作者:老Z,雲原生愛好者,目前專注於雲原生運維,KubeSphere Ambassador。
Spring Cloud Alibaba 全家桶之 RocketMQ 是一款典型的分佈式架構下的消息中間件產品,使用異步通信方式和發佈訂閱的消息傳輸模型。
很多基於 Spring Cloud 開發的項目都喜歡採用 RocketMQ 作爲消息中間件。
RocketMQ 常用的部署模式如下:
- 單 Master 模式
- 多 Master 無 Slave 模式
- 多 Master 多 Slave 模式-異步複製
- 多 Master 多 Slave 模式-同步雙寫
更多的部署方案詳細信息可以參考官方文檔。
本文重點介紹 單 Master 模式和多 Master 多 Slave-異步複製模式在 K8s 集羣上的部署方案。
單 Master 模式
這種部署方式風險較大,僅部署一個 NameServer 和一個 Broker,一旦 Broker 重啓或者宕機時,會導致整個服務不可用,不建議線上生產環境使用,僅可以用於開發和測試環境。
部署方案參考官方rocketmq-docker項目中使用的容器化部署方案涉及的鏡像、啓動方式、定製化配置。
多 Master 多 Slave-異步複製模式
每個 Master 配置一個 Slave,有多對 Master-Slave,HA 採用異步複製方式,主備有短暫消息延遲(毫秒級),這種模式的優缺點如下:
- 優點:即使磁盤損壞,消息丟失的非常少,且消息實時性不會受影響,同時 Master 宕機後,消費者仍然可以從 Slave 消費,而且此過程對應用透明,不需要人工干預,性能同多 Master 模式幾乎一樣;
- 缺點:Master 宕機,磁盤損壞情況下會丟失少量消息。
多 Master 多 Slave-異步複製模式適用於生產環境,部署方案採用官方提供的 RocketMQ Operator。
離線鏡像製作
此過程爲可選項,離線內網環境可用,如果不配置內網鏡像,後續的資源配置清單中注意容器的 image 參數請使用默認值。
本文分別介紹了單 Master 模式和多 Master 多 Slave-異步複製模式部署 RocketMQ 使用的離線鏡像的製作方式。
-
單 Master 模式直接採用 RocketMQ 官方文檔中介紹的容器化部署方案中使用的鏡像。
-
多 Master 多 Slave-異步複製模式的離線鏡像製作方式採用 RocketMQ Operator 官方自帶的鏡像製作工具製作打包,製作過程中很多包都需要到國外網絡下載,但是受限於國外網絡訪問,默認成功率較低,需要多次嘗試或採取特殊手段 ( 懂的都懂)。
也可以用傳統的方式手工的 Pull Docker Hub 上已有的鏡像,然後再 Push 到私有鏡像倉庫。
在一臺能同時訪問互聯網和內網 Harbor 倉庫的服務器上進行下面的操作。
在 Harbor 中創建項目
本人習慣內網離線鏡像的命名空間跟應用鏡像默認的命名空間保持一致,因此,在 Harbor 中創建 apache 和 apacherocketmq 兩個項目,可以在 Harbor 管理界面中手工創建項目,也可以用下面命令行的方式自動化創建。
curl -u "admin:Harbor12345" -X POST -H "Content-Type: application/json" https://registry.zdevops.com.cn/api/v2.0/projects -d '{ "project_name": "apache", "public": true}'
curl -u "admin:Harbor12345" -X POST -H "Content-Type: application/json" https://registry.zdevops.com.cn/api/v2.0/projects -d '{ "project_name": "apacherocketmq", "public": true}'
安裝 Go 1.16
RocketMQ Operator 自定義鏡像製作需要用到 Go 環境,需要先安裝配置。
下載 Go 1.16 系列的最新版:
cd /opt/
wget https://golang.google.cn/dl/go1.16.15.linux-amd64.tar.gz
解壓源代碼到指定目錄:
tar zxvf go1.16.15.linux-amd64.tar.gz -C /usr/local/
配置環境變量:
cat >> /etc/profile.d/go.sh << EOF
# go environment
export GOROOT=/usr/local/go
export GOPATH=/srv/go
export PATH=$PATH:$GOROOT/bin:$GOPATH/bin
EOF
GOPATH 爲工作目錄也是代碼的存放目錄,可以根據自己的習慣配置
配置 Go:
go env -w GO111MODULE=on
go env -w GOPROXY=https://goproxy.cn,direct
驗證:
source /etc/profile.d/go.sh
go verison
獲取 RocketMQ Operator
從 Apache 官方 GitHub 倉庫獲取 rocketmq-operator 代碼。
cd /srv
git clone -b 0.3.0 https://github.com/apache/rocketmq-operator.git
製作 RocketMQ Operator Image
修改 DockerFile:
cd /srv/rocketmq-operator
vi Dockerfile
Notice: 構建鏡像的過程需訪問國外的軟件源和鏡像倉庫,在國內訪問有時會受限制,因此可以提前修改爲國內的軟件源和鏡像倉庫。
此操作爲可選項,如果訪問不受限則不需要配置。
必要的修改內容:
# 第 10 行(修改代理地址爲國內地址,加速訪問)
# 修改前
RUN go mod download
# 修改後
RUN go env -w GOPROXY=https://goproxy.cn,direct && go mod download
# 第 25 行(修改源地址爲國內源)
# 修改前
RUN apk add --no-cache bash gettext nmap-ncat openssl busybox-extras
# 修改後
RUN sed -i 's/dl-cdn.alpinelinux.org/mirrors.tuna.tsinghua.edu.cn/g' /etc/apk/repositories && \
apk add --no-cache bash gettext nmap-ncat openssl busybox-extras
可選的修改內容:
# 默認安裝的 ROCKETMQ版本爲 4.9.4,可以修改爲指定版本
# 第 28 行,修改 4.9.4
ENV ROCKETMQ_VERSION 4.9.4
製作鏡像:
yum install gcc
cd /srv/rocketmq-operator
go mod tidy
IMAGE_URL=registry.zdevops.com.cn/apacherocketmq/rocketmq-operator:0.3.0
make docker-build IMG=${IMAGE_URL}
驗證鏡像構建成功:
docker images | grep rocketmq-operator
推送鏡像:
make docker-push IMG=${IMAGE_URL}
清理臨時鏡像
docker rmi registry.zdevops.com.cn/apacherocketmq/rocketmq-operator:0.3.0
製作 RocketMQ Broker Image
修改 DockerFile(可選):
cd /srv/rocketmq-operator/images/broker/alpine
vi Dockerfile
此操作爲可選項,主要是爲了安裝軟件加速,如果訪問不受限則不需要配置。
# 第 20 行(修改源地址爲國內源)
# 修改前
RUN apk add --no-cache bash gettext nmap-ncat openssl busybox-extras
