deployment-safety

name	deployment-safety
description	Deployment best practices — canary patterns, rollback triggers, and health check interpretation.

Deployment Safety

Pre-Flight Checks

Before any deployment, verify:

1. Service health: All health checks passing (status = "healthy")
2. Error rate baseline: Current error rate < 2% (deploying into an already-degraded service is risky)
3. CPU headroom: CPU < 70% (deployment causes brief CPU spikes)
4. No active incidents: Check if the service is currently under investigation
5. Recent deploys: No other deployment to the same service in the last 10 minutes

Rollback Triggers

Automatically recommend rollback if ANY of these occur post-deploy:

Metric	Threshold	Window
Error rate increase	> 2% above baseline	2 minutes
Health check failure	Any check fails	Immediate
CPU spike	> 95% sustained	1 minute
Response time increase	> 50% above baseline	2 minutes

Deployment Monitoring Pattern

Deploy → Wait 30s → Check metrics → Wait 30s → Check metrics → Verdict

• Minimum 2 check cycles before declaring success
• If first check shows degradation, wait for second check to confirm
• If both checks show degradation, recommend rollback

Rollback Procedure

1. Call rollback_service with the deployment ID
2. Wait 30 seconds for rollback to complete
3. Verify health checks are passing again
4. Report final status to the user