Stop 12 · Monitoring & ObservabilityTopics · 5/8

Dashboards for Cloud Infrastructure

In short

During an incident, nobody wants to write queries from scratch. A good dashboard answers "is it us, and where?" in ten seconds flat.

Building a focused, recurring view of key infrastructure health metrics — CPU, memory, error rates, latency — that a team can check at a glance without digging through raw logs.

The blob character in a commentary box looks at screens of colourful charts.

Think of it like this

The commentary box screens

Commentators don't read the full scorebook. Their screens show run rate, wickets, required rate and the wagon wheel, chosen to tell the story at a glance. A cloud dashboard picks the few numbers that tell your system's story.

Watch it happen

Laying out one service's dashboard

Requests/s

Error %

p95 latency

Saturation

DB conns

Queue depth

Deploys

1/5Top row first. These four are the "golden signals": traffic, errors, latency, saturation.

The big ideas

What you need to know

One per service

Each service gets a dashboard its team owns, plus one overview for the whole system.

Percentiles, not averages

Averages hide slow users. p95 and p99 latency show what the unlucky ones feel.

Same time window

All panels share one time range so you can line up a spike in errors with a spike in DB load.

Dashboards as code

Define them in Terraform or JSON so they're reviewed, versioned and rebuilt easily.

See it in code

A tiny example

text

1fields @timestamp, durationMs2| filter path = "/api/orders"3| stats pct(durationMs, 95) as p95, count(*) as requests by bin(5m)

A CloudWatch Logs Insights query you can add as a dashboard widget: p95 latency and request count every 5 minutes.

Save yourself the pain

Common mistakes

Oops

A dashboard with 40 panels that nobody can read during an outage.

Do this instead

Keep the main view to under ten panels. Move details to linked drill-down dashboards.

Oops

Only showing average latency.

Do this instead

Show p95 or p99 as well. A fine average can hide the 5% of users waiting seconds.

Your mission

Build a golden-signals board

~35 min

Quick check

Why show p95 latency instead of just the average?

Go deeper

1Build a simple dashboard covering 5-6 key metrics for a resource you've deployed.
2Limit a first-draft dashboard to the most important signals before adding more.
3Practice designing a dashboard someone else could understand without you explaining it.