前提:kubelet已经关闭kmem accounting
sh build.sh 会生成可执行文件cgroup_fix
- 停止kubelet
- 打开task migration 开关(虽然kernel doc上说打开目标的task migration即可，我们都打开）
- 创建新的group /sys/fs/cgroup/memory/kubepods2
- 将/sys/fs/cgroup/memory/kubepods下的子group对应的memory.limit_in_bytes和tasks 迁移到新的/sys/fs/cgroup/memory/kubepods2
- 迁移过程中逐步删除各个子group, 直到/sys/fs/cgroup/memory/kubepods被删除，重新创建/sys/fs/cgroup/memory/kubepods
- 将/sys/fs/cgroup/memory/kubepods2的子group迁移回/sys/fs/cgroup/memory/kubepods
- 启动kubelet
- 重启cadvisor