这篇文章主要介绍了Kubernetes垃圾回收机制的示例分析,具有一定借鉴价值,感兴趣的朋友可以参考下,希望大家阅读完这篇文章之后大有收获,下面让小编带着大家一起了解一下。
测试环境上,node节点上面镜像,突然被删除,故查明下删除的原理机制
Kubelet的GC功能将清理未使用的image和container。Kubelet每分钟对container执行一次GC,每5分钟对image执行一次GC。不建议使用外部垃圾收集工具,因为这些工具可能破坏Kubelet。
Kubernetes与Cadvisor配合,通过ImageManager管理所有image的生命周期。 image的GC策略包含高阈值和低阈值,高于高阈值的磁盘使用率将触发GC,删除最近最少使用的图像,直到满足低阈值为止。
container的GC主要有3个用户定义变量: MinAge:容器被GC的最短时间 MaxPerPodContainer: 允许每个PodContainer中死容器的最大数目,PodContainer指1个Container而非pod MaxContainers:死容器的最大数目 Minage=0,MaxPerPodContainer和MaxContainers <0, 表示禁用这些变量 GC用于unidentified、deleted或超出边界的容器(3个用户定义变量)。 最旧的container通常首先被移除。 如果MaxPerPodContainer>MaxContainers,maxperpodcontainer会进行调整,直至降级为1,并逐出最旧的容器。 pods所拥有的已删除的容器一旦超过MinAge,就会被删除。 未由Kubelet管理的容器不受容器垃圾收集的约束。
通过修改kubelet flags来实现。 image gc参数: image-gc-high-threshold image GC 高阈值百分比,缺省为85% image-gc-low-threshold image GC 低阈值百分比,缺省为80% minimum-container-ttl-duration MinAge参数,缺省为0 maximum-dead-containers-per-container MaxPerPodContainer,缺省为1. maximum-dead-containers MaxContainers,缺省为-1, 也就是没有限制 Container可能在其到期之前被GC,Container包含日志和其他对故障排除有用的数据。强烈建议MaxPerPodContainer和maximum-dead-containers 足够大。
配置一个10s后会失败的container 配置文件pod-gc.yaml
apiVersion: v1 kind: Pod metadata: name: gc-test spec: containers: - name: busybox-gc-1 image: busybox:v1 command: - /bin/sh - -c - 'sleep 10 && hello'
使用yaml创建pod
kubectl apply -f pod-gc.yaml
查看docker的状态,当出现2个exit的container时,就会出发GC,回收最早的一个container
# docker ps -a | grep busybox-gc 8ddbfcf4ebee 59788edf1f3e "/bin/sh -c 'sleep 1…" 2 seconds ago Up 1 second k8s_busybox-gc-1_gc-test_default_3651c167-75f0-11e9-bc74-52540005f38a_5 7d7c1ade82af 59788edf1f3e "/bin/sh -c 'sleep 1…" About a minute ago Exited (127) About a minute ago k8s_busybox-gc-1_gc-test_default_3651c167-75f0-11e9-bc74-52540005f38a_4 # docker ps -a | grep busybox-gc 8ddbfcf4ebee 59788edf1f3e "/bin/sh -c 'sleep 1…" 10 seconds ago Up 9 seconds k8s_busybox-gc-1_gc-test_default_3651c167-75f0-11e9-bc74-52540005f38a_5 7d7c1ade82af 59788edf1f3e "/bin/sh -c 'sleep 1…" About a minute ago Exited (127) About a minute ago k8s_busybox-gc-1_gc-test_default_3651c167-75f0-11e9-bc74-52540005f38a_4 # docker ps -a | grep busybox-gc 8ddbfcf4ebee 59788edf1f3e "/bin/sh -c 'sleep 1…" 11 seconds ago Exited (127) Less than a second ago k8s_busybox-gc-1_gc-test_default_3651c167-75f0-11e9-bc74-52540005f38a_5 7d7c1ade82af 59788edf1f3e "/bin/sh -c 'sleep 1…" About a minute ago Exited (127) About a minute ago k8s_busybox-gc-1_gc-test_default_3651c167-75f0-11e9-bc74-52540005f38a_4 # docker ps -a | grep busybox-gc 8ddbfcf4ebee 59788edf1f3e "/bin/sh -c 'sleep 1…" 12 seconds ago Exited (127) 1 second ago k8s_busybox-gc-1_gc-test_default_3651c167-75f0-11e9-bc74-52540005f38a_5
修改配置文件/var/lib/kubelet/kubeadm-flags.env增加参数–maximum-dead-containers-per-container=2
KUBELET_KUBEADM_ARGS=--cgroup-driver=systemd --network-plugin=cni --pod-infra-container-image=k8s.gcr.io/pause:3.1 --maximum-dead-containers-per-container=2
systemctl restart kubelet ##重启kubelet服务 kubectl delete -f pod-gc.yaml ##删除前面的pod kubectl apply -f pod-gc.yaml ##重新添加pod
可以看到,当exit的容器达到3个时,才会触发gc
# docker ps -a | grep busybox-gc c8677098e9aa af2f74c517aa "/bin/sh -c 'sleep 1…" 8 seconds ago Up 8 seconds k8s_busybox-gc-1_gc-test_default_3425a55b-75f5-11e9-bc74-52540005f38a_2 f0bc1b7893bd af2f74c517aa "/bin/sh -c 'sleep 1…" 34 seconds ago Exited (127) 23 seconds ago k8s_busybox-gc-1_gc-test_default_3425a55b-75f5-11e9-bc74-52540005f38a_1 9e7fa118b1d3 af2f74c517aa "/bin/sh -c 'sleep 1…" 45 seconds ago Exited (127) 35 seconds ago k8s_busybox-gc-1_gc-test_default_3425a55b-75f5-11e9-bc74-52540005f38a_0 # docker ps -a | grep busybox-gc c8677098e9aa af2f74c517aa "/bin/sh -c 'sleep 1…" 9 seconds ago Up 9 seconds k8s_busybox-gc-1_gc-test_default_3425a55b-75f5-11e9-bc74-52540005f38a_2 f0bc1b7893bd af2f74c517aa "/bin/sh -c 'sleep 1…" 35 seconds ago Exited (127) 24 seconds ago k8s_busybox-gc-1_gc-test_default_3425a55b-75f5-11e9-bc74-52540005f38a_1 9e7fa118b1d3 af2f74c517aa "/bin/sh -c 'sleep 1…" 46 seconds ago Exited (127) 36 seconds ago k8s_busybox-gc-1_gc-test_default_3425a55b-75f5-11e9-bc74-52540005f38a_0 # docker ps -a | grep busybox-gc c8677098e9aa af2f74c517aa "/bin/sh -c 'sleep 1…" 10 seconds ago Exited (127) Less than a second ago k8s_busybox-gc-1_gc-test_default_3425a55b-75f5-11e9-bc74-52540005f38a_2 f0bc1b7893bd af2f74c517aa "/bin/sh -c 'sleep 1…" 36 seconds ago Exited (127) 25 seconds ago k8s_busybox-gc-1_gc-test_default_3425a55b-75f5-11e9-bc74-52540005f38a_1 9e7fa118b1d3 af2f74c517aa "/bin/sh -c 'sleep 1…" 47 seconds ago Exited (127) 36 seconds ago k8s_busybox-gc-1_gc-test_default_3425a55b-75f5-11e9-bc74-52540005f38a_0 # docker ps -a | grep busybox-gc c8677098e9aa af2f74c517aa "/bin/sh -c 'sleep 1…" 11 seconds ago Exited (127) 1 second ago k8s_busybox-gc-1_gc-test_default_3425a55b-75f5-11e9-bc74-52540005f38a_2 f0bc1b7893bd af2f74c517aa "/bin/sh -c 'sleep 1…" 37 seconds ago Exited (127) 26 seconds ago k8s_busybox-gc-1_gc-test_default_3425a55b-75f5-11e9-bc74-52540005f38a_1
感谢你能够认真阅读完这篇文章,希望小编分享的“Kubernetes垃圾回收机制的示例分析”这篇文章对大家有帮助,同时也希望大家多多支持亿速云,关注亿速云行业资讯频道,更多相关知识等着你来学习!
免责声明:本站发布的内容(图片、视频和文字)以原创、转载和分享为主,文章观点不代表本网站立场,如果涉及侵权请联系站长邮箱:is@yisu.com进行举报,并提供相关证据,一经查实,将立刻删除涉嫌侵权内容。