# 修改後
RUN sed -i 's/dl-cdn.alpinelinux.org/mirrors.tuna.tsinghua.edu.cn/g' /etc/apk/repositories && \
apk add --no-cache bash gettext nmap-ncat openssl busybox-extras
修改鏡像構建腳本:
# 修改鏡像倉庫地址爲內網地址
sed -i 's#apacherocketmq#registry.zdevops.com.cn/apacherocketmq#g' build-broker-image.sh
構建並推送鏡像:
./build-broker-image.sh 4.9.4
驗證鏡像構建成功:
docker images | grep rocketmq-broker
清理臨時鏡像:
docker rmi registry.zdevops.com.cn/apacherocketmq/rocketmq-broker:4.9.4-alpine-operator-0.3.0
製作 RocketMQ Name Server Image
修改 DockerFile(可選):
cd /srv/rocketmq-operator/images/namesrv/alpine
vi Dockerfile
此操作爲可選項,主要是爲了安裝軟件加速,如果訪問不受限則不需要配置。
# 第 20 行(修改源地址爲國內源)
# 修改前
RUN apk add --no-cache bash gettext nmap-ncat openssl busybox-extras
# 修改後
RUN sed -i 's/dl-cdn.alpinelinux.org/mirrors.tuna.tsinghua.edu.cn/g' /etc/apk/repositories && \
apk add --no-cache bash gettext nmap-ncat openssl busybox-extras
修改鏡像構建腳本:
# 修改鏡像倉庫地址爲內網地址
sed -i 's#apacherocketmq#registry.zdevops.com.cn/apacherocketmq#g' build-namesrv-image.sh
構建並推送鏡像:
./build-namesrv-image.sh 4.9.4
驗證鏡像構建成功:
docker images | grep rocketmq-nameserver
清理臨時鏡像:
docker rmi registry.zdevops.com.cn/apacherocketmq/rocketmq-nameserver:4.9.4-alpine-operator-0.3.0
根據官方已有鏡像製作離線鏡像
上面的 RocketMQ 多 Master 多 Slave-異步複製模式部署方案中用到的離線鏡像製作方案更適合於本地修改定製的場景,如果單純的只想把官方已有鏡像不做修改的下載並推送到本地倉庫,可以參考下面的方案。
下載鏡像:
docker pull apache/rocketmq-operator:0.3.0
docker pull apacherocketmq/rocketmq-nameserver:4.5.0-alpine-operator-0.3.0
docker pull apacherocketmq/rocketmq-broker:4.5.0-alpine-operator-0.3.0
Notice: 官方倉庫最新版的鏡像是 2 年前的 4.5.0.
重新打 tag:
docker tag apache/rocketmq-operator:0.3.0-snapshot registry.zdevops.com.cn/apacherocketmq/rocketmq-operator:0.3.0
docker tag apacherocketmq/rocketmq-nameserver:4.5.0-alpine-operator-0.3.0 registry.zdevops.com.cn/apacherocketmq/rocketmq-nameserver:4.5.0-alpine-operator-0.3.0
docker tag apacherocketmq/rocketmq-broker:4.5.0-alpine-operator-0.3.0 registry.zdevops.com.cn/apacherocketmq/rocketmq-broker:4.5.0-alpine-operator-0.3.0
推送到私有鏡像倉庫:
docker push registry.zdevops.com.cn/apacherocketmq/rocketmq-operator:0.3.0
docker push registry.zdevops.com.cn/apacherocketmq/rocketmq-nameserver:4.9.4-alpine-operator-0.3.0
docker push registry.zdevops.com.cn/apacherocketmq/rocketmq-broker:4.9.4-alpine-operator-0.3.0
清理臨時鏡像:
docker rmi apache/rocketmq-operator:0.3.0
docker rmi apacherocketmq/rocketmq-nameserver:4.5.0-alpine-operator-0.3.0
docker rmi apacherocketmq/rocketmq-broker:4.5.0-alpine-operator-0.3.0
docker rmi registry.zdevops.com.cn/apacherocketmq/rocketmq-operator:0.3.0
docker rmi registry.zdevops.com.cn/apacherocketmq/rocketmq-nameserver:4.5.0-alpine-operator-0.3.0
docker rmi registry.zdevops.com.cn/apacherocketmq/rocketmq-broker:4.5.0-alpine-operator-0.3.0
製作 RocketMQ Console Image
本文直接拉取官方鏡像作爲本地離線鏡像,如果需要修改內容並重構,可以參考 RocketMQ Console 使用的 官方 Dockerfile自行構建。
下載鏡像:
docker pull apacherocketmq/rocketmq-console:2.0.0
重新打 tag:
docker tag apacherocketmq/rocketmq-console:2.0.0 registry.zdevops.com.cn/apacherocketmq/rocketmq-console:2.0.0
推送到私有鏡像倉庫:
docker push registry.zdevops.com.cn/apacherocketmq/rocketmq-console:2.0.0
清理臨時鏡像:
docker rmi apacherocketmq/rocketmq-console:2.0.0
docker rmi registry.zdevops.com.cn/apacherocketmq/rocketmq-console:2.0.0
準備單 Master RocketMQ 部署方案涉及的離線鏡像
單 Master RocketMQ 部署方案涉及的鏡像跟集羣模式部署方案採用的 RocketMQ Operator 中使用的鏡像不同,在製作離線鏡像時,直接從官方鏡像庫拉取然後重新打 tag,再推送本地鏡像倉庫。
二者具體不同說明如下:
-
單 Master 方案使用的是 Docker Hub 中 apache 命名空間下的鏡像,並且鏡像名稱不區分 nameserver 和 broker,RocketMQ Operator 使用的是 apacherocketmq 命名空間下的鏡像,鏡像名稱區分 nameserver 和 broker。
-
單 Master 方案和 RocketMQ Operator 方案中管理工具使用的鏡像也不同,單 Master 方案使用的是 apacherocketmq 命名空間下的 rocketmq-dashboard 鏡像,RocketMQ Operator 使用的是 apacherocketmq 命名空間下的 rocketmq-console 鏡像。
