【Kubernetes系列十】Kubernetes StatefulSet控制器
1. StatefulSet介紹
前面使用Deployment建立的Pod是無狀態的,當掛載了volume之後,如果該Pod掛了,Replication Controller會再啟動一個Pod來保證可用性,但是由於Pod是無狀態的,pod掛了就會和之前的Volume的關係斷開,新建立的Pod無法找到之前的Pod。但是對於使用者來說,他們對底層的Pod掛了是沒有感知的,但是當Pod掛了之後就無法再使用之前掛載的儲存卷。為了解決這一問題,就引入了StatefulSet用於保留Pod的狀態資訊。
StatefulSet是Pod資源控制器的一種實現,用於部署和擴充套件有狀態應用的Pod資源,確保它們的執行順序及每個Pod資源的唯一性。其應用場景包括:
穩定的持久化儲存,即Pod重新排程後還是能訪問到相同的持久化資料,基於PVC來實現。
穩定的網路標識,即Pod重新排程後其PodName和HostName不變,基於Headless Service(即沒有Cluster IP的Service)來實現
有序部署,有序擴充套件,即Pod是有順序的,在部署或者擴充套件的時候要依據定義的順序依次進行(即從0到N-1,在下一個Pod執行之前的所有之前的Pod必須都是Running和Ready狀態),基於init Containers來實現
有序收縮,有序刪除(即從N-1到0)
StatefulSet由以下幾個部分組成:
用於定義網路標誌(DNS domain)和Headless Service
用於建立PersistentVolumes和VolumeClaimTemplates
定義具體應用的StatefulSet
StatefulSet中的每個Pod的DNS格式為statefulSetName-{0..N-1}.serviceName.namespace.svc.cluster.local,其中
serviceName:為Headless Service的名字
0..N-1:為Pod所在的序號,從0開始到N-1
statefulSetName:為StatefulSet的名字
namespace:為服務所在的namaspace,Headless Service和StatefulSet必須在相同的namespace
.cluster.local:為Cluster Domain
疑問解惑:
(1) 為什麼要有headless?
在Deployment中,每一個pod是沒有名稱,是隨機字串,是無序的。而statefulSet中是要求有序的,每一個Pod的名稱必須是固定的。當節點掛了,重建之後的識別符號是不變的,每一個節點的節點名稱是不會改變的。Pod名稱是作為Pod識別的唯一識別符號,必須保證其識別符號的穩定並且唯一。
為了實現識別符號的穩定,這時候就需要一個headless service解析直達到Pod,還需要給Pod配置一個唯一的名稱。
(2) 為什麼要有volumeClainTemplate?
大部分有狀態副本集都會用到持久儲存,比如分散式系統來說,由於資料是不一樣的,每個節點都需要自己專用的儲存節點。而在Deployment中Pod模板中建立的儲存卷是一個共享的儲存卷,多個Pod使用同一個儲存卷,而statefulSet定義中的每一個Pod都不能使用同一個儲存卷,由此基於Pod模板建立Pod是不適應的,這就需要引入volumeClainTemplate,當在使用StatefulSet建立Pod時,會自動生成一個PVC,從而請求繫結一個PV,從而有自己專用的儲存卷。
2. StatefulSet定義
在建立StatefulSet之前需要準備的東西,建立順序非常關鍵,如下
a、Volume
b、Persistent Volume
b、Persistent Volume Clain
d、Service
e、StatefulSet
Volume可以有很多中型別,比如nfs、gluster等,下面使用nfs
statefulSet欄位說明:
~]# kubectl explain statefulset.spec
podManagementPolicy <string> #Pod管理策略
replicas <integer> #Pod副本數量
revisionHistoryLimit <integer> #歷史版本限制
selector <Object> -required- #標籤選擇器,根據標籤選擇管理的Pod資源;必選欄位
serviceName <string> -required- #服務名稱,必選欄位
template <Object> -required- #模板,定義pod資源,必選欄位
updateStrategy <Object> #更新策略
volumeClaimTemplates <[]Object> #儲存卷申請模板,列表物件形式
示列:清單定義StatefulSet
通過上面的描述,下面示例定義StatefulSet資源,在定義之前首先得準備PV資源物件。這裡同樣使用NFS作為後端儲存。
(1) 準備NFS
~]# mkdir /data/volumes/v{1..5} -p
~]# vim /etc/exports
/data/volumes/v1 10.157.27.0/24(rw,no_root_squash)
/data/volumes/v2 10.157.27.0/24(rw,no_root_squash)
/data/volumes/v3 10.157.27.0/24(rw,no_root_squash)
/data/volumes/v4 10.157.27.0/24(rw,no_root_squash)
/data/volumes/v5 10.157.27.0/24(rw,no_root_squash)
~]# exportfs -arv
exporting 10.157.27.0/24:/data/volumes/v5
