ChatDiagram
software-delivery · devops

Canary deployment pipeline

A canary deployment reduces the initial exposure of a release by sending only a small share of production traffic to it. This diagram starts with pre-release tests, then deploys the candidate to five percent of traffic. The team watches explicit guardrails instead of treating deployment as the final step. Passing guardrails permits an expansion to twenty-five percent and a second observation period. Failing guardrails trigger removal of the canary and restoration of the prior version. A hold path gives the team time to investigate when the expanded rollout is not ready to continue. Replace the traffic percentages and measurements with the limits that match the service and its release policy.

Type OrganigrammeNorme Sugiyama layered DAG + orthogonal routingMoteur schematex-flowchartMis à jour 24/09/2026
Canary deployment pipeline
Drawing preview
La demande

“Create a canary deployment pipeline: a release candidate passes pre-release tests, deploys to 5% of traffic, and is monitored for errors, latency, and conversion. If guardrails fail, remove the canary, restore the prior version, and review the issue. If they pass, expand to 25%, monitor again, then either hold and investigate or deploy to all traffic and complete the rollout.”

Essayez ensuiteAdd a 50% traffic stage before the full rollout.Add a customer-support review after expanded monitoring.
Ce que contient ce dessin

Comprenez les décisions qui le sous-tendent.

01Initial exposure

Limit the first deployment to a small traffic share.

02Guardrails

Define which error, latency, and conversion changes are acceptable.

03Expansion

Increase traffic only after the first observation period passes.

04Recovery

Remove a failing canary and restore the prior version.

Use this for releases that can be routed to a controlled share of production traffic.

Voir tous les modèles de organigramme →

FAQ

Questions fréquentes

What is a canary deployment?

It is a release approach that sends a new version to a limited share of production traffic before making it available to everyone.

What are deployment guardrails?

They are measurable limits, such as error rate or latency, that determine whether the rollout can continue.

Why monitor after expansion?

A version can behave differently under more traffic, so a second observation period checks the release before full rollout.