Highly Available Services During Maintenance Events
summary
Explains how to keep services available while nodes are rebooted or upgraded by using PodDisruptionBudget to limit concurrent evictions and graceful termination with longer terminationGracePeriodSeconds so long-lived requests and flushes complete. Shows node drain behavior with kubectl drain obeying PDBs (cloud upgrades cap waits at one hour), demos a 5‑replica Cassandra statefulset plus 5 web pods surviving rolling node reboots with no dropped requests. Discusses edge cases like readiness vs quorum and local-volume decommissioning.
AI-written summary · about
chapters (11)
- 0:00Introduction
- 1:26Maintenance is fundamentally disruptive
- 2:46Options for maintenance
- 4:53Introducing Eric Tune
- 7:15Graceful Termination
- 8:49How to handle graceful termination
- 9:54Pod disruption budgets
- 11:47Deleting pods
- 13:24Draining a node
- 16:02Demo
- 21:17QA
related talks
25mWhat goes down the drain?
26mWait! Can Your Pod Survive a Restart?
41mAfterlife Tales
30mIs Your Kubernetes Disaster Recovery Actually Ready?
35mHow to Safely and Reliably (...) While Maintaining 100% Uptime
40mStrategies for Zero Down Time, Frequent Deployments
25mAutomating Kubernetes Upgrades for Zero Downtime
37mZero downtime Deployment on Kubernetes