前言
上篇文章介绍了k8s使用pv/pvc 的方式使用cephfs,
k8s(十一)、分布式存储Cephfs使用
Ceph存储有三种存储接口,分别是:
对象存储 Ceph Object Gateway
块设备 RBD
文件系统 CEPHFS
Kubernetes支持后两种存储接口,支持的接入模式如下图:
在本篇将测试使用ceph rbd作持久化存储后端
RBD创建测试
rbd的使用分为3个步骤:
1.服务端/客户端创建块设备image
2.客户端将image映射进linux系统内核,内核识别出该块设备后生成dev下的文件标识
3.格式化块设备并挂载使用
# 创建
[root@h020112 ~]# rbd create --size 10240000 rbd/test1 #创建指定大小的node
[root@h020112 ~]# rbd info test1
rbd image 'test1':
size 10000 GB in 2560000 objects
order 22 (4096 kB objects)
block_name_prefix: rbd_data.31e3b6b8b4567
format: 2
features: layering, exclusive-lock,object-map, fast-diff, deep-flatten
flags:
#映射进内核操作之前,首先查看内核版本,jw版本的ceph默认format为2,默认开启5种特性,2.x及之前的内核版本需手动调整format为1
,4.x之前要关闭object-map, fast-diff, deep-flatten功能才能成功映射到内核,这里使用的是centos7.4,内核版本3.10
[root@h020112 ~]# rbd feature disable test1 object-map fast-diff deep-flatten
# 映射进内核
[root@h020112 ~]# rbd map test1
[root@h020112 ~]# ls /dev/rbd0
/dev/rbd0
#挂载使用
[root@h020112 ~]# mkdir /mnt/cephrbd
[root@h020112 ~]# mkfs.ext4 -m0 /dev/rbd0
[root@h020112 ~]# mount /dev/rbd0 /mnt/cephrbd/
Kubernetes Dynamic Pv 使用
首先回顾一下上篇cephfs在k8s中的使用方式:
绑定使用流程: pv指定存储类型和接口,pvc绑定pv,pod挂载指定pvc.这种一一对应的方式,即为静态pv
例如cephfs的使用流程:
那么什么是动态pv呢?
在现有PV不满足PVC的请求时,可以使用存储分类(StorageClass),pvc对接sc后,k8s会动态地创建pv,并且向相应的存储后端发出资源申请.
描述具体过程为:首先创建sc分类,PVC请求已创建的sc的资源,来自动创建pv,这样就达到动态配置的效果。即通过一个叫 Storage Class的对象由存储系统根据PVC的要求自动创建pv并使用存储资源
Kubernetes的pv支持rbd,从第一部分测试使用rbd的过程可以看出,存储实际使用的单位是image,首先必须在ceph端创建image,才能创建pv/pvc,再在pod里面挂载使用.这是一个标准的静态pv的绑定使用流程.但是这样的流程就带来了一个弊端,即一个完整的存储资源挂载使用的操作被分割成了两段,一段在ceph端划分rbd image,一段由k8s端绑定pv,这样使得管理的复杂性提高,不利于自动化运维.
