Best Practice on Windows Workload Management In Kubernetes
Talkat KubeCon + CloudNativeCon + Open Source Summit China 2021 · on Friday, 2021-12-10· at 14:05· Kubecon + CloudNativeCon 演讲厅, 虚拟会议 (Virtual)·
summary
Windows containers allocate 20 GB default scratch per container and keeping the container runtime root on C: causes OS disk exhaustion; relocate the runtime root to a larger volume (example move runtime root to E:\\ via symlink) and ensure PVs are deleted before cluster removal to avoid orphaned disks. Rolling upgrades work one node at a time and measure time (50 pods × 3 min = 150 min) so throttle concurrency (example set max in-flight to 5 (example)) and wait for all block volumes to detach before removing a node. gMSA requires AD + KDS root key and integrates via install CredentialSpec CRD + mutating/validating webhook that expands and authorizes gMSA creds per pod.
AI-written summary · about
chapters (5)
- 0:00Start
- 0:43Agenda
- 1:13Background - Windows Clusters .
- 1:51Operation tips - ephemeral disks.
- 2:58Operation tips - rolling update
related talks
38mDay 2 Operations with Windows Containers
59mMigrate and Modernize with Kubernetes and Windows Containers
01h 04mIntro + Deep Dive: Windows SIG
35mIntroduction to Windows Containers in Kubernetes
32mLessons From Scheduling 20 Million Windows Containers a Month
01h 02mAKS Survival Pack: what to consider before going all-in with AKS
35mHA and Data Mobility for persistent Volumes
32mYour Pods can be stateless, your VMs cannot! How to manage state in k8s