1. Traffic Split
Direct 5% of production traffic to the new version. The rest remains on the stable version. This is usually done via a load balancer or service mesh configuration.
Tutorial
Oct 24, 2023 · By Alex Chen
Canary Deployments: The Safest Way to Ship to Production
A practical guide to incrementally rolling out changes and catching regressions before they hit your entire user base.
Concept
What is a canary deployment — and why the name?
A canary deployment is a risk mitigation technique where you release a new version of an application to a small subset of users before rolling it out to everyone. Named after the canary in a coal mine, the idea is that if the canary dies, the miners know to get out — or at least, to turn back.
Unlike a rolling update where every change affects the entire fleet immediately, a canary lets you validate behavior in the wild. If the new code introduces a memory leak, you'll see it in the canary metrics long before it crashes your stable production environment.
Workflow
Step-by-step guide
We recommend a 5% → 20% → 100% rollout strategy, validating key metrics at each step.
2. Metric Observation
Monitor latency, error rates, and custom business metrics. Launchpad aggregates these in real-time. Look for anomalies in the canary pod's logs compared to the stable baseline.
3. Promote or Abort
If metrics look good, increase traffic to 20% or 50%. If an error spike occurs, Launchpad can automatically roll back. If confidence is high, proceed to 100%.
Safety
Common failure modes and configuration
The most dangerous part of a canary is the "blind spot" where metrics might look good but the user experience is degraded. Configure these gates to catch them.
- Latency Spikes: Set a 200ms threshold on p95. If the new version is slower, pause traffic immediately.
- Error Rate Thresholds: Define an absolute limit (e.g., 0.5% 5xx errors). Launchpad's gate engine enforces this.
- Database Connection Pool Exhaustion: Monitor connection counts. A sudden drop in connections on the new pod often means it's crashing under load.
Implementation
Code sample: Canary block
Here is a minimal configuration for a canary rollout on Launchpad. We use the canary stage to define the rollout strategy.
1pipeline: canary-rollout
2on: [push]
3
4stages:
5 - deploy:stable
6 - canary:
7 traffic: 5%
8 metric: http_5xx
9 threshold: 0.5
10 pause: 60s
11 on_fail: rollback
12 on_success: promote
13 notify: slack#alerts
Strategy
Comparison: Canary vs. Blue-Green vs. Rolling
Understanding the trade-offs helps you choose the right strategy for your stack.
Canary
Gradual traffic shift. Best for gradual confidence building. Slightly complex to implement at scale.
Blue-Green
Switch traffic instantly between two identical environments. Zero downtime, but double infrastructure costs.
Rolling
Update pods one by one. Simple to understand. Higher risk of partial failure affecting a subset of users.
Related resources
Launchpad Docs
Read the full documentation on deployment strategies and orchestration.
Read docs →Changelog
See the latest features added to the platform, including our new canary engine.
View changelog →Ship with confidence
Automate canary deployments in minutes.
Launchpad handles the traffic splitting, metrics monitoring, and automatic rollbacks so you don't have to think about it during deploy time.
Ship code. Not excuses.