Ready? A Deep Dive into Pod Readiness Gates for Service Health Management
summary
Kubernetes adds a pod readiness gate API that requires an extra pod condition to be true before a pod counts as ready, implemented so kubelet patches only its owned status fields and kubectl keeps the original Ready column; it reaches GA in 1.14. Google uses it to avoid traffic loss when programming external load balancers by targeting pods via network endpoint groups because external LB programming can take well over the typical <5s endpoint update, and Intuit’s Foremast auto-rollback demo shows the readiness gate preventing bad canaries and enabling automated remediation.
AI-written summary · about
chapters (16)
- 0:00Intro
- 0:37Container Status
- 1:28Pod Status
- 1:51Pod LifeCycle
- 2:53Pod Readiness Consumer: Workload
- 3:20Deployment Rolling Update
- 4:49Pod Readiness Consumer Service
- 7:29Rendezvous
- 8:18Workloads vs. Network Abstractions
- 11:01Pod Ready++?
- 12:18Constraints
- 14:58Pod Life Cycle with Readiness Gate
- 15:39Pod Readiness Gate
- 16:29Seamless UX
- 19:25Container Native Load Balancing
- 29:24Foremast Demo
related talks
39mKubernetes is awesome – but …
35mWhat Does “Production Ready” Really Mean for a Kubernetes Cluster?
32mHow to make kubernetes rhyme with prod readiness
29mHow to make Kubernetes rhyme with prod readiness
33mMaking an Internal Kubernetes Offering Generally Available
16mKubernetes SIG Apps Updates
30mManaged Kubernetes Service Day Zero Survival Pack
34mKeynote: Kubernetes1.6 and the Open Source Roadmap