Dashboards you can read in thirty seconds

Open the dashboard list of any organization that has been using Grafana for three years. You'll find between sixty and three hundred. Now look at the view statistics: three or four are opened every week, a dozen have been opened once since they were created, and the rest have never been reopened by anyone, including the people who built them. This isn't a team discipline problem. It's the predictable consequence of how the dashboards get built.

The mechanism is almost always the same. During an incident, someone looks for a piece of information, finds it, and creates a panel so they never have to look for it again. They add six more "while they're at it," just in case. Six months later, nobody remembers which question the dashboard was supposed to answer, the incident that motivated it won't happen again in the same form, and twenty-two panels on a single page are impossible to scan under stress. A useful dashboard isn't distinguished by how much information it presents, but by how quickly it lets you rule out hypotheses. This lesson describes how to build such dashboards, and how to stop producing the other kind.