因此,动态pv是一个很好的选择,ceph rbd接口的动态pv的实现流程是这样的:
根据pvc及sc的声明,自动生成pv,且向sc指定的ceph后端发出申请自动创建相应大小的rbd image,最终绑定使用
工作流程图:
下面开始演示
1.创建secret
key获取方式参考上篇文章
~/mytest/ceph/rbd# cat ceph-secret-rbd.yaml
apiVersion: v1
kind: Secret
metadata:
name: ceph-secret-rbd
type: "kubernetes.io/rbd"
data:
key: QVFCL3E1ZGIvWFdxS1JBQTUyV0ZCUkxldnRjQzNidTFHZXlVYnc9PQ==
2.创建sc
~/mytest/ceph/rbd# cat ceph-secret-rbd.yaml
kind: StorageClass
apiVersion: storage.k8s.io/v1
metadata:
name: cephrbd-test2
provisioner: kubernetes.io/rbd
reclaimPolicy: Retain
parameters:
monitors: 192.168.20.112:6789,192.168.20.113:6789,192.168.20.114:6789
adminId: admin
adminSecretName: ceph-secret-rbd
pool: rbd
userId: admin
userSecretName: ceph-secret-rbd
fsType: ext4
imageFormat: "2"
imageFeatures: "layering"
创建pvc
~/mytest/ceph/rbd# cat cephrbd-pvc.yaml
kind: PersistentVolumeClaim
apiVersion: v1
metadata:
name: cephrbd-pvc2
spec:
storageClassName: cephrbd-test2
accessModes:
- ReadWriteOnce
resources:
requests:
storage: 20Gi
创建deploy,容器挂载使用创建的pvc
:~/mytest/ceph/rbd# cat deptest13dbdm.yaml
apiVersion: extensions/v1beta1
kind: Deployment
metadata:
labels:
app: deptest13dbdm
name: deptest13dbdm
namespace: default
spec:
replicas: 1
selector:
matchLabels:
app: deptest13dbdm
strategy:
rollingUpdate:
maxSurge: 1
maxUnavailable: 1
type: RollingUpdate
template:
metadata:
labels:
app: deptest13dbdm
spec:
containers:
image: registry.xxx.com:5000/mysql:56v6
imagePullPolicy: Always
name: deptest13dbdm
ports:
- containerPort: 3306
protocol: TCP
resources:
limits:
memory: 16Gi
requests:
memory: 512Mi
terminationMessagePath: /dev/termination-log
terminationMessagePolicy: File
volumeMounts:
- mountPath: /var/lib/mysql
name: datadir
subPath: deptest13dbdm
resources: {}
terminationMessagePath: /dev/termination-log
terminationMessagePolicy: File
dnsPolicy: ClusterFirst
restartPolicy: Always
schedulerName: default-scheduler
securityContext: {}
terminationGracePeriodSeconds: 30
volumes:
- name: datadir
persistentVolumeClaim:
claimName: cephrbd-pvc2
这里会遇到一个大坑:
~# kubectl describe pvc cephrbd-pvc2
...
Error creating rbd image: executable file not found in $PATH
这是因为kube-controller-manager需要调用rbd客户端api来创建rbd image,集群的每个节点都已经安装ceph-common,但是集群是使用kubeadm部署的,kube-controller-manager组件是以容器方式工作的,内部不包含rbd的执行文件,后面在这个issue里,找到了解决的办法:
Github issue
使用外部的rbd-provisioner,提供给kube-controller-manager以rbd的执行入口,注意,issue内的demo指定的image版本有bug,不能指定rbd挂载后使用的文件系统格式,需要使用最新的版本demo,项目地址:rbd-provisioner
完整demo yaml文件:
root@h009027:~/mytest/ceph/rbd/rbd-provisioner# cat rbd-provisioner.yaml
kind: ClusterRole
apiVersion: rbac.authorization.k8s.io/v1
metadata:
name: rbd-provisioner
rules:
- apiGroups: [""]
resources: ["persistentvolumes"]
verbs: ["get", "list", "watch", "create", "delete"]
- apiGroups: [""]
resources: ["persistentvolumeclaims"]
verbs: ["get", "list", "watch", "update"]
- apiGroups: ["storage.k8s.io"]
resources: ["storageclasses"]