具體的離線鏡像製作流程如下:
下載鏡像
docker pull apache/rocketmq:4.9.4
docker pull apacherocketmq/rocketmq-dashboard:1.0.0
重新打 tag
docker tag apache/rocketmq:4.9.4 registry.zdevops.com.cn/apache/rocketmq:4.9.4
docker tag apacherocketmq/rocketmq-dashboard:1.0.0 registry.zdevops.com.cn/apacherocketmq/rocketmq-dashboard:1.0.0
推送到私有鏡像倉庫
docker push registry.zdevops.com.cn/apache/rocketmq:4.9.4
docker push registry.zdevops.com.cn/apacherocketmq/rocketmq-dashboard:1.0.0
清理臨時鏡像
docker rmi apache/rocketmq:4.9.4
docker rmi apacherocketmq/rocketmq-dashboard:1.0.0
docker rmi registry.zdevops.com.cn/apache/rocketmq:4.9.4
docker rmi registry.zdevops.com.cn/apacherocketmq/rocketmq-dashboard:1.0.0
單 Master 模式部署
思路梳理
根據 RocketMQ 服務使用的組件,需要部署以下資源
- Broker StatefulSet
- NameServer StatefulSet
- NameServer Cluster Service:內部服務
- Dashboard Deployment
- Dashboard External Service:Dashboard 外部管理用
- ConfigMap:Broker 自定義配置文件
資源配置清單
參考 GitHub 中 Apache rocketmq-docker項目中介紹的容器化啓動示例配置,編寫適用於 K8S 的資源配置清單。
Notice: 每個人技術能力、技術習慣、服務環境有所不同,這裏介紹的只是我採用的一種簡單方式,並不一定是最優的方案,大家可以根據實際情況編寫適合自己的配置。
rocketmq-cm.yaml:
kind: ConfigMap
apiVersion: v1
metadata:
name: rocketmq-broker-config
namespace: zdevops
data:
BROKER_MEM: ' -Xms2g -Xmx2g -Xmn1g '
broker-common.conf: |-
brokerClusterName = DefaultCluster
brokerName = broker-0
brokerId = 0
deleteWhen = 04
fileReservedTime = 48
brokerRole = ASYNC_MASTER
flushDiskType = ASYNC_FLUSH
rocketmq-name-service-sts.yaml:
kind: StatefulSet
apiVersion: apps/v1
metadata:
name: rocketmq-name-service
namespace: zdevops
spec:
replicas: 1
selector:
matchLabels:
app: rocketmq-name-service
name_service_cr: rocketmq-name-service
template:
metadata:
labels:
app: rocketmq-name-service
name_service_cr: rocketmq-name-service
spec:
volumes:
- name: host-time
hostPath:
path: /etc/localtime
type: ''
containers:
- name: rocketmq-name-service
image: 'registry.zdevops.com.cn/apache/rocketmq:4.9.4'
command:
- /bin/sh
args:
- mqnamesrv
ports:
- name: tcp-9876
containerPort: 9876
protocol: TCP
resources:
limits:
cpu: 500m
memory: 1Gi
requests:
cpu: 250m
memory: 512Mi
volumeMounts:
- name: rocketmq-namesrv-storage
mountPath: /home/rocketmq/logs
subPath: logs
- name: host-time
readOnly: true
mountPath: /etc/localtime
imagePullPolicy: Always
volumeClaimTemplates:
- kind: PersistentVolumeClaim
apiVersion: v1
metadata:
name: rocketmq-namesrv-storage
spec:
accessModes:
- ReadWriteOnce
resources:
requests:
storage: 1Gi
storageClassName: glusterfs
volumeMode: Filesystem
serviceName: ''
---
kind: Service
apiVersion: v1
metadata:
name: rocketmq-name-server-service
namespace: zdevops
spec:
ports:
- name: tcp-9876
protocol: TCP
port: 9876
targetPort: 9876
selector:
name_service_cr: rocketmq-name-service
type: ClusterIP
rocketmq-broker-sts.yaml:
kind: StatefulSet
apiVersion: apps/v1
metadata:
name: rocketmq-broker-0-master
namespace: zdevops
spec:
replicas: 1
selector:
matchLabels:
app: rocketmq-broker
broker_cr: rocketmq-broker
template:
metadata:
labels:
app: rocketmq-broker
broker_cr: rocketmq-broker
spec:
volumes:
- name: rocketmq-broker-config
configMap:
name: rocketmq-broker-config
items:
- key: broker-common.conf
path: broker-common.conf
defaultMode: 420
- name: host-time
hostPath:
path: /etc/localtime
type: ''
containers:
- name: rocketmq-broker
image: 'apache/rocketmq:4.9.4'
command:
- /bin/sh
args:
- mqbroker
- "-c"
- /home/rocketmq/conf/broker-common.conf
ports:
- name: tcp-vip-10909
containerPort: 10909
protocol: TCP
- name: tcp-main-10911
containerPort: 10911
protocol: TCP
- name: tcp-ha-10912
containerPort: 10912
protocol: TCP
env:
- name: NAMESRV_ADDR
value: 'rocketmq-name-server-service.zdevops:9876'
