel knows about it. A server belongs on it when it serves websites whose databases are here (Needed for websites there). A server nobody needs any more shows No longer needed — left over from an older move or from a hand-made corectl db peer add; the panel does not remove such a row by itself. A server the websites need that is missing from the list is named in a red notice at the top, because those websites cannot reach their database.

When a website moves to another server, the panel puts the new server on the list before the switch and takes the old one off once the old server no longer serves the website — unless other accounts' websites there still keep their databases on the same server.

Why there is a ceiling, and how high

The ceiling lives in the product itself and was measured rather than guessed: at the drawing's own geometry, seventy-five machines is about six thousand pixels of column — long, but still a picture. Two hundred is seventeen thousand, and that is not.

If you want the numbers themselves — what every fleet-wide screen costs on a fleet of two hundred machines and ten thousand hosting accounts — they are in the technical documentation at docs/scale.md, with the measured figure and the ceiling around it per screen. The measurement can be run yourself:

corecp-panel bench fleet   --config /etc/corecp-panel/panel.yaml
corecp-panel bench screens --config /etc/corecp-panel/panel.yaml \
    --url https://panel1.example --admins 8 --seconds 15
corecp-panel bench fleet   --config /etc/corecp-panel/panel.yaml --remove

The first line creates two hundred machines as records — there is no hardware behind them and no machine is contacted. The last one removes them again. Do this on a test installation, not on the one your customers run on.

When something looks wrong

• The pool list is not there. It appears only when there is more than one pool; with a single pool there is nothing to choose.
• The drawing is empty after picking a pool. Then no machine you may see is in that pool. The screen says so in as many words.
• You see fewer machines than a colleague. That is not a fault: the diagram shows what your permissions let you see, and the pool counts only count those machines too.