Technical

Design your CI/CD pipeline with every failure planned for

Give each pipeline stage its own branch, plan what happens when it fails, and set the gates between environments.

Explore a reference map

A reference map of deployment at the scale of a product like Uber, zoomed out to show the whole system. Zoom in on any layer for the detail.

Open the full map

How it works

Describe your services and environments, and the model maps a pipeline. Each stage gets its own branch, and every question under it keeps your setup in context.

  1. Describe the system. List your services, their dependencies, your environments and how you deploy today. Ask where the bottlenecks, manual steps and missing checks are.

  2. Give each stage a branch. Build, unit tests, integration tests, security scan, staging deploy, smoke tests, production deploy and canary analysis each get one. Under each, settle parallelization, caching, artifact storage and timeouts.

  3. Plan each failure. Under every stage, ask what happens when it fails. A failed unit test blocks the merge, and a failed canary needs automatic rollback, alert routing and traffic draining.

  4. Set the promotion gates. Between environments, list what must be true to move on, such as passing tests, a clean security scan and benchmarks within threshold. Mark which gates are automated and which need a person.

Why a map instead of pipeline YAML

A 500-line pipeline file does not show that a staging deploy waits on three parallel test suites and a security scan. On a map, each of those dependencies is a branch you can read.

For a monorepo, add a branch for a change to a shared library and ask which services, test suites and deploy targets it touches.

Example

Say you are setting up Kubernetes deployments for a service that processes payments. The map splits into container build, config management, deployment strategy, observability and disaster recovery. Under deployment strategy you compare rolling updates, blue-green and canary, and under canary you plan traffic splitting, success metrics and rollback triggers. A branch on multi-region failover shows your Helm charts hard-code one cluster’s service mesh config, so you add region-specific values under config management.

Now try it yourself

Design a CI/CD pipeline for a monorepo with three microservices, a shared library, and a React frontend
The pipeline needs change detection to avoid rebuilding unchanged services. Use a DAG-based pipeline where the shared library triggers downstream builds for any service that depends on it. Parallelize service builds but gate deployments behind integration tests that run against all affected services together.
How do we handle the shared library triggering too many rebuilds?
Version the shared library with semantic versioning and only trigger downstream builds on minor or major bumps, not patches. Use lockfile-based change detection — if a service's lockfile hasn't changed, skip its build even if the library updated. For critical shared changes, add a CI label that forces a full rebuild across all consumers.

Ready to try devops workflows?