- name: BROKER_MEM
valueFrom:
configMapKeyRef:
name: rocketmq-broker-config
key: BROKER_MEM
resources:
limits:
cpu: 500m
memory: 12Gi
requests:
cpu: 250m
memory: 2Gi
volumeMounts:
- name: host-time
readOnly: true
mountPath: /etc/localtime
- name: rocketmq-broker-storage
mountPath: /home/rocketmq/logs
subPath: logs/broker-0-master
- name: rocketmq-broker-storage
mountPath: /home/rocketmq/store
subPath: store/broker-0-master
- name: rocketmq-broker-config
mountPath: /home/rocketmq/conf/broker-common.conf
subPath: broker-common.conf
imagePullPolicy: Always
volumeClaimTemplates:
- kind: PersistentVolumeClaim
apiVersion: v1
metadata:
name: rocketmq-broker-storage
spec:
accessModes:
- ReadWriteOnce
resources:
requests:
storage: 8Gi
storageClassName: glusterfs
volumeMode: Filesystem
serviceName: ''
rocketmq-dashboard.yaml:
kind: Deployment
apiVersion: apps/v1
metadata:
name: rocketmq-dashboard
namespace: zdevops
spec:
replicas: 1
selector:
matchLabels:
app: rocketmq-dashboard
template:
metadata:
labels:
app: rocketmq-dashboard
spec:
containers:
- name: rocketmq-dashboard
image: 'registry.zdevops.com.cn/apacherocketmq/rocketmq-dashboard:1.0.0'
ports:
- name: http-8080
containerPort: 8080
protocol: TCP
env:
- name: JAVA_OPTS
value: >-
-Drocketmq.namesrv.addr=rocketmq-name-server-service.zdevops:9876
-Dcom.rocketmq.sendMessageWithVIPChannel=false
resources:
limits:
cpu: 500m
memory: 2Gi
requests:
cpu: 50m
memory: 512Mi
imagePullPolicy: Always
---
kind: Service
apiVersion: v1
metadata:
name: rocketmq-dashboard-service
namespace: zdevops
spec:
ports:
- name: http-8080
protocol: TCP
port: 8080
targetPort: 8080
nodePort: 31080
selector:
app: rocketmq-dashboard
type: NodePort
GitOps
本操作爲可選項,本人習慣在個人開發服務器上編輯或修改資源配置清單,然後提交到 Git 服務器 (Gitlab、Gitee、GitHub 等),然後在 k8s 節點上從 Git 服務器拉取資源配置清單並執行,從而實現資源配置清單的版本化管理,簡單的實現運維 GitOps。
本系列文檔的所有 k8s 資源配置清單文件,爲了演示和操作方便,都放在了統一的 k8s-yaml 倉庫中,實際工作中都是一個應用一個 Git 倉庫,更便於應用配置的版本控制。
大家在實際使用中可以忽略本步驟,直接在 k8s 節點上編寫資源配置清單並執行,也可以參考我的使用方式,實現簡單的 GitOps。
在個人運維開發服務器上操作:
# 在已有代碼倉庫創建 rocketmq/single 目錄
mkdir -p rocketmq/single
# 編輯資源配置清單
vi rocketmq/single/rocketmq-cm.yaml
vi rocketmq/single/rocketmq-name-service-sts.yaml
vi rocketmq/single/rocketmq-broker-sts.yaml
vi rocketmq/single/rocketmq-dashboard.yaml
# 提交 Git
git add rocketmq
git commit -am '添加 rocketmq 單節點資源配置清單'
git push
部署資源
在 k8s 集羣 Master 節點上或是獨立的運維管理服務器上操作。
更新鏡像倉庫代碼
cd /srv/k8s-yaml
git pull
部署資源 (分步式,二選一)
測試環境使用分步單獨部署的方式,以便測試資源配置清單的準確性。
cd /srv/k8s-yaml
kubectl apply -f rocketmq/single/rocketmq-cm.yaml
kubectl apply -f rocketmq/single/rocketmq-name-service-sts.yaml
kubectl apply -f rocketmq/single/rocketmq-broker-sts.yaml
kubectl apply -f rocketmq/single/rocketmq-dashboard.yaml
部署資源 (一鍵式,二選一)
實際使用中,可以直接 apply 整個目錄,實現一鍵式自動部署,在正式研發和生產環境中使用目錄的方式實現快速部署。
kubectl apply -f rocketmq/single/
驗證
ConfigMap:
$ kubectl get cm -n zdevops
NAME DATA AGE
kube-root-ca.crt 1 17d
rocketmq-broker-config 2 22s
StatefulSet:
$ kubectl get sts -o wide -n zdevops
NAME READY AGE CONTAINERS IMAGES
rocketmq-broker-0-master 1/1 11s rocketmq-broker registry.zdevops.com.cn/apache/rocketmq:4.9.4
rocketmq-name-service 1/1 12s rocketmq-name-service registry.zdevops.com.cn/apache/rocketmq:4.9.4
Deployment:
$ kubectl get deploy -o wide -n zdevops
NAME READY UP-TO-DATE AVAILABLE AGE CONTAINERS IMAGES SELECTOR
rocketmq-dashboard 1/1 1 1 31s rocketmq-dashboard registry.zdevops.com.cn/apacherocketmq/rocketmq-dashboard:1.0.0 app=rocketmq-dashboard
Pods:
$ kubectl get pods -o wide -n zdevops
NAME READY STATUS RESTARTS AGE IP NODE NOMINATED NODE READINESS GATES
rocketmq-broker-0-master-0 1/1 Running 0 77s 10.233.116.103 ks-k8s-master-2 <none> <none>
