Greenplum on Kubernetes
2021-03-10 本文已影响0人
程序员王旺
Greenplum 是一款基于postsqlgre实现的分布式关系型数据库，我们今天尝试下如何把它部署到kubernetes上。
docker-compose部署方式
在部署到kkubernetes之前，我们先来演示一个使用docker部署的例子，本例是基于这篇文章，感谢提供的基础镜像。
1. 下载greenplum
https://github.com/greenplum-db/gpdb/releases
2. 编译greenplum镜像
定义Dockerfile文件
# lyasper/gphost 基础镜像提供了sshd服务，及一些初始化脚本
FROM lyasper/gphost
# open-source-greenplum-xx...rpm为上面下载的greenplum安装包
COPY open-source-greenplum-db-6.14.1-rhel7-x86_64.rpm /home/gpadmin/greenplum-db.rpm
RUN yum install -y /home/gpadmin/greenplum-db.rpm
RUN chown -R gpadmin /usr/local/greenplum-db*
RUN rm -f /home/gpadmin/greenplum-db.rpm
docker build -t greenplum:6 .
3. 编写一个docker-compose.yaml文件
version: '3'
services:
mdw:
hostname: mdw
image: "greenplum:6"
ports:
- "6222:22"
- "6432:5432"
sdw1:
hostname: sdw1
image: "greenplum:6"
sdw2:
hostname: sdw2
image: "greenplum:6"
etl:
hostname: etl
image: "greenplum:6"
docker-compose up
4. 初始化greenplum
# 登陆master节点
ssh -p 6222 gpadmin@127.0.0.1
# 或者 ssh -p 6222 gpadmin@0.0.0.0
# 密码: changeme
# 初始化配置文件
source /usr/local/greenplum-db/greenplum_path.sh
# 配置greenplum文件
artifact/prepare.sh -s 2 -n 2
# -s: segment(容器)的个数
# -n: 每个segment(容器)上primary的个数
# 初始化集群，会生成env.sh 文件(greenplum所需的环境变量)
gpinitsystem -a -c gpinitsystem_config
source env.sh
# 开启远程无密码访问
artifact/postinstall.sh
# 查看安装结果
ps -ef | grep postgres
# 查看集群状态
gpstate -s
image.png
image.png
k8s 部署方式
制作k8s专用镜像（可选）
在部署到k8s之前，我们需要对现有的镜像做一些处理，生成一个k8s专用的greenplum镜像。稍后我会把制作好的镜像放到阿里云上，方便大家拉取。
- 先启动一个容器并进入
注意：docker run 后面一定不要加/bin/bash等参数，否则后面commit容器时会覆盖以前镜像的CMD参数，所以这里采用先启用再进入到容器的方式
$ docker run -it --rm greenplum:6
$ docker exec -it a2b9a823d845 /bin/bash
- 修改 artifact/prepare.sh 脚本，需修改3个地方
- 修改第7行master地址末尾加.gp：MASTERHOST=
hostname.gp - 修改第8行数据节点的前缀：SEG_PREFIX=gp-
- 修改87、90行，分别在地址后面加个 .gp
#!/bin/bash
set -x
if [ -z ${GPHOME+x} ]; then echo "GPHOME is unset";exit 1 ; fi
MASTERHOST=`hostname`.gp
SEG_PREFIX=gp-
SEG_HOSTNUM=0 # 0 means muster only
SEG_NUMPERHOST=1
VERBOSE=0
.....
if [ $SEG_HOSTNUM -eq 0 ];then
echo $MASTERHOST.gp > $HOSTFILE
else
for i in $(seq 1 $SEG_HOSTNUM); do
echo $SEG_PREFIX$i.gp >> $HOSTFILE
done
fi
- 修改/usr/local/greenplum-db-6.14.1/bin/gpinitsystem 脚本第1320行，在解析主机地址时，末尾添加.gp，这样就可以用gp-1.gp去连接segment服务器，而不是gp-1。
1314 HOST_LOOKUP() {
1315 res=`echo $1 | $GPHOSTCACHELOOKUP`
1316 err_index=`echo $res | awk '{print index($0,"__lookup_of_hostname_failed__")}'`
1317 if [ $err_index -ne 0 ]; then
1318 echo "__lookup_of_hostname_failed__"
1319 else
1320 echo $res.gp # 在主机
1321 fi
1322 }
- 创建一个初始化脚本
新建一个脚本 /home/gpadmin/init.sh，这个脚本仅在master上运行。
#!/bin/bash
# whoami: gpadmin
# only run on master
host=`hostname`
if [ $host = "gp-0" ];then
source /usr/local/greenplum-db/greenplum_path.sh
artifact/prepare.sh -s 2 -n 2
gpinitsystem -a -c gpinitsystem_config
source env.sh
artifact/postinstall.sh
gpstate -s
fi
- 生成一个新的镜像
我已经把镜像放到了阿里云上，所以后面直接拉取即可
$ docker commit -a "wangjc" -m "Greenplum on Kubernetes" 60875d7be058 greenplum-k8s:6.14.18
$ docker tag greenplum-k8s:6.14.18 registry.cn-hangzhou.aliyuncs.com/9c/greenplum-k8s:6.14.18
$ docker push registry.cn-hangzhou.aliyuncs.com/9c/greenplum-k8s:6.14.18
部署到k8s
我的 k8s版本为： 1.18.0，低版本是不支持PVC的，注意下
- 创建Local Volume
这里我们使用本地的存储卷来保存greenplum的数据，在部署之前，先保证 /var/greenplum目录存在且有足够的权限。
# gp-lv.yaml
### greenplum storage ###
apiVersion: v1
kind: PersistentVolume
metadata:
name: greenplum-pv
spec:
capacity:
storage: 2Gi
volumeMode: Filesystem
accessModes:
- ReadWriteOnce
persistentVolumeReclaimPolicy: Delete
storageClassName: greenplum-storage
local:
path: /var/greenplum