exporting 10.157.27.0/24:/data/volumes/v4
exporting 10.157.27.0/24:/data/volumes/v3
exporting 10.157.27.0/24:/data/volumes/v2
exporting 10.157.27.0/24:/data/volumes/v1
~]# showmount -e
Export list for storage:
/data/volumes/v5 10.157.27.0/24
/data/volumes/v4 10.157.27.0/24
/data/volumes/v3 10.157.27.0/24
/data/volumes/v2 10.157.27.0/24
/data/volumes/v1 10.157.27.0/24
(2) 建立pv
這裡建立5個PV,儲存大小各不相等,是否可讀也不相同,這裡新建立一個目錄用於存放statefulset所有的資源清單檔案等
~]# mkdir statefulset && cd statefulset
~]# vim pv-nfs.yaml
apiVersion: v1
kind: PersistentVolume
metadata:
name: pv-nfs-001
labels:
name: pv001
spec:
nfs:
path: /data/volumes/v1
server: 10.157.27.114
readOnly: false
accessModes: ["ReadWriteOnce","ReadWriteMany"]
capacity:
storage: 5Gi
persistentVolumeReclaimPolicy: Retain
---
apiVersion: v1
kind: PersistentVolume
metadata:
name: pv-nfs-002
labels:
name: pv002
spec:
nfs:
path: /data/volumes/v2
server: 10.157.27.114
readOnly: false
accessModes: ["ReadWriteOnce"]
capacity:
storage: 5Gi
persistentVolumeReclaimPolicy: Retain
---
apiVersion: v1
kind: PersistentVolume
metadata:
name: pv-nfs-003
labels:
name: pv003
spec:
nfs:
path: /data/volumes/v3
server: 10.157.27.114
readOnly: false
accessModes: ["ReadWriteOnce","ReadWriteMany"]
capacity:
storage: 5Gi
persistentVolumeReclaimPolicy: Retain
---
apiVersion: v1
kind: PersistentVolume
metadata:
name: pv-nfs-004
labels:
name: pv004
spec:
nfs:
path: /data/volumes/v4
server: 10.157.27.114
readOnly: false
accessModes: ["ReadWriteOnce","ReadWriteMany"]
capacity:
storage: 5Gi
persistentVolumeReclaimPolicy: Retain
---
apiVersion: v1
kind: PersistentVolume
metadata:
name: pv-nfs-005
labels:
name: pv005
spec:
nfs:
path: /data/volumes/v5
server: 10.157.27.114
readOnly: false
accessModes: ["ReadWriteOnce","ReadWriteMany"]
capacity:
storage: 5Gi
persistentVolumeReclaimPolicy: Retain
~]# kubectl apply -f pv-nfs.yaml
persistentvolume/pv-nfs-001 created
persistentvolume/pv-nfs-002 created
persistentvolume/pv-nfs-003 created
persistentvolume/pv-nfs-004 created
persistentvolume/pv-nfs-005 created
(3) 編寫定義StatefulSet的資源清單
首先我們要定義一個Headless Service,這裡headless Service和StatefulSet寫在一個檔案。
~]# vim statefulset-demo.yaml
#定義一個Headless Service
apiVersion: v1
kind: Service
metadata:
name: nginx-svc
labels:
app: nginx-svc
spec:
ports:
- name: http
port: 80
clusterIP: None
selector:
app: nginx-pod
---
#定義StatefulSet
apiVersion: apps/v1
kind: StatefulSet
metadata:
name: nginx-statefulset
spec:
serviceName: nginx-svc #指定service,和上面定義的service對應
replicas: 5 #指定副本數量
selector: #指定標籤選擇器,和後面的pod的標籤對應
matchLabels:
app: nginx-pod
template: #定義後端Pod的模板