rocketmq-dashboard-b5dbb9d88-cwhqc 1/1 Running 0 3s 10.233.87.115 ks-k8s-master-1 <none> <none>
rocketmq-name-service-0 1/1 Running 0 78s 10.233.116.102 ks-k8s-master-2 <none> <none>
Service:
$ kubectl get svc -o wide -n zdevops
NAME TYPE CLUSTER-IP EXTERNAL-IP PORT(S) AGE SELECTOR
rocketmq-dashboard-service NodePort 10.233.5.237 <none> 8080:31080/TCP 74s app=rocketmq-dashboard
rocketmq-name-server-service ClusterIP 10.233.3.61 <none> 9876/TCP 2m29s name_service_cr=rocketmq-name-service
通過瀏覽器打開 K8S 集羣中任意節點的 IP:31080,可以看到 RocketMQ 控制檯的管理界面。
清理資源
卸載 RocketMQ 或是安裝失敗需要清理後重新安裝,可以在 K8S 集羣上使用下面的流程清理資源。
清理 StatefulSet:
kubectl delete sts rocketmq-broker-0-master -n zdevops
kubectl delete sts rocketmq-name-service -n zdevops
清理 Deployment:
kubectl delete deployments rocketmq-dashboard -n zdevops
清理 ConfigMap:
kubectl delete cm rocketmq-broker-config -n zdevops
清理服務:
kubectl delete svc rocketmq-name-server-service -n zdevops
kubectl delete svc rocketmq-dashboard-service -n zdevops
清理存儲卷:
kubectl delete pvc rocketmq-namesrv-storage-rocketmq-name-service-0 -n zdevops
kubectl delete pvc rocketmq-broker-storage-rocketmq-broker-0-master-0 -n zdevops
當然,也可以利用資源配置清單清理資源,更簡單快捷 (存儲卷無法自動清理,需要手工清理)。
$ kubectl delete -f rocketmq/single/
statefulset.apps "rocketmq-broker-0-master" deleted
configmap "rocketmq-broker-config" deleted
deployment.apps "rocketmq-dashboard" deleted
service "rocketmq-dashboard-service" deleted
statefulset.apps "rocketmq-name-service" deleted
service "rocketmq-name-server-service" deleted
多 Master 多 Slave-異步複製模式部署
思路梳理
多 Master 多 Slave-異步複製模式的 RocketMQ 部署,使用官方提供的 RocketMQ Operator,部署起來比較快速便捷,擴容也比較方便。
默認配置會部署 1 個 Master 和 1 個對應的 Slave,部署完成後可以根據需求擴容 Master 和 Slave。
獲取 RocketMQ Operator
# git 獲取代碼時指定版本
cd /srv
git clone -b 0.3.0 https://github.com/apache/rocketmq-operator.git
準備資源配置清單
本文演示的資源配置清單都是直接修改 rocketmq-operator 默認的配置,生產環境應根據默認配置修改一套適合自己環境的標準配置文件,並存放於 git 倉庫中。
爲 deploy 資源配置清單文件增加或修改命名空間:
cd /srv/rocketmq-operator
sed -i 'N;8 a \ namespace: zdevops' deploy/crds/rocketmq.apache.org_brokers.yaml
sed -i 'N;8 a \ namespace: zdevops' deploy/crds/rocketmq.apache.org_consoles.yaml
sed -i 'N;8 a \ namespace: zdevops' deploy/crds/rocketmq.apache.org_nameservices.yaml
sed -i 'N;8 a \ namespace: zdevops' deploy/crds/rocketmq.apache.org_topictransfers.yaml
sed -i 'N;18 a \ namespace: zdevops' deploy/operator.yaml
sed -i 'N;18 a \ namespace: zdevops' deploy/role_binding.yaml
sed -i 's/namespace: default/namespace: zdevops/g' deploy/role_binding.yaml
sed -i 'N;18 a \ namespace: zdevops' deploy/service_account.yaml
sed -i 'N;20 a \ namespace: zdevops' deploy/role.yaml
切記此步驟只能執行一次,如果失敗了則需要刪掉後重新執行。
執行完成後一定要查看一下結果是否符合預期
grep -r zdevops deploy/*
。
修改 example 資源配置清單文件中的命名空間:
sed -i 's/namespace: default/namespace: zdevops/g' example/rocketmq_v1alpha1_rocketmq_cluster.yaml
sed -i 'N;18 a \ namespace: zdevops' example/rocketmq_v1alpha1_cluster_service.yaml
修改鏡像地址爲內網地址:
sed -i 's#apache/rocketmq-operator:0.3.0#registry.zdevops.com.cn/apacherocketmq/rocketmq-operator:0.3.0#g' deploy/operator.yaml
sed -i 's#apacherocketmq#registry.zdevops.com.cn/apacherocketmq#g' example/rocketmq_v1alpha1_rocketmq_cluster.yaml
修改 RocketMQ 版本 (可選):
sed -i 's/4.5.0/4.9.4/g' example/rocketmq_v1alpha1_rocketmq_cluster.yaml
Notice: 默認的資源配置清單示例中部署 RocketMQ 集羣的版本爲 4.5.0, 實際使用時請根據需求調整。
修改 NameService 網絡模式 (可選):
sed -i 's/hostNetwork: true/hostNetwork: false/g' example/rocketmq_v1alpha1_rocketmq_cluster.yaml
sed -i 's/dnsPolicy: ClusterFirstWithHostNet/dnsPolicy: ClusterFirst/g' example/rocketmq_v1alpha1_rocketmq_cluster.yaml
Notice: 官方示例默認配置使用 hostNetwork 模式 , 適用於同時給 K8S 集羣內、外應用提供服務 , 實際使用時請根據需求調整 .