nodeAffinity:
required:
nodeSelectorTerms:
- matchExpressions:
- key: kubernetes.io/hostname
operator: In
values:
- master
- node1
- node2
---
kind: PersistentVolumeClaim
apiVersion: v1
metadata:
name: greenplum-claim
spec:
accessModes:
- ReadWriteOnce
resources:
requests:
storage: 2Gi
storageClassName: greenplum-storage
$ kubectl create -f gp-lv.yaml
$ kubectl get pv
$ kubectl get pvc
- 部署greenplum文件
- 这里采用StatefulSet方式部署，部署三个服务： gp-0作为master，gp-1、gp-2作为segment
- master服务开放两个端口：8432为gp数据库的端口、8222为ssh服务端口
# gp.yaml
### greenplum db ###
apiVersion: apps/v1
kind: StatefulSet
metadata:
name: gp
spec:
selector:
matchLabels:
app: gp
serviceName: gp
replicas: 3
template:
metadata:
labels:
app: gp
spec:
securityContext:
# 允许对非root用户访问PersistentVolumeClaim，否则无法写Local PVC
#runAsUser: 1000
fsGroup: 1000
volumes:
- name: gp-pv-storage
persistentVolumeClaim:
claimName: greenplum-claim
containers:
- name: gp-container
image: registry.cn-hangzhou.aliyuncs.com/9c/greenplum-k8s:6.14.18
imagePullPolicy: IfNotPresent
volumeMounts:
- mountPath: "/home/gpadmin/master"
name: gp-pv-storage
- mountPath: "/home/gpadmin/data"
name: gp-pv-storage
ports:
- name: gp-port
containerPort: 5432
- name: ssh-port
containerPort: 22
---
#service
apiVersion: v1
kind: Service
metadata:
name: gp
spec:
selector:
app: gp
type: ClusterIP
clusterIP: None
---
#service
apiVersion: v1
kind: Service
metadata:
name: gp-out
spec:
ports:
- name: gp-port
port: 8432
nodePort: 8432
protocol: TCP
targetPort: gp-port
- name: ssh-port
port: 8222
nodePort: 8222
protocol: TCP
targetPort: ssh-port
selector:
app: gp
statefulset.kubernetes.io/pod-name: gp-0
type: NodePort
$ kubectl create -f gp.yaml
$ kubectl get pods
- 初始化gp数据库
通过ssh连接上master服务器（密码是：changeme），进入到服务器内部进行初始化，192.168.x.x为k8s任意一个节点。初始化成功后，会在每个segment节点上生成两个段。
$ ssh -p 8222 gpadmin@192.168.x.x
$ ./init.sh
20210310:02:50:04:002506 gpstate:gp-0:gpadmin-[INFO]:-Obtaining Segment details from master...
20210310:02:50:04:002506 gpstate:gp-0:gpadmin-[INFO]:-Gathering data from segments...
20210310:02:50:05:002506 gpstate:gp-0:gpadmin-[INFO]:-----------------------------------------------------
20210310:02:50:05:002506 gpstate:gp-0:gpadmin-[INFO]:--Master Configuration & Status
20210310:02:50:05:002506 gpstate:gp-0:gpadmin-[INFO]:-----------------------------------------------------
20210310:02:50:05:002506 gpstate:gp-0:gpadmin-[INFO]:- Master host = gp-0
20210310:02:50:05:002506 gpstate:gp-0:gpadmin-[INFO]:- Master postgres process ID = 2345
20210310:02:50:05:002506 gpstate:gp-0:gpadmin-[INFO]:- Master data directory = /home/gpadmin/master/gpseg-1
20210310:02:50:05:002506 gpstate:gp-0:gpadmin-[INFO]:- Master port = 5432
20210310:02:50:05:002506 gpstate:gp-0:gpadmin-[INFO]:- Master current role = dispatch
20210310:02:50:05:002506 gpstate:gp-0:gpadmin-[INFO]:- Greenplum initsystem version = 6.14.1 build commit:5ef30dd4c9878abadc0124e0761e4b988455a4bd Open Source
20210310:02:50:05:002506 gpstate:gp-0:gpadmin-[INFO]:- Greenplum current version = PostgreSQL 9.4.24 (Greenplum Database 6.14.1 build commit:5ef30dd4c9878abadc0124e0761e4b988455a4bd Open Source) on x86_64-unknown-linux-gnu, compiled by gcc (GCC) 6.4.0, 64-bit compiled on Feb 22 2021 22:11:57
20210310:02:50:05:002506 gpstate:gp-0:gpadmin-[INFO]:- Postgres version = 9.4.24
20210310:02:50:05:002506 gpstate:gp-0:gpadmin-[INFO]:- Master standby = No master standby configured
20210310:02:50:05:002506 gpstate:gp-0:gpadmin-[INFO]:-----------------------------------------------------
20210310:02:50:05:002506 gpstate