metadata:
labels:
app: nginx-pod
spec:
containers:
- name: nginx
image: nginx:1.12
imagePullPolicy: IfNotPresent
ports:
- name: http
containerPort: 80
volumeMounts:
- name: nginxdata
mountPath: /usr/share/nginx/html
volumeClaimTemplates: #定義儲存卷申請模板
- metadata:
name: nginxdata
spec:
accessModes: ["ReadWriteOnce"]
resources:
requests:
storage: 5Gi
#---
#解析上面的資源清單:由於StatefulSet資源依賴於一個事先存在的Service資源,所以需要先定義一個名為nginx-svc的Headless Service資源,用於關聯到每個Pod資源建立DNS資源記錄。接著定義了一個名為nginx-statefulset的StatefulSet資源,它通過Pod模板建立了5個Pod資源副本,並基於volumeClaiTemplate向前面建立的PV進行了請求大小為5Gi的專用儲存卷。
(4) 建立StatefulSet資源
~]# kubectl apply -f statefulset-demo.yaml
service/nginx-svc created
statefulset.apps/nginx-statefulset created
~]# kubectl get svc #檢視建立的無頭服務nginx-svc
NAME TYPE CLUSTER-IP EXTERNAL-IP PORT(S) AGE
nginx-svc ClusterIP None <none> 80/TCP 8s
~]# kubectl get pv #檢視PV繫結
NAME CAPACITY ACCESS MODES RECLAIM POLICY STATUS CLAIM STORAGECLASS REASON AGE
pv-nfs-001 5Gi RWO,RWX Retain Available 3m25s
pv-nfs-002 5Gi RWO Retain Bound default/nginxdata-nginx-statefulset-0 3m25s
pv-nfs-003 5Gi RWO,RWX Retain Available 3m25s
pv-nfs-004 5Gi RWO,RWX Retain Available 3m25s
pv-nfs-005 5Gi RWO,RWX Retain Available
~]# kubectl get pvc #檢視PVC繫結
NAME STATUS VOLUME CAPACITY ACCESS MODES STORAGECLASS AGE
nginxdata-nginx-statefulset-0 Bound pv-nfs-002 5Gi RWO 17s
~]# kubectl get statefulset #檢視StatefulSet
NAME DESIRED CURRENT AGE
nginx-statefulset 5 2 72s
~]# kubectl get pods #檢視Pod資訊
(5) 刪除測試
~]# kubectl delete -f statefulset-demo.yaml
service "nginx-svc" deleted
statefulset.apps "nginx-statefulset" deleted
~]# kubectl get pods -w #動態檢視刪除過程,可以也是按照順序刪除,逆向關閉。
NAME READY STATUS RESTARTS AGE
nginx-statefulset-0 1/1 Running 0 18m
nginx-statefulset-1 1/1 Running 0 18m
nginx-statefulset-2 1/1 Running 0 18m
nginx-statefulset-2 1/1 Terminating 0 18m
nginx-statefulset-0 1/1 Terminating 0 18m
nginx-statefulset-1 1/1 Terminating 0 18m
nginx-statefulset-2 0/1 Terminating 0 18m
nginx-statefulset-0 0/1 Terminating 0 18m
nginx-statefulset-1 0/1 Terminating 0 18m
nginx-statefulset-2 0/1 Terminating 0 18m
nginx-statefulset-2 0/1 Terminating 0 18m
nginx-statefulset-2 0/1 Terminating 0 18m
nginx-statefulset-1 0/1 Terminating 0 18m
nginx-statefulset-1 0/1 Terminating 0 18m
nginx-statefulset-0 0/1 Terminating 0 18m
nginx-statefulset-0 0/1 Terminating 0 18m
此時PVC依舊存在的,再重新建立pod時,依舊會重新去繫結原來的PVC
~]# kubectl apply -f statefulset-demo.yaml
service/nginx-svc created
statefulset.apps/nginx-statefulset created
~]# kubectl get pvc #檢視PVC繫結
NAME STATUS VOLUME CAPACITY ACCESS MODES STORAGECLASS AGE