個人傾向於禁用 hostNetwork 模式 , 不跟外部應用混用 . 如果需要混用 , 則傾向於在外部獨立部署 RocketMQ。
修改 storageClassName 爲 glusterfs:
sed -i 's/storageClassName: rocketmq-storage/storageClassName: glusterfs/g' example/rocketmq_v1alpha1_rocketmq_cluster.yaml
sed -i 's/storageMode: EmptyDir/storageMode: StorageClass/g' example/rocketmq_v1alpha1_rocketmq_cluster.yaml
Notice: 演示環境 GlusterFS 存儲對應的 storageClassName 爲 glusterfs,請根據實際情況修改。
修改 nameServers 爲域名的形式:
sed -i 's/nameServers: ""/nameServers: "name-server-service.zdevops:9876"/g' example/rocketmq_v1alpha1_rocketmq_cluster.yaml
Notice: name-server-service.zdevops 是 NameServer service 名稱 + 項目名稱的組合
默認配置採用 pod [ip:port] 的形式 , 一旦 Pod IP 發生變化 ,Console 就沒法管理集羣了 , 且 Console 不會自動變更配置,如果設置爲空的話可能還會出現隨便配置的情況,因此一定要提前修改。
修改 RocketMQ Console 外部訪問的 NodePort:
sed -i 's/nodePort: 30000/nodePort: 31080/g' example/rocketmq_v1alpha1_cluster_service.yaml
Notice: 官方示例默認配置端口號爲 30000, 實際使用時請根據需求調整。
修改 RocketMQ NameServer 和 Console 的 service 配置:
sed -i '32,46s/^#//g' example/rocketmq_v1alpha1_cluster_service.yaml
sed -i 's/nodePort: 30001/nodePort: 31081/g' example/rocketmq_v1alpha1_cluster_service.yaml
sed -i 's/namespace: default/namespace: zdevops/g' example/rocketmq_v1alpha1_cluster_service.yaml
NameServer 默認使用了 NodePort 的形式,單純在 K8S 集羣內部使用的話,可以修改爲集羣模式。
GitOps
生產環境實際使用時建議將上面編輯整理後的資源配置清單,單獨整理,刪除 rocketmq-operator 項目中多餘的文件,行成一套適合於自己業務需要的資源配置清單,並使用 Git 進行版本控制。
單 Master 模式部署方案中已經詳細介紹過操作流程,此處不再多做介紹。
4.5. 部署 RocketMQ Operator (自動)
官方介紹的自動部署方法,適用於能連接互聯網的環境,部署過程中需要下載 controller-gen 和 kustomize 二進制文件,同時會下載一堆 go 依賴。
不適合於內網離線環境,這裏只是簡單介紹,本文重點採用後面的手動部署的方案。
部署 RocketMQ Operator:
make deploy
部署 RocketMQ Operator (手動)
部署 RocketMQ Operator:
kubectl create -f deploy/crds/rocketmq.apache.org_brokers.yaml
kubectl create -f deploy/crds/rocketmq.apache.org_nameservices.yaml
kubectl create -f deploy/crds/rocketmq.apache.org_consoles.yaml
kubectl create -f deploy/crds/rocketmq.apache.org_topictransfers.yaml
kubectl create -f deploy/service_account.yaml
kubectl create -f deploy/role.yaml
kubectl create -f deploy/role_binding.yaml
kubectl create -f deploy/operator.yaml
驗證 CRDS:
$ kubectl get crd | grep rocketmq.apache.org
brokers.rocketmq.apache.org 2022-11-09T02:54:52Z
consoles.rocketmq.apache.org 2022-11-09T02:54:54Z
nameservices.rocketmq.apache.org 2022-11-09T02:54:53Z
topictransfers.rocketmq.apache.org 2022-11-09T02:54:54Z
驗證 RocketMQ Operator:
$ kubectl get deploy -n zdevops -o wide
NAME READY UP-TO-DATE AVAILABLE AGE CONTAINERS IMAGES SELECTOR
rocketmq-operator 1/1 1 1 6m46s manager registry.zdevops.com.cn/apacherocketmq/rocketmq-operator:0.3.0 name=rocketmq-operator
$ kubectl get pods -n zdevops -o wide
NAME READY STATUS RESTARTS AGE IP NODE NOMINATED NODE READINESS GATES
rocketmq-operator-7cc6b48796-htpk8 1/1 Running 0 2m28s 10.233.116.70 ks-k8s-master-2 <none> <none>
部署 RocketMQ 集羣
創建服務:
$ kubectl apply -f example/rocketmq_v1alpha1_cluster_service.yaml
service/console-service created
service/name-server-service created
創建集羣:
$ kubectl apply -f example/rocketmq_v1alpha1_rocketmq_cluster.yaml
configmap/broker-config created
broker.rocketmq.apache.org/broker created
nameservice.rocketmq.apache.org/name-service created
console.rocketmq.apache.org/console created
驗證
StatefulSet:
$ kubectl get sts -o wide -n zdevops
NAME READY AGE CONTAINERS IMAGES
broker-0-master 1/1 27s broker registry.zdevops.com.cn/apacherocketmq/rocketmq-broker:4.9.4-alpine-operator-0.3.0
broker-0-replica-1 1/1 27s broker registry.zdevops.com.cn/apacherocketmq/rocketmq-broker:4.9.4-alpine-operator-0.3.0
name-service 1/1 27s name-service registry.zdevops.com.cn/apacherocketmq/rocketmq-nameserver:4.9.4-alpine-operator-0.3.0
Deployment:
$ kubectl get deploy -o wide -n zdevops
NAME READY UP-TO-DATE AVAILABLE AGE CONTAINERS IMAGES SELECTOR
console 1/1 1 1 52s console registry.zdevops.com.cn/apacherocketmq/rocketmq-console:2.0.0 app=rocketmq-console
rocketmq-operator 1/1 1 1 4h43m manager registry.zdevops.com.cn/apacherocketmq/rocketmq-operator:0.3.0 name=rocketmq-operator
Pod:
$ kubectl get pods -o wide -n zdevops
NAME READY STATUS RESTARTS AGE IP NODE NOMINATED NODE READINESS GATES
broker-0-master-0 1/1 Running 0 47s 10.233.87.24 ks-k8s-master-1 <none> <none>
broker-0-replica-1-0 1/1 Running 0 17s 10.233.117.28 ks-k8s-master-0 <none> <none>
console-8d685798f-5pwct 1/1 Running 0 116s 10.233.116.84 ks-k8s-master-2 <none> <none>
name-service-0 1/1 Running 0 96s 10.233.116.85 ks-k8s-master-2 <none> <none>
rocketmq-operator-7cc6b48796-htpk8 1/1 Running 2 (98s ago) 4h39m 10.233.116.70 ks-k8s-master-2 <none> <none>
Services:
$ kubectl get svc -o wide -n zdevops
NAME TYPE CLUSTER-IP EXTERNAL-IP PORT(S) AGE SELECTOR
