Part 5 of 8 · 1 min read
Dashboards
One audience, one purpose per dashboard.
Metrics also feed real-time dashboards. UDS warns that dashboards easily become a dumping ground for forgotten charts. Good ones do not happen by chance. The first decision is the audience: ask who will look and what they want, then work backward to the charts. UDS gives three examples of kinds, and says they are its own way of organising, not a standard.
| Dashboard | Audience | What it shows |
|---|---|---|
| SLO summary | Many people across the company | System health as SLOs; in an incident, how many users are hurt |
| Public API | Operators finding bad paths | Per endpoint: requests, sizes, auth problems, handling time, dependency health, counts per response type |
| Service | The team that owns the service | Internal details, plus upstream things (load balancers, queues) and downstream things (data stores) |
The service dashboard is a first stop when debugging. From there the operator splits metrics into smaller groups and finally reads raw logs and traces, which are covered in module 3.
Dashboard
A dashboard is a screen of charts built for one audience and one purpose.
Example: a wall display of SLO health for managers, and a separate page of queue depths for the team that owns the queue.
Check yourself: 1. What is the first decision when building a dashboard? 2. Which dashboard would a manager open during an outage? 3. Where does a debugger go after the service dashboard?