Evolution of Spinnaker Internals & Operations
summary
Netflix migrates Spinnaker services from Redis to relational stores (MySQL-backed persistence for Orca, Clouddriver, Front50, Echo) to cut operational complexity and improve efficiency, reducing pages from ~300 in a crisis period to under 50 in six months. Traffic: ~40,000 pipelines/day, ~10,000 EC2 and ~6,000 container deployments, peaks of 5–6k concurrent tasks. Built saga-style idempotent retries that avoided failures in a few hundred pipelines, disabled costly force-cache-refreshes to speed stages, added PF4J plugin support and active-active multi-region experiments with CockroachDB for durable front50 storage.
AI-written summary · about
chapters (21)
- 0:00Spinnaker Summit
- 2:08Three Takeaways
- 7:21People
- 8:41Deployments Per Day!
- 9:09Number of Accounts
- 10:05More Than Just Cloud Deployments!
- 14:30Performant
- 18:14The Future is - A lot of relational data SQL
- 20:15SQL, Again
- 21:57Force Cache Refresh
- 24:49Idempotency
- 30:11Extensibility
- 32:36Multi-Region
- 34:42Must Haves Monitoring & dashboards Spinnaker for · Alerting - Basic app & system telemetry
- 35:42Easy Wins
- 36:37Spinnaker as a Platform
- 37:17Simplify the Developer Experience
- 37:49Managed Delivery
- 38:07Spinnaker at Netflix in 2020
- 38:54Incentives and Motivators
- 40:24Ways to Keep Special interest Group In Touch! - Technical Oversight Committee Open Forum
related talks
36mSpinnaker - Land of a 1000 Builds
25mEnabling Diverse Application Use Cases using Spinnaker
33mPanel Discussion: Spinnaker Success - CD as a Service using Spinnaker
27mSpinnaker at Salesforce: Enterprise-Level Multi-Substrate Continuous Delivery and Release Management
26mSpinnaker Summit
28mSpinnaker at Airbnb, one year in
58mBuilding a Culture of Continuous Delivery
52mContinuous Deployment to the Cloud using Spinnaker