console-service NodePort 10.233.38.15 <none> 8080:31080/TCP 21m app=rocketmq-console
name-server-service NodePort 10.233.56.238 <none> 9876:31081/TCP 21m name_service_cr=name-service
通過瀏覽器打開 K8S 集羣中任意節點的 IP:31080,可以看到 RocketMQ 控制檯的管理界面。
清理資源
清理 RocketMQ Cluster
部署集羣失敗或是需要重新部署時,採用下面的順序清理刪除。
kubectl delete -f example/rocketmq_v1alpha1_rocketmq_cluster.yaml
kubectl delete -f example/rocketmq_v1alpha1_cluster_service.yaml
清理 RocketMQ Operator
kubectl delete -f deploy/crds/rocketmq.apache.org_brokers.yaml
kubectl delete -f deploy/crds/rocketmq.apache.org_nameservices.yaml
kubectl delete -f deploy/crds/rocketmq.apache.org_consoles.yaml
kubectl delete -f deploy/crds/rocketmq.apache.org_topictransfers.yaml
kubectl delete -f deploy/service_account.yaml
kubectl delete -f deploy/role.yaml
kubectl delete -f deploy/role_binding.yaml
kubectl delete -f deploy/operator.yaml
清理存儲卷
需要手工查找 Broker 和 NameServer 相關的存儲卷並刪除。
# 查找存儲卷
$ kubectl get pvc -n zdevops
NAME STATUS VOLUME CAPACITY ACCESS MODES STORAGECLASS AGE
broker-storage-broker-0-master-0 Bound pvc-6a78b573-d72a-47ca-9012-5bc888dfcb0f 8Gi RWO glusterfs 3m54s
broker-storage-broker-0-replica-1-0 Bound pvc-4f096942-505d-4e34-ac7f-b871b9f33df3 8Gi RWO glusterfs 3m54s
namesrv-storage-name-service-0 Bound pvc-2c45a77e-3ca1-4eab-bb57-8374aa9068d3 1Gi RWO glusterfs 3m54s
# 刪除存儲卷
kubectl delete pvc namesrv-storage-name-service-0 -n zdevops
kubectl delete pvc broker-storage-broker-0-master-0 -n zdevops
kubectl delete pvc broker-storage-broker-0-replica-1-0 -n zdevops
擴容 NameServer
如果當前的 name service 集羣規模不能滿足您的需求,您可以簡單地使用 RocketMQ-Operator 來擴大或縮小 name service 集羣的規模。
擴容 name service 需要編寫並執行獨立的資源配置清單,參考官方示例Name Server Cluster Scale,並結合自己實際環境的 rocketmq-operator 配置修改。
Notice: 不要在已部署的資源中直接修改副本數,直接修改不會生效,會被 Operator 幹掉。
編輯擴容 NameServer 資源配置清單 , rocketmq_v1alpha1_nameservice_cr.yaml:
apiVersion: rocketmq.apache.org/v1alpha1
kind: NameService
metadata:
name: name-service
namespace: zdevops
spec:
size: 2
nameServiceImage: registry.zdevops.com.cn/apacherocketmq/rocketmq-nameserver:4.9.4-alpine-operator-0.3.0
imagePullPolicy: Always
hostNetwork: false
dnsPolicy: ClusterFirst
resources:
requests:
memory: "512Mi"
cpu: "250m"
limits:
memory: "1024Mi"
cpu: "500m"
storageMode: StorageClass
hostPath: /data/rocketmq/nameserver
volumeClaimTemplates:
- metadata:
name: namesrv-storage
spec:
accessModes:
- ReadWriteOnce
storageClassName: glusterfs
resources:
requests:
storage: 1Gi
執行擴容操作:
kubectl apply -f rocketmq/cluster/rocketmq_v1alpha1_nameservice_cr.yaml
驗證 StatefulSet:
$ kubectl get sts name-service -o wide -n zdevops
NAME READY AGE CONTAINERS IMAGES
name-service 2/2 16m name-service registry.zdevops.com.cn/apacherocketmq/rocketmq-nameserver:4.9.4-alpine-operator-0.3.0
驗證 Pods:
$ kubectl get pods -o wide -n zdevops
NAME READY STATUS RESTARTS AGE IP NODE NOMINATED NODE READINESS GATES
broker-0-master-0 1/1 Running 0 18m 10.233.87.117 ks-k8s-master-1 <none> <none>
broker-0-replica-1-0 1/1 Running 0 43s 10.233.117.99 ks-k8s-master-0 <none> <none>
console-8d685798f-hnmvg 1/1 Running 0 18m 10.233.116.113 ks-k8s-master-2 <none> <none>
name-service-0 1/1 Running 0 18m 10.233.116.114 ks-k8s-master-2 <none> <none>
name-service-1 1/1 Running 0 110s 10.233.87.120 ks-k8s-master-1 <none> <none>
rocketmq-operator-6db8ccc685-5hkk8 1/1 Running 0 18m 10.233.116.112 ks-k8s-master-2 <none> <none>
特別說明
NameServer 擴容一定要慎重,在實際驗證測試中發現 NameServer 擴容會導致重建已有的除了 Broker-0 的 Master 之外的其他 Broker 的 Master 和 所有的 Slave。按官方文檔上的說明,應該是 Operator 通知所有的 Broker 更新 name service list parameters,以便它們可以註冊到新的 NameServer Service。
同時,在 allowRestart: true 策略下,Broker 將逐漸更新,因此更新過程也不會被生產者和消費者客戶端感知,也就是說理論上不會影響業務(未實際測試)。
但是,所有 Broker 的 Master 和 Slave 重建後,查看集羣狀態時,集羣節點的信息不穩定,有的時候能看到 3 個節點,有的時候則能看到 4 個節點。
因此,生產環境最好在初次部署的時候就配置 NameServer 的副本數爲 2 或是 3,儘量不要在後期擴容,除非你能搞定擴容造成的一切後果。
擴容 Broker
通常情況下,隨着業務的發展,現有的 Broker 集羣規模可能不再滿足您的業務需求。你可以簡單地使用 RocketMQ-Operator 來升級、擴容 Broker 集羣。
擴容 Broker 需要編寫並執行獨立的資源配置清單,參考官方示例Broker Cluster Scale,並結合自己實際環境的 rocketmq-operator 配置修改。
編輯擴容 Broker 資源配置清單 , rocketmq_v1alpha1_broker_cr.yaml:
apiVersion: rocketmq.apache.org/v1alpha1
kind: Broker
metadata:
name: broker
namespace: zdevops
spec:
size: 2
nameServers: "name-server-service.zdevops::9876"
replicaPerGroup: 1
brokerImage: registry.zdevops.com.cn/apacherocketmq/rocketmq-broker:4.9.4-alpine-operator-0.3.0
imagePullPolicy: Always
resources:
requests:
memory: "2048Mi"
cpu: "250m"
limits:
memory: "12288Mi"
cpu: "500m"
allowRestart: true
storageMode: StorageClass
hostPath: /data/rocketmq/broker
# scalePodName is [Broker name]-[broker group number]-master-0
scalePodName: broker-0-master-0
env:
- name: BROKER_MEM
valueFrom:
configMapKeyRef:
name: broker-config
key: BROKER_MEM