verbs: ["get", "list", "watch"]
- apiGroups: [""]
resources: ["events"]
verbs: ["create", "update", "patch"]
- apiGroups: [""]
resources: ["services"]
resourceNames: ["kube-dns","coredns"]
verbs: ["list", "get"]
---
kind: ClusterRoleBinding
apiVersion: rbac.authorization.k8s.io/v1
metadata:
name: rbd-provisioner
subjects:
- kind: ServiceAccount
name: rbd-provisioner
namespace: default
roleRef:
kind: ClusterRole
name: rbd-provisioner
apiGroup: rbac.authorization.k8s.io
---
apiVersion: rbac.authorization.k8s.io/v1
kind: Role
metadata:
name: rbd-provisioner
rules:
- apiGroups: [""]
resources: ["secrets"]
verbs: ["get"]
- apiGroups: [""]
resources: ["endpoints"]
verbs: ["get", "list", "watch", "create", "update", "patch"]
---
apiVersion: rbac.authorization.k8s.io/v1
kind: RoleBinding
metadata:
name: rbd-provisioner
roleRef:
apiGroup: rbac.authorization.k8s.io
kind: Role
name: rbd-provisioner
subjects:
- kind: ServiceAccount
name: rbd-provisioner
namespace: default
---
apiVersion: extensions/v1beta1
kind: Deployment
metadata:
name: rbd-provisioner
spec:
replicas: 1
strategy:
type: Recreate
template:
metadata:
labels:
app: rbd-provisioner
spec:
containers:
- name: rbd-provisioner
image: quay.io/external_storage/rbd-provisioner:latest
env:
- name: PROVISIONER_NAME
value: ceph.com/rbd
serviceAccount: rbd-provisioner
---
apiVersion: v1
kind: ServiceAccount
metadata:
name: rbd-provisioner
部署完成后,重新创建上方的sc/pv,这个时候,pod就正常运行起来了,查看pvc,可以看到此pvc触发自动生成了一个pv且绑定了此pv,pv绑定的StorageClass则是刚创建的sc:
# pod
~/mytest/ceph/rbd/rbd-provisioner# kubectl get pods | grep deptest13
deptest13dbdm-c9d5bfb7c-rzzv6 2/2 Running 0 6h
# pvc
~/mytest/ceph/rbd/rbd-provisioner# kubectl get pvc cephrbd-pvc2
NAME STATUS VOLUME CAPACITY ACCESS MODES STORAGECLASS AGE
cephrbd-pvc2 Bound pvc-d976c2bf-c2fb-11e8-878f-141877468256 20Gi RWO cephrbd-test2 23h
# 自动创建的pv
~/mytest/ceph/rbd/rbd-provisioner# kubectl get pv pvc-d976c2bf-c2fb-11e8-878f-141877468256
NAME CAPACITY ACCESS MODES RECLAIM POLICY STATUS CLAIM STORAGECLASS REASON AGE
pvc-d976c2bf-c2fb-11e8-878f-141877468256 20Gi RWO Retain Bound default/cephrbd-pvc2 cephrbd-test2 23h
# sc
~/mytest/ceph/rbd/rbd-provisioner# kubectl get sc cephrbd-test2
NAME PROVISIONER AGE
cephrbd-test2 ceph.com/rbd 23h
再回到ceph节点上看看是否自动创建了rbd image
~# rbd ls
kubernetes-dynamic-pvc-d988cfb1-c2fb-11e8-b2df-0a58ac1a084e
test1
~# rbd info kubernetes-dynamic-pvc-d988cfb1-c2fb-11e8-b2df-0a58ac1a084e
rbd image 'kubernetes-dynamic-pvc-d988cfb1-c2fb-11e8-b2df-0a58ac1a084e':
size 20480 MB in 5120 objects
order 22 (4096 kB objects)
block_name_prefix: rbd_data.47d0b6b8b4567
format: 2
features: layering
flags:
可以发现,动态创建pv/rbd成功
挂载测试
回顾最上方的kubernetes支持的接入模式截图,rbd是不支持ReadWriteMany的,而k8s的理念是像牲口一样管理你的容器,容器节点有可能随时横向扩展,横向缩减,在节点间漂移,因此,不支持多点挂载的RBD模式,显然不适合做多应用节点之间的共享文件存储,这里进行一次进一步的验证
多点挂载测试
创建一个新的deploy,并挂载同一个pvc:
root@h009027:~/mytest/ceph/rbd# cat deptest14dbdm.yaml
