Storage on Kubernetes - Learning From Failures
Talkat KubeCon + CloudNativeCon North America 2019 · on Wednesday, 2019-11-20· at 15:20· Ballroom Sec 20AB - San Diego Convention Center, Upper Level·
summary
Several real failures and fixes are described with concrete fixes and remaining gaps: restoring PV/PVC YAMLs without sanitizing UIDs triggers reclaimPolicy=Delete and deletes backend data; finalizers plus storage in-use protection prevent deleting PVCs while pods run; kubelet garbage-collector once wiped mounted volumes so reconstruction of /var/lib/kubelet state and disruptive tests are added; access-mode enforcement lands in 1.7/1.8 but needs attachable drivers and dummy attach implementations; AWS fixes include an LRU device-name cache of three and an uncertain attach state; FSGroup chown and node-shutdown detach remain unsolved.
AI-written summary · about
chapters (11)
- 0:00Intro
- 4:33Persistent wouldnt claim while running
- 6:52Cubelet lost data
- 12:17Kubernetes vulnerability
- 13:53File system corruption
- 20:09AWS attached issues
- 23:28FS Group
- 24:57Shutdown
- 27:31Issues
- 28:15Summary
- 29:30Questions
related talks
36mTips and tricks for Kubernetes storage
37mKubernetes Storage 101
05mProject Lightning Talk: What's New in Kubernetes Storage
49mDevOps: Kubernetes Storage - Myths, Facts and Tips
28mKubernetes Storage Evolution: Enabling High Performance Distributed Datastores
01h 16mIntro + Deep Dive: Kubernetes Storage SIG
27mKubernetes SIG Storage: Intro & Deep Dive
35mKubernetes Storage Lingo 101