volumes:
- name: broker-config
configMap:
name: broker-config
items:
- key: broker-common.conf
path: broker-common.conf
volumeClaimTemplates:
- metadata:
name: broker-storage
spec:
accessModes:
- ReadWriteOnce
storageClassName: glusterfs
resources:
requests:
storage: 8Gi
Notice: 注意重點字段 scalePodName: broker-0-master-0。
選擇源 Broker pod,將從其中將主題和訂閱信息數據等舊元數據傳輸到新創建的 Broker。
執行擴容 Broker:
kubectl apply -f rocketmq/cluster/rocketmq_v1alpha1_broker_cr.yaml
Notice: 執行成功後將部署一個新的 Broker Pod 組,同時 Operator 將在啓動新 Broker 之前將源 Broker Pod 中的元數據複製到新創建的 Broker Pod 中,因此新 Broker 將重新加載已有的主題和訂閱信息。
驗證 StatefulSet:
$ kubectl get sts -o wide -n zdevops
NAME READY AGE CONTAINERS IMAGES
broker-0-master 1/1 43m broker registry.zdevops.com.cn/apacherocketmq/rocketmq-broker:4.9.4-alpine-operator-0.3.0
broker-0-replica-1 1/1 43m broker registry.zdevops.com.cn/apacherocketmq/rocketmq-broker:4.9.4-alpine-operator-0.3.0
broker-1-master 1/1 27s broker registry.zdevops.com.cn/apacherocketmq/rocketmq-broker:4.9.4-alpine-operator-0.3.0
broker-1-replica-1 1/1 27s broker registry.zdevops.com.cn/apacherocketmq/rocketmq-broker:4.9.4-alpine-operator-0.3.0
name-service 2/2 43m name-service registry.zdevops.com.cn/apacherocketmq/rocketmq-nameserver:4.9.4-alpine-operator-0.3.0
驗證 Pods:
$ kubectl get pods -o wide -n zdevops
NAME READY STATUS RESTARTS AGE IP NODE NOMINATED NODE READINESS GATES
broker-0-master-0 1/1 Running 0 44m 10.233.87.117 ks-k8s-master-1 <none> <none>
broker-0-replica-1-0 1/1 Running 0 26m 10.233.117.99 ks-k8s-master-0 <none> <none>
broker-1-master-0 1/1 Running 0 72s 10.233.116.117 ks-k8s-master-2 <none> <none>
broker-1-replica-1-0 1/1 Running 0 72s 10.233.117.100 ks-k8s-master-0 <none> <none>
console-8d685798f-hnmvg 1/1 Running 0 44m 10.233.116.113 ks-k8s-master-2 <none> <none>
name-service-0 1/1 Running 0 44m 10.233.116.114 ks-k8s-master-2 <none> <none>
name-service-1 1/1 Running 0 27m 10.233.87.120 ks-k8s-master-1 <none> <none>
rocketmq-operator-6db8ccc685-5hkk8 1/1 Running 0 44m 10.233.116.112 ks-k8s-master-2 <none> <none>
在 KubeSphere 控制檯驗證:
在 RocketMQ 控制檯驗證:
常見問題
沒裝 gcc 編譯工具
報錯信息:
[root@zdevops-master rocketmq-operator]# make docker-build IMG=${IMAGE_URL}
/data/k8s-yaml/rocketmq-operator/bin/controller-gen rbac:roleName=rocketmq-operator crd:generateEmbeddedObjectMeta=true webhook paths="./..." output:dir=deploy output:crd:artifacts:config=deploy/crds
head -n 14 deploy/role_binding.yaml > deploy/role.yaml.bak
cat deploy/role.yaml >> deploy/role.yaml.bak
rm deploy/role.yaml && mv deploy/role.yaml.bak deploy/role.yaml
/data/k8s-yaml/rocketmq-operator/bin/controller-gen object:headerFile="hack/boilerplate.go.txt" paths="./..."
/usr/local/go/src/net/cgo_linux.go:12:8: no such package located
Error: not all generators ran successfully
run `controller-gen object:headerFile=hack/boilerplate.go.txt paths=./... -w` to see all available markers, or `controller-gen object:headerFile=hack/boilerplate.go.txt paths=./... -h` for usage
make: *** [generate] Error 1
解決方案:
$ yum install gcc
go mod 錯誤
報錯信息:
# 執行 make docker-build IMG=${IMAGE_URL} 報錯
go: creating new go.mod: module tmp
Downloading sigs.k8s.io/controller-tools/cmd/[email protected]
go get: added sigs.k8s.io/controller-tools v0.7.0
/data/build/rocketmq-operator/bin/controller-gen rbac:roleName=rocketmq-operator crd:generateEmbeddedObjectMeta=true webhook paths="./..." output:dir=deploy output:crd:artifacts:config=deploy/crds
Error: err: exit status 1: stderr: go: github.com/google/[email protected]: missing go.sum entry; to add it:
go mod download github.com/google/uuid
Usage:
controller-gen [flags]
......
output rules (optionally as output:<generator>:...)
+output:artifacts[:code=<string>],config=<string> package outputs artifacts to different locations, depending on whether they're package-associated or not.
+output:dir=<string> package outputs each artifact to the given directory, regardless of if it's package-associated or not.
+output:none package skips outputting anything.
+output:stdout package outputs everything to standard-out, with no separation.
run `controller-gen rbac:roleName=rocketmq-operator crd:generateEmbeddedObjectMeta=true webhook paths=./... output:dir=deploy output:crd:artifacts:config=deploy/crds -w` to see all available markers, or `controller-gen rbac:roleName=rocketmq-operator crd:generateEmbeddedObjectMeta=true webhook paths=./... output:dir=deploy output:crd:artifacts:config=deploy/crds -h` for usage
make: *** [manifests] Error 1
解決方案:
go mod tidy
結束語
本文只是初步介紹了 RocketMQ 在 K8s 平臺上的單 Master 節點和多 Master 多 Slave-異步複製模式部署的方法,屬於入門級。
在生產環境中還需要根據實際環境優化配置,例如調整集羣的 Broker 數量、Master 和 Slave 的分配、性能調優、配置優化等。
本文由博客一文多發平臺 OpenWrite 發佈!