koordinator 资源限制源码

资源限制入口pkg/koordlet/resmanager/cpu_suppress.go中

func (r *CPUSuppress) suppressBECPU() {
  ...
    if nodeSLO.Spec.ResourceUsedThresholdWithBE.CPUSuppressPolicy == slov1alpha1.CPUCfsQuotaPolicy {
        adjustByCfsQuota(suppressCPUQuantity, node)
        r.suppressPolicyStatuses[string(slov1alpha1.CPUCfsQuotaPolicy)] = policyUsing
        r.recoverCPUSetIfNeed()
    } else {
        adjustByCPUSet(suppressCPUQuantity, nodeCPUInfo)
        r.suppressPolicyStatuses[string(slov1alpha1.CPUSetPolicy)] = policyUsing
        r.recoverCFSQuotaIfNeed()
    }
  ...
}

拿cfsquota举例pkg/koordlet/resmanager/cpu_suppress.go中

func adjustByCfsQuota(cpuQuantity *resource.Quantity, node *corev1.Node) {
  ...
    if err := sysutil.CgroupFileWrite(beCgroupPath, sysutil.CPUCFSQuota, strconv.FormatInt(newBeQuota, 10)); err != nil {
        klog.Errorf("suppressBECPU: failed to write cfs_quota_us for offline pods, error: %v", err)
        return
    }
  ...
}

其实最终的操作就是写入相应的值到/sys/fs/cgroup/cpu/kubepods.slice/kubepods-besteffort.slice/cpu.cfs_quota_us这个文件中
由于
1cgroup是层级的,子cgroup会继承父cgroup
2 所有的best effort pod都在此目录下
因此可以限制best effort pod的总cpu开销