nginxdata-nginx-statefulset-0 Bound pv-nfs-002 5Gi RWO 30m
nginxdata-nginx-statefulset-1 Bound pv-nfs-003 5Gi RWO,RWX 30m
nginxdata-nginx-statefulset-2 Bound pv-nfs-004 5Gi RWO,RWX 30m
(6) 名稱解析
~]# kubectl get pods -o wide
NAME READY STATUS RESTARTS AGE IP NODE NOMINATED NODE
nginx-statefulset-0 1/1 Running 0 10m 10.244.1.196 node01.ilinux.io <none>
nginx-statefulset-1 1/1 Running 0 9m13s 10.244.2.167 node02.ilinux.io <none>
nginx-statefulset-2 1/1 Running 0 2m9s 10.244.3.42 node03.ilinux.io <none>
nginx-statefulset-3 1/1 Running 0 2m6s 10.244.0.28 master.ilinux.io <none>
nginx-statefulset-4 1/1 Running 0 54s 10.244.3.43 node03.ilinux.io <none>
~]# dig -t A nginx-statefulset-0.nginx-svc.default.svc.cluster.local @10.96.0.10
nginx-statefulset-0.nginx-svc.default.svc.cluster.local. 5 IN A 10.244.1.196
3.StatefulSet資源擴縮容
StatefulSet資源的擴縮容與Deployment資源相似,即通過修改資源的副本數來改動其目標Pod資源數量。對StatefulSet資源來說,kubectl scale和kubectl patch命令均可以實現此功能,也可以使用kubectl edit命令直接修改其副本數,或者修改資源清單檔案,由kubectl apply命令重新宣告
(1) 通過scale將nginx-statefulset資源副本數量擴容為4個
~]# kubectl scale statefulset/nginx-statefulset --replicas=4 #擴容副本增加到4個
~]# kubectl get pods #檢視pod資訊
NAME READY STATUS RESTARTS AGE
nginx-statefulset-0 1/1 Running 0 13m
nginx-statefulset-1 1/1 Running 0 12m
nginx-statefulset-2 1/1 Running 0 5m30s
nginx-statefulset-3 1/1 Running 0 5m27s
~]# kubectl get pv #檢視pv繫結
(2) 通過patch將nginx-statefulset資源副本數量縮容為3個
~]# kubectl patch sts/nginx-statefulset -p '{"spec":{"replicas":2}}' #通過patch打補丁方式縮容
~]# kubectl get pods -w #動態檢視縮容過程
NAME READY STATUS RESTARTS AGE
nginx-statefulset-0 1/1 Running 0 17m
nginx-statefulset-1 1/1 Running 0 17m
nginx-statefulset-2 1/1 Running 0 17m
nginx-statefulset-3 1/1 Running 0 1m
nginx-statefulset-3 1/1 Terminating 0 20s
nginx-statefulset-3 0/1 Terminating 0 20s
nginx-statefulset-3 0/1 Terminating 0 22s
nginx-statefulset-3 0/1 Terminating 0 22s
nginx-statefulset-2 1/1 Terminating 0 24s
nginx-statefulset-2 0/1 Terminating 0 24s
nginx-statefulset-2 0/1 Terminating 0 36s
nginx-statefulset-2 0/1 Terminating 0 36s
4.更新策略
StatefulSet的預設更新策略為滾動更新,也可以暫停更新
(1) 滾動更新示例:
~]# kubectl patch sts/nginx-statefulset -p '{"spec":{"replicas":4}}' #這裡先將副本擴容到4個。方便測試
~]# kubectl set image statefulset nginx-statefulset nginx=nginx:1.14 #更新映象版本
statefulset.apps/nginx-statefulset image updated
~]# kubectl get pods -w #動態檢視更新
NAME READY STATUS RESTARTS AGE
nginx-statefulset-0 1/1 Running 0 18m
nginx-statefulset-1 1/1 Running 0 18m
nginx-statefulset-2 1/1 Running 0 13m
nginx-statefulset-3 1/1 Running 0 13m
nginx-statefulset-3 1/1 Terminating 0 13m
