Planet-Scale Dashboards

How do you provide powerful monitoring for thousands of services without the toil? Learn how Google built a zero-configuration dashboard system that just works.

#1about 3 minutes

The challenge of creating monitoring dashboards from scratch

Monitoring is often an afterthought, leading to painful incident response without the necessary dashboards for troubleshooting.

#2about 3 minutes

Understanding Google's unique observability scaling challenges

Google's massive scale, global distribution, and monorepo architecture created a unique need for a scalable, reusable monitoring solution.

#3about 5 minutes

Building reusable dashboards with templated dimensions

Replace hardcoded values in queries with template variables, called dimensions, to create a single dashboard that can be reused for any service.

#4about 6 minutes

Solving dashboard discovery with scopes and traits

Address the problem of too many dashboards by having users select a "scope" (e.g., a service), which then uses discovered "traits" to show only relevant dashboards.

#5about 2 minutes

Modeling different entities with scope types

Introduce "scope types" to create namespaces for different kinds of monitorable entities, such as servers, databases, or machine learning models.

#6about 4 minutes

Why infrastructure as code is not the right solution

Static provisioning with infrastructure-as-code or dashboards-as-code is insufficient because it lacks dynamic runtime information and creates a stale second source of truth.

#7about 3 minutes

Improving performance at scale with query variants

Use pre-aggregated metrics and define multiple query "variants" within a graph, allowing the system to automatically select the most performant query based on the user's drill-down level.

#8about 1 minute

Visualizing dependencies with a service graph

Leverage the scope and dependency information to build a service graph that helps engineers quickly navigate between related systems during an incident.

#9about 1 minute