k8s生成StarRocks集群模版

集群由1个fe+3个be组成，满足以下要求：

1、由3个pod组成，每pod分配2c4g

2、第一个pod里有一个be与一个fe，同在一个容器里，fe配置jvm内存设置为1024mb，be的jvm内存设置为1024MB

3、第二第三个pod里分别有一个be，jvm内存设置为2048MB

4、starrocks使用3.4.2镜像，已内置 Java 11

5、需要把数据挂载出来

6、pod需要再不同的node机器上

1. 第一个 Pod（FE + BE 共享 Pod）​

# pod-fe-with-be.yaml
apiVersion: v1
kind: Pod
metadata:name: starrocks-fe-with-belabels:app: starrockscomponent: fe
spec:containers:# FE 容器- name: feimage: starrocks/starrocks:3.4.2imagePullPolicy: IfNotPresentports:- containerPort: 9030  # HTTP 端口- containerPort: 9010  # BEACON 端口env:- name: JAVA_OPTSvalue: "-Xmx1024m -Xms1024m"  # FE JVM 1GBvolumeMounts:- name: fe-storagemountPath: /var/lib/starrocks/feresources:requests:cpu: "1"memory: "2Gi"limits:cpu: "1"memory: "2Gi"# 与 FE 同 Pod 的 BE 容器- name: beimage: starrocks/starrocks:3.4.2imagePullPolicy: IfNotPresentports:- containerPort: 9060  # BE 主端口env:- name: JAVA_OPTSvalue: "-Xmx1024m -Xms1024m"  # BE JVM 1GBvolumeMounts:- name: be-storage-1mountPath: /var/lib/starrocks/beresources:requests:cpu: "1"memory: "2Gi"limits:cpu: "1"memory: "2Gi"affinity:podAntiAffinity:requiredDuringSchedulingIgnoredDuringExecution:- labelSelector:matchExpressions:- key: appoperator: Invalues: ["starrocks"]topologyKey: kubernetes.io/hostnamevolumes:# FE 数据卷- name: fe-storagepersistentVolumeClaim:claimName: fe-pvc# BE 数据卷- name: be-storage-1persistentVolumeClaim:claimName: be-pvc-1

2. 第二个 Pod（独立 BE）

# pod-be-2.yaml
apiVersion: v1
kind: Pod
metadata:name: starrocks-be-2labels:app: starrockscomponent: be
spec:containers:- name: beimage: starrocks/starrocks:3.4.2imagePullPolicy: IfNotPresentports:- containerPort: 9060env:- name: JAVA_OPTSvalue: "-Xmx2048m -Xms2048m"  # BE JVM 2GBvolumeMounts:- name: be-storage-2mountPath: /var/lib/starrocks/beresources:requests:cpu: "1"memory: "2Gi"limits:cpu: "1"memory: "2Gi"affinity:podAntiAffinity:requiredDuringSchedulingIgnoredDuringExecution:- labelSelector:matchExpressions:- key: appoperator: Invalues: ["starrocks"]topologyKey: kubernetes.io/hostnamevolumes:- name: be-storage-2persistentVolumeClaim:claimName: be-pvc-2

3. 第三个 Pod（独立 BE）

# pod-be-3.yaml
apiVersion: v1
kind: Pod
metadata:name: starrocks-be-3labels:app: starrockscomponent: be
spec:containers:- name: beimage: starrocks/starrocks:3.4.2imagePullPolicy: IfNotPresentports:- containerPort: 9060env:- name: JAVA_OPTSvalue: "-Xmx2048m -Xms2048m"  # BE JVM 2GBvolumeMounts:- name: be-storage-3mountPath: /var/lib/starrocks/beresources:requests:cpu: "1"memory: "2Gi"limits:cpu: "1"memory: "2Gi"affinity:podAntiAffinity:requiredDuringSchedulingIgnoredDuringExecution:- labelSelector:matchExpressions:- key: appoperator: Invalues: ["starrocks"]topologyKey: kubernetes.io/hostnamevolumes:- name: be-storage-3persistentVolumeClaim:claimName: be-pvc-3

​4. 数据卷声明（PVC）

# pvc.yaml
apiVersion: v1
kind: PersistentVolumeClaim
metadata:name: fe-pvc
spec:accessModes:- ReadWriteOncestorageClassName: "ssd-storage"  # 替换为实际存储类resources:requests:storage: 100Gi---
apiVersion: v1
kind: PersistentVolumeClaim
metadata:name: be-pvc-1
spec:accessModes:- ReadWriteOncestorageClassName: "ssd-storage"resources:requests:storage: 500Gi---
apiVersion: v1
kind: PersistentVolumeClaim
metadata:name: be-pvc-2
spec:accessModes:- ReadWriteOncestorageClassName: "ssd-storage"resources:requests:storage: 500Gi---
apiVersion: v1
kind: PersistentVolumeClaim
metadata:name: be-pvc-3
spec:accessModes:- ReadWriteOncestorageClassName: "ssd-storage"resources:requests:storage: 500Gi

配置说明​​

1. ​​资源分配​​

   • ​​FE + BE Pod​​：FE 和 BE 容器各占 1C/2G，总和为 2C/4G。
   • ​​独立 BE Pod​​：BE 容器独占 2C/4G，但 JVM 内存限制为 2GB（需确保容器内存足够）。

2. ​​节点隔离​​

   • 使用 podAntiAffinity 强制每个 Pod 调度到不同节点（通过 topologyKey: kubernetes.io/hostname）。

3. ​​存储隔离​​

   • FE 和每个 BE 均挂载独立 PVC，路径分别为 /var/lib/starrocks/fe 和 /var/lib/starrocks/be。

4. ​​版本验证​​

   • 镜像 starrocks:3.4.2 已内置 Java 11，无需额外配置。

部署步骤​​

1. 创建 StorageClass
2. 应用 PVC 和 Pod 配置：

   kubectl apply -f pvc.yaml
   kubectl apply -f pod-fe-with-be.yaml
   kubectl apply -f pod-be-2.yaml
   kubectl apply -f pod-be-3.yaml

注意事项​​

• ​​内存风险​​：BE 容器的 JVM 内存（2GB）接近 Pod 内存限制（4GB），建议实际部署时预留更多内存或优化 JVM 参数。
• ​​存储类​​：替换 ssd-storage 为实际存储类型（如 gp2、standard）。
• ​​网络策略​​：确保 Kubernetes 集群节点间网络互通（尤其是 BE 之间的通信）。