kubelet 最佳配置

这两天试着重新去理解kubelet的每一个配置，并整理出我们最有可能要配置的项作为我们上生产的要考虑的kubelet最佳配置，如下表所示。（有些flag的value待确定）

author: garnett.wang@gmail.com

kubernetes version: 1.6.2

Kubelet Configurations We Should Care About

flagvalue–address0.0.0.0–allow-privilegedfalse–cadvisor-port int324194–cgroup-driver stringcgroupfs–cluster-dns stringSlice10.0.0.10 //todo–cluster-domain stringcaas.vivo.com–cni-bin-dir string/opt/cni/bin–cni-conf-dir string/etc/cni/net.d–docker-endpoint stringunix:///var/run/docker.sock–eviction-hard stringmemory.available<4Gi, nodefs.available<20Gi, imagefs.available<5Gi–eviction-max-pod-grace-period int3230–eviction-minimum-reclaim stringmemory.available=500Mi, nodefs.available=2Gi,, imagefs.available=2Gi–eviction-pressure-transition-period duration5m0s–eviction-soft stringmemory.available<8Gi, nodefs.available<100Gi, imagefs.available<20Gi–eviction-soft-grace-period stringmemory.available=30s, nodefs.available=2m, imagefs.available=2m–experimental-fail-swap-on+–experimental-kernel-memcg-notification+–feature-gates stringAllAlpha=false–file-check-frequency duration20s–hairpin-mode stringpromiscuous-bridge–healthz-port int3210248–image-gc-high-threshold int3260–image-gc-low-threshold int3240–image-pull-progress-deadline duration2m0s–kube-api-qps int325–kube-reserved mapStringStringcpu=200m,memory=16G–kubeconfig string/var/lib/kubelet/kubeconfig–max-pods int3250–minimum-image-ttl-duration duration1h–network-plugin stringcni–pod-infra-container-image stringvivo.registry.com/google_containers/pause-amd64:3.0–pod-manifest-path string/var/lib/kubelet/pod_manifest–port int3210250–protect-kernel-defaults+–read-only-port int3210255–require-kubeconfig+–root-dir string/var/lib/kubelet–runtime-request-timeout duration2m0s–serialize-image-pullsfalse–sync-frequency duration1m0s–system-reserved mapStringStringcpu=100m,memory=32G–volume-plugin-dir string/usr/libexec/kubernetes/kubelet-plugins/volume/exec/–volume-stats-agg-period duration1m0s

Kubelet Configuration Best Practicies

下面是我最终梳理的，认为需要真正显示设置的flag，所谓的最佳配置（当然真实的value需要根据实际情况做调整）：

/usr/bin/kubelet —address=0.0.0.0 --port=10250 --allow-privileged=false --cluster-dns=10.0.0.1 --cluster-domain=caas.vivo.com --max-pods=50 --network-plugin=cni --require-kubeconfig --pod-manifest-path=/etc/kubelet.d/ --pod-infra-container-image=vivo.registry.com/google_containers/pause-amd64:3.0 --eviction-hard=memory.available<4Gi,nodefs.available<20Gi,imagefs.available<5Gi --eviction-max-pod-grace-period=30 --eviction-minimum-reclaim=memory.available=500Mi,nodefs.available=2Gi,imagefs.available=2Gi --eviction-pressure-transition-period=5m0s --eviction-soft=memory.available<8Gi,nodefs.available<100Gi,imagefs.available<20Gi --eviction-soft-grace-period=memory.available=30s,nodefs.available=2m,imagefs.available=2m --experimental-kernel-memcg-notification --experimental-fail-swap-on --system-reserved=cpu=100m,memory=8G --kube-reserved=cpu=200m,memory=16G --hairpin-mode=promiscuous-bridge --image-gc-high-threshold=60 --image-gc-low-threshold=40 --serialize-image-pulls=false --protect-kernel-defaults --feature-gates=AllAlpha=false