Brave new world of unified cgroup hierarchy
summary
systemd switches the compile-time default to cgroups v2 after about five years of kernel work, and Fedora 31 runs it as default; Docker still refuses to start while podman/cri-o and many PRs add partial support. v2 gives hierarchical controllers that allow safe delegation (unit property delegate=yes or list of controllers), threaded mode restores thread-level CPU control for libvirt, BPF replaces devices/firewall controllers, PSI and new controllers deliver better CPU/memory/io accounting and the Fedora switch proves surprisingly smooth.
AI-written summary · about
chapters (13)
- 0:00Start
- 1:58Linux C Groups
- 3:41History
- 13:48Freezer Controller
- 16:09C Group Tree
- 21:33Threaded Mode for C Groups
- 25:11Limits
- 25:36Protections
- 26:11Allocations
- 27:12Soft Limit
- 30:40I / O Controller
- 32:25Latency Target Protection
- 37:21New Kernel Documentation
related talks
45mLinux Kernel Control Group v2
27msystemd cgroup delegation and control processes
57mAn introduction to control groups (cgroups) version 2
35mFuture-Proof Your Applications: Avoid Breakdowns With Cgroup Chan...
27mPostgreSQL container groups, aka cgroups down the road
28mFedora CoreOS - best host for your containers
34mCgroups and Enterprise Users
38msystemd @ Meta: From upstream to production