apiVersion: extensions/v1beta1
kind: Deployment
metadata:
annotations:
deployment.kubernetes.io/revision: "4"
creationTimestamp: 2018-09-26T07:32:41Z
generation: 4
labels:
app: deptest14dbdm
domain: dev.deptest14dbm.kokoerp.com
nginx_server: ""
name: deptest14dbdm
namespace: default
resourceVersion: "26647927"
selfLink: /apis/extensions/v1beta1/namespaces/default/deployments/deptest14dbdm
uid: 5a1a62fe-c15e-11e8-878f-141877468256
spec:
replicas: 1
selector:
matchLabels:
app: deptest14dbdm
strategy:
rollingUpdate:
maxSurge: 1
maxUnavailable: 1
type: RollingUpdate
template:
metadata:
creationTimestamp: null
labels:
app: deptest14dbdm
domain: dev.deptest14dbm.kokoerp.com
nginx_server: ""
spec:
containers:
- env:
- name: DB_NAME
value: deptest14dbdm
image: registry.youkeshu.com:5000/mysql:56v6
imagePullPolicy: Always
name: deptest14dbdm
ports:
- containerPort: 3306
protocol: TCP
resources:
limits:
memory: 16Gi
requests:
memory: 512Mi
terminationMessagePath: /dev/termination-log
terminationMessagePolicy: File
volumeMounts:
- mountPath: /var/lib/mysql
name: datadir
subPath: deptest14dbdm
- env:
- name: DATA_SOURCE_NAME
value: exporter:6qQidF3F5wNhrLC0@(127.0.0.1:3306)/
image: registry.youkeshu.com:5000/mysqld-exporter:v0.10
imagePullPolicy: Always
name: mysql-exporter
ports:
- containerPort: 9104
protocol: TCP
resources: {}
terminationMessagePath: /dev/termination-log
terminationMessagePolicy: File
dnsPolicy: ClusterFirst
nodeSelector:
192.168.20.109: ""
restartPolicy: Always
schedulerName: default-scheduler
securityContext: {}
terminationGracePeriodSeconds: 30
volumes:
- name: datadir
persistentVolumeClaim:
claimName: cephrbd-pvc2
使用nodeselector将此deptest13和deptest14限制在同一个node中:
root@h009027:~/mytest/ceph/rbd# kubectl get pods -o wide | grep deptest1[3-4]
deptest13dbdm-c9d5bfb7c-rzzv6 2/2 Running 0 6h 172.26.8.82 h020109
deptest14dbdm-78b8bb7995-q84dk 2/2 Running 0 2m 172.26.8.84 h020109
不难发现在同一个节点中,是可以支持多个pod共享rbd image运行的,在容器的工作节点上查看mount条目也可以证实这一点:
192.168.20.109:~# mount | grep rbd
/dev/rbd0 on /var/lib/kubelet/plugins/kubernetes.io/rbd/rbd/rbd-image-kubernetes-dynamic-pvc-d988cfb1-c2fb-11e8-b2df-0a58ac1a084e type ext4 (rw,relatime,stripe=1024,data=ordered)
/dev/rbd0 on /var/lib/kubelet/pods/68b8e3c5-c38d-11e8-878f-141877468256/volumes/kubernetes.io~rbd/pvc-d976c2bf-c2fb-11e8-878f-141877468256 type ext4 (rw,relatime,stripe=1024,data=ordered)
/dev/rbd0 on /var/lib/kubelet/pods/1e9b04b1-c3c3-11e8-878f-141877468256/volumes/kubernetes.io~rbd/pvc-d976c2bf-c2fb-11e8-878f-141877468256 type ext4 (rw,relatime,stripe=1024,data=ordered)
修改新deploy的nodeselector指向其他的node:
~/mytest/ceph/rbd# kubectl get pods -o wide| grep deptest14db
deptest14dbdm-78b8bb7995-q84dk 0/2 Terminating 0 7m 172.26.8.84 h020109
deptest14dbdm-79fd495579-rkl92 0/2 ContainerCreating 0 49s <none> h009028
# kubectl describe pod deptest14dbdm-79fd495579-rkl92
...
