Test in Production, Safely
Canary deployment rulează versiunea nouă pentru 1% din trafic. Dacă metricile sunt ok, crește treptat. Dacă nu, rollback instant.
Rollout Progress
Version 2.5.0 Rollout
Stage 2 of 5Canary Metrics Comparison
✓ All metrics within acceptable range - proceed to next stage
Rollout Configuration
# canary-config.yaml
deployment:
name: voice-ai-v2.5.0
type: canary
stages:
- percentage: 1
duration: 2h
metrics_gate:
error_rate: < 3%
latency_p95: < 800ms
- percentage: 5
duration: 4h
metrics_gate:
error_rate: < 2.5%
latency_p95: < 750ms
- percentage: 25
duration: 8h
metrics_gate:
error_rate: < 2%
latency_p95: < 700ms
- percentage: 50
duration: 12h
requires_approval: true
- percentage: 100
requires_approval: true
rollback:
automatic: true
triggers:
- error_rate > 5%
- latency_p99 > 2000ms
- manual_triggerRollback Scenario
Canary error rate jumped to 6.2% (threshold: 5%)
Traffic redirected back to v2.4.0 within 30 seconds
Error rate back to baseline (2.1%). Incident logged for investigation.
Total impact: 5% of traffic × 4 minutes = ~200 affected calls
Traffic Routing
Random Distribution
5% of all calls go to canary randomly
Sticky Sessions
Same caller always gets same version