In-Place PVC Re-Binding: Zero-Downtime Disk Migration Using Only the Kub...
Talkat KubeCon + CloudNativeCon + OpenInfra Summit + PyTorch Conference China 2026 · 2026-09-07..09-09·
summary
Migrates ~500 Azure managed disks to the CSI driver across many clusters in ~2 months with no severity-1 incidents by using only standard Kubernetes APIs: delete storageclass and recreate with same name; patch PV reclaimPolicy to Retain; create a new CSI PV for the same disk in Available state (omit UID/resourceVersion), then delete PVC then pod in sequence so the new PVC binds to the CSI PV; delete old PV afterward. Handles edge cases: rare Azure multi-attach needs manual force-detach; PVC protection finalizer and sequencing prevent race conditions.
AI-written summary · about
chapters (15)
- 0:00Zero-downtime PVC rebinding intro
- 1:04Legacy storage to CSI migration
- 2:26Kubernetes storage APIs
- 4:55Key migration observations
- 7:08PVC rebinding algorithm
- 9:16Migration results and alternatives
- 10:20Why we chose this approach
- 12:05Speed bumps and failures
- 14:11Migration principles
- 15:34Automation and approval
- 15:55About the company
- 16:39Q&A: Other storage migrations
- 17:33Q&A: Node affinity for failures
- 18:17Q&A: PVC race condition fix
- 19:48Q&A: True zero downtime?
related talks
48mCSI in Kubernetes
43mDeploy & manage storage volumes with Azure Container Storage
28mKubernetes Storage Evolution: Enabling High Performance Distributed Datastores
19mFrom PVC to Mount Point: Dissecting a...
24mCapacity-aware Dynamic Volume Provisioning For LVM-based Local Storage
37mKubernetes Storage 101
21mKeep Your VM Dancing With Volume Migration
37mStorage on Kubernetes - Learning From Failures