Cgroups V2: Before You Jump In
summary
Adobe enables Flatcar 2969.0.0 and finds cgroups v2 breaks telemetry (metrics missing, HPA stops) and causes Java/runtime issues so clusters had to be rolled back, then fixed by making kubelet and runtime aware and updating cAdvisor. Requirements and fixes: kernel 5.8+, kubernetes 1.25 (stable), cAdvisor >= 0.44 (0.55 recommended), systemd for kubelet and runtime, language/runtime patches (Go auto‑max‑procs, JDK >=11.0.16 or >=15), test separate node pools and expect memory QoS, PSI metrics and future container‑aware OOM behavior.
AI-written summary · about
chapters (10)
- 0:00Introduction
- 2:15Investigating cgroups V2
- 5:42What are cgroups
- 8:10Key changes in V2
- 13:50Kubernetes timeline for V2
- 15:57What V2 means for you
- 20:30Preparing cluster for V2
- 22:38Preparing apps for V2
- 24:22Demo: V2 impact on Go apps
- 27:09Key takeaways and closing
related talks
32mWhat is cgroups v2? and how it affects our application
08mKeynote: Kubernetes Project Updates
34mKubernetes SIG Node Intro and Deep Dive
23mPractice of Fine-grained Cgroups Resources Scheduling in Kubernetes
47mCPU & Memory, Requests & Limits, QoS & CFS: The JVM in the Kubernetes Pod
01h 59mKubernetes for Java Developers
50mJVM and Kubernetes walk into the bar...
35mSIG Cluster Lifecycle (kops)
