Canary Releases

Strategy

Canary release gradually routes small percentage of traffic to new version. If metrics are good, gradually increase percentage until all traffic routes to new version.

Day 1: 98% Blue, 2% Green (2 instances) Day 2: 95% Blue, 5% Green (5 instances) Day 3: 90% Blue, 10% Green (10 instances) Day 4: 50% Blue, 50% Green (rollout halfway) Day 5: 0% Blue, 100% Green (fully deployed)

Implementation

Manual Canary (Using Instance Counts)

```bash

Deploy new version

$ cf push my-app-green -i 1 # 1 instance = ~9%

Keep old version

$ cf scale my-app-blue -i 10 # 10 instances = ~91%

Monitor metrics

$ cf stats my-app-green $ cf stats my-app-blue

Gradually increase green

$ cf scale my-app-green -i 3 # ~23% traffic $ cf scale my-app-green -i 5 # ~33% traffic $ cf scale my-app-green -i 10 # ~50% traffic

Once stable, finish migration

$ cf scale my-app-blue -i 0 # No traffic to blue $ cf delete my-app-blue # Remove old ```

Traffic Splitting (Advanced)

PCF supports weighted routing:

```bash

Deploy both versions

$ cf push my-app-blue -i 1

Configure traffic weighting (if supported)

$ cf route-service example.com \ --hostname my-app \ my-app-blue:0.9 \ my-app-green:0.1 # 10% to green ```

Monitoring During Canary

Key Metrics:

```bash

Error rate

$ cf logs my-app-green | grep -i error | wc -l

Response time (from app logs)

$ cf logs my-app-green --recent | grep "duration_ms" | \ sed 's/.duration_ms":([0-9]).*/\1/' | \ awk '{sum+=$1; count++} END {print sum/count " ms avg"}'

Capacity

$ cf stats my-app-green | grep memory

CPU usage

$ cf stats my-app-blue | grep cpu ```

Advantages

Aspect	Benefit
Safety	Minimal risk with small percentage
Validation	Real production data before full rollout
Rollback	Easy to revert if issues found
Gradual	Smooth transition for users
Observation	Time to monitor behavior

Disadvantages

Aspect	Issue
Duration	Takes longer than blue-green
Complexity	More monitoring required
Resources	Run both versions during rollout
User Impact	Some users on old, some on new

Deciding Canary vs Blue-Green

Use Canary if: - Changes may have unknown impact - Want to validate with real traffic - Risk of issues justifies slower rollout

Use Blue-Green if: - Confident in changes (tested) - Need quick deployment - Can tolerate resource overhead - Want immediate rollback

---

See Lab 4: Blue-Green Deployment for hands-on practice with traffic management.