Breaking Changes Without Breaking Things
summary
Alaska Airlines moves from rolling Kubernetes deploys to rainbow deployments on Azure Container Apps, packaging services and Temporal workers into a single release so one traffic shift replaces the previous five-step, multi-week process. They use Temporal worker versioning with build IDs and workflow pinning (versioningBehavior.pinned and pinVersioningOverride) to keep in-flight workflows on their originating release. Result: fewer than-before workflow patches (they had 50+), instant rollbacks, much faster deploys for ~3M Temporal actions/day and 32 PRs/week, at the cost of rebuilt CI/CD, routing plumbing, tagging/observability, and occasional manual workflow migrations.
AI-written summary · about
chapters (7)
- 0:00Introduction and speakers
- 1:07Problem with rolling deployments
- 10:14Five-phase deployment breakdown
- 15:50Solution overview rainbow deployments
- 22:56Challenges and trade-offs
- 27:10Team and acknowledgments
- 27:53Q&A session
related talks
01h 01mFrom one release per quarter to 30 times a day
01h 04mZero Downtime Application Deployments on Azure PaaS
40mStrategies for Zero Down Time, Frequent Deployments
36mPractical Cloud Native: What Works, What Doesn't
54mFrom Code to Customer
46mContinuous Deployment Workflows
24mExclude Human – Continuous Deployment and OpenShift
41mSurviving Continuous Deployment in Distributed Systems