nginx-statefulset-3 0/1 Terminating 0 13m
nginx-statefulset-3 0/1 Terminating 0 13m
nginx-statefulset-3 0/1 Terminating 0 13m
nginx-statefulset-3 0/1 Pending 0 0s
nginx-statefulset-3 0/1 Pending 0 0s
nginx-statefulset-3 0/1 ContainerCreating 0 0s
nginx-statefulset-3 1/1 Running 0 2s
nginx-statefulset-2 1/1 Terminating 0 13m
nginx-statefulset-2 0/1 Terminating 0 13m
nginx-statefulset-2 0/1 Terminating 0 14m
nginx-statefulset-2 0/1 Terminating 0 14m
nginx-statefulset-2 0/1 Pending 0 0s
nginx-statefulset-2 0/1 Pending 0 0s
nginx-statefulset-2 0/1 ContainerCreating 0 0s
nginx-statefulset-2 1/1 Running 0 1s
nginx-statefulset-1 1/1 Terminating 0 18m
nginx-statefulset-1 0/1 Terminating 0 18m
nginx-statefulset-1 0/1 Terminating 0 18m
nginx-statefulset-1 0/1 Terminating 0 18m
nginx-statefulset-1 0/1 Pending 0 0s
nginx-statefulset-1 0/1 Pending 0 0s
nginx-statefulset-1 0/1 ContainerCreating 0 0s
nginx-statefulset-1 1/1 Running 0 2s
nginx-statefulset-0 1/1 Terminating 0 18m
nginx-statefulset-0 0/1 Terminating 0 18m
nginx-statefulset-0 0/1 Terminating 0 18m
nginx-statefulset-0 0/1 Terminating 0 18m
nginx-statefulset-0 0/1 Pending 0 0s
nginx-statefulset-0 0/1 Pending 0 0s
nginx-statefulset-0 0/1 ContainerCreating 0 0s
nginx-statefulset-0 1/1 Running 0 2s
~]# kubectl get pods -l app=nginx-pod -o custom-columns=NAME:metadata.name,IMAGE:spec.containers[0].image #檢視更新完成後的映象版本
NAME IMAGE
nginx-statefulset-0 nginx:1.14
nginx-statefulset-1 nginx:1.14
nginx-statefulset-2 nginx:1.14
nginx-statefulset-3 nginx:1.14
通過上面示例可以看出,預設為滾動更新,倒序更新,更新完成一個接著更新下一個。
(2) 暫停更新示例:
有時候設定了一個更新操作,但是又不希望一次性全部更新完成,想先更新幾個,觀察其是否穩定,然後再更新所有的。這時候只需要將.spec.spec.updateStrategy.rollingUpdate.partition欄位的值進行修改即可。(預設值為0,所以我們看到了更新效果為上面那樣,全部更新)。該欄位表示如果設定為2,那麼只有當編號大於等於2的才會進行更新。類似於金絲雀的釋出方式。示例如下:
~]# kubectl patch sts/nginx-statefulset -p '{"spec":{"updateStrategy":{"rollingUpdate":{"partition":2}}}}}' #將更新值partition設定為2
~]# kubectl set image statefulset nginx-statefulset nginx=nginx:1.12 #更新映象版本
statefulset.apps/nginx-statefulset image updated
~]# kubectl get pods -w
~]# kubectl get pods -l app=nginx-pod -o custom-columns=NAME:metadata.name,IMAGE:spec.containers[0].image #檢視更新完成後的映象版本,可以發現只有當編號大於等於2的進行了更新。
NAME IMAGE
nginx-statefulset-0 nginx:1.14
nginx-statefulset-1 nginx:1.14
nginx-statefulset-2 nginx:1.12
nginx-statefulset-3 nginx:1.12
將剩餘的也全部更新,只需要將更新策略的partition的值改為0即可,如下:
~]# kubectl patch sts/nginx-statefulset -p '{"spec":{"updateStrategy":{"rollingUpdate":{"partition":0}}}}}' #將更新值partition設定為0
statefulset.apps/nginx-statefulset patche
~]# kubectl get pods -w #動態檢視更新