Kubernetes CronJobs - Does Anyone Actually Use This [in Production]?
Talkat KubeCon + CloudNativeCon North America 2020 Virtual · on Thursday, 2020-11-19· at 15:45· Intrado Virtual Event Platform,·
summary
Lyft runs 500+ CronJobs and discovers Kubernetes’ controller does a full sync loop every 30s that lists all CronJobs, causing API client rate‑limiting when hundreds exist and making many schedules miss their startingDeadlineSeconds. Pods often disappear before debugging, and Kubernetes’ “too many failed starts” stops retries. Platform fixes include tracepoint metrics for start delay and exit codes, a web “run cron” ad‑hoc button, and a forked Kubernetes patch to avoid permanent disablement.
AI-written summary · about
chapters (8)
- 0:00Start
- 1:04Motivation
- 1:47Why should you care?
- 14:01What we did
- 19:21What should the User Experience be?
- 19:59What should the Operator Experience be?
- 21:48To Conclude...
- 22:30Kubernetes is no silver bullet
related talks
31mThe Long, Winding and Bumpy Road to CronJob’s GA
35m10 Weird Ways to Blow Up Your Kubernetes
27mRunning Kubernetes in Production A Million Ways to Crash Your Cluster
25mScale your Batch / Big Data / AI Workloads Beyond the Kubernetes Scheduler
14mKeynote: Kubernetes as a Foundational Layer of Infrastructure
05mReading 100+ Kubernetes Post-Mortems
23mDon't Let Users Find Your Outages: Synthetic Monitoring for Kubernetes Platforms
24mKubernetes Batch Processing at Scale - A Scheduling Perspective