Warning FailedAttachVolume 1m attachdetach-controller Multi-Attach error for volume "pvc-d976c2bf-c2fb-11e8-878f-141877468256" Volume is already exclusively attached to one node and can't be attached to another
pod将阻塞在ContainerCreating状态,描述报错很清晰地指出控制器不支持多点挂载此pv.
进行下一步测试前先删除此临时新增的deploy
滚动更新测试
在deptest13dbdm的deploy文件中指定的更新方式为滚动更新:
strategy:
rollingUpdate:
maxSurge: 1
maxUnavailable: 1
type: RollingUpdate
这里我们尝试一下,将正在运行的pod,从一个节点调度到另一个节点,rbd 是否能迁移到新的节点上挂载:
# 修改nodeselector
root@h009027:~/mytest/ceph/rbd# kubectl edit deploy deptest13dbdm
deployment "deptest13dbdm" edited
# 观察pod调度迁移过程
root@h009027:~/mytest/ceph/rbd# kubectl get pods -o wide| grep deptest13
deptest13dbdm-6999f5757c-fk9dp 0/2 ContainerCreating 0 4s <none> h009028
deptest13dbdm-c9d5bfb7c-rzzv6 2/2 Terminating 0 6h 172.26.8.82 h020109
root@h009027:~/mytest/ceph/rbd# kubectl describe pod deptest13dbdm-6999f5757c-fk9dp
...
Events:
Type Reason Age From Message
---- ------ ---- ---- -------
Normal Scheduled 2m default-scheduler Successfully assigned deptest13dbdm-6999f5757c-fk9dp to h009028
Normal SuccessfulMountVolume 2m kubelet, h009028 MountVolume.SetUp succeeded for volume "default-token-7q8pq"
Warning FailedMount 45s kubelet, h009028 Unable to mount volumes for pod "deptest13dbdm-6999f5757c-fk9dp_default(28e9cdd6-c3c5-11e8-878f-141877468256)": timeout expired waiting for volumes to attach/mount for pod "default"/"deptest13dbdm-6999f5757c-fk9dp". list of unattached/unmounted volumes=[datadir]
Normal SuccessfulMountVolume 35s (x2 over 36s) kubelet, h009028 MountVolume.SetUp succeeded for volume "pvc-d976c2bf-c2fb-11e8-878f-141877468256"
Normal Pulling 27s kubelet, h009028 pulling image "registry.youkeshu.com:5000/mysql:56v6"
Normal Pulled 27s kubelet, h009028 Successfully pulled image "registry.youkeshu.com:5000/mysql:56v6"
Normal Pulling 27s kubelet, h009028 pulling image "registry.youkeshu.com:5000/mysqld-exporter:v0.10"
Normal Pulled 26s kubelet, h009028 Successfully pulled image "registry.youkeshu.com:5000/mysqld-exporter:v0.10"
Normal Created 26s kubelet, h009028 Created container
Normal Started 26s kubelet, h009028 Started container
root@h009027:~/mytest/ceph/rbd# kubectl get pods -o wide| grep deptest13
deptest13dbdm-6999f5757c-fk9dp 2/2 Running 0 3m 172.26.0.87 h009028
#在原node上查看内核中映射的rbd image块设备描述文件已消失,挂载已取消
root@h020109:~/tpcc-mysql# ls /dev/rbd*
ls: cannot access '/dev/rbd*': No such file or directory
root@h020109:~/tpcc-mysql# mount | grep rbd
root@h020109:~/tpcc-mysql#
可以看出,是可以自动完成在节点间迁移的滚动更新的,整个调度的过程有些漫长,接近3分钟,这是因为其中至少涉及如下几个步骤:
1.原pod终结,释放资源
2.原node取消挂载rbd,释放内核映射
3.新node内核映射rbd image,挂载rbd到文件目录
4.创建pod,挂载指定目录
结论:
ceph rbd不支持多节点挂载,支持滚动更新.
后续
在完成了使用cephfs和ceph rbd这两种k8s分布式存储的对接方式测试之后,下一篇文章将针对这两种方式进行数据库性能测试及适用场景讨论