k8s namespace限制调研

1.创建namespace gpu

2.增加限制

[root@tensorflow1 gpu-namespace]# cat compute-resources.yaml apiVersion: v1 kind: ResourceQuota metadata: name: compute-resources namespace: gpu spec: hard: pods: "5" requests.cpu: "1" requests.memory: 1Gi limits.cpu: "2" limits.memory: 2Gi [root@tensorflow1 gpu-namespace]# kubectl describe namespace gpu Name: gpu Labels: <none> Annotations: <none> Status: Active Resource Quotas Name: compute-resources Resource Used Hard -------- --- --- limits.cpu 0 2 limits.memory 0 2Gi pods 4 5 requests.cpu 0 1 requests.memory 0 1Gi No resource limits. 

3.检查限制情况

在已经创建好容器的情况下再增加限制，发现限 制并没有起作用，预期是memory限制到2g，结果是 从容器内仍然能看到8g内存

容器内：

root@tensorflow-ps-rc-cm9c8:/notebooks# free -m total used free shared buff/cache available Mem: 7783 1615 274 251 5893 5383 Swap: 0 0 0 

宿主机：

[root@tensorflow0 ~]# free -m total used free shared buff/cache available Mem: 7783 1616 272 251 5894 5382 Swap: 0 0 0 

4.杀掉容器

容器启动失败，要求对容器添加限制

[root@tensorflow1 gpu-namespace]# kubectl describe rc/tensorflow-ps-rc -n gpu ... Warning FailedCreate 2m replication-controller Error creating: pods "tensorflow-ps-rc-jrxxl" is forbidden: failed quota: compute-resources: must specify limits.cpu,limits.memory,requests.cpu,requests.memory Warning FailedCreate 23s (x9 over 2m) replication-controller (combined from similar events): Error creating: pods "tensorflow-ps-rc-sw9wx" is forbidden: failed quota: compute-resources: must specify limits.cpu,limits.memory,requests.cpu,requests.memory 

5.配置好限制，重启启动

增加配置：

resources: requests: memory: "1024Mi" cpu: "250m" limits: memory: "1024Mi" cpu: "500m" 

只启动了一个[root@tensorflow1 tf_gpu]# kubectl get all -o wide -n gpu

NAME READY STATUS RESTARTS AGE IP NODE po/tensorflow-ps-rc-9m8zj 1/1 Running 0 1h 10.244.2.91 tensorflow0 po/tensorflow-worker-rc-5zq9q 1/1 Running 0 11d 10.244.2.61 tensorflow0 po/tensorflow-worker-rc-mhncr 1/1 Running 0 11d 10.244.1.87 tensorflow2 NAME DESIRED CURRENT READY AGE CONTAINERS IMAGES SELECTOR rc/tensorflow-ps-rc 2 1 1 1h ps nfs:5000/tensorflow/tensorflow:nightly name=tensorflow-ps rc/tensorflow-worker-rc 2 2 2 11d worker nfs:5000/tensorflow/tensorflow:nightly-gpu name=tensorflow-worker NAME TYPE CLUSTER-IP EXTERNAL-IP PORT(S) AGE SELECTOR svc/tensorflow-ps-service ClusterIP 10.99.156.187 <none> 2222/TCP 11d name=tensorflow-ps svc/tensorflow-wk-service ClusterIP 10.102.251.161 <none> 2222/TCP 11d name=tensorflow-worker 

第二个不满足条件了

[root@tensorflow1 tf_gpu]# kubectl describe namespace gpu Name: gpu Labels: <none> Annotations: <none> Status: Active Resource Quotas Name: compute-resources Resource Used Hard -------- --- --- limits.cpu 500m 2 limits.memory 1Gi 2Gi pods 3 5 requests.cpu 250m 1 requests.memory 1Gi 1Gi No resource limits. [root@tensorflow1 tf_gpu]# kubectl describe rc/tensorflow-ps-rc -n gpu Warning FailedCreate 3m replication-controller Error creating: pods "tensorflow-ps-rc-cbt6c" is forbidden: exceeded quota: compute-resources, requested: requests.memory=1Gi, used: requests.memory=1Gi, limited: requests.memory=1Gi 

6.进入启动成功的那个容器

宿主机内存

[root@tensorflow0 ~]# free -m total used free shared buff/cache available Mem: 7783 1433 450 251 5899 5567 Swap: 0 0 0 

容器内存，与外面看到的一致。

root@tensorflow-ps-rc-9m8zj:/notebooks# free -m total used free shared buff/cache available Mem: 7783 1433 450 251 5899 5567 Swap: 0 0 0 

虽然限制了1G内存，但是仍能看到8G内存

本文转自CSDN-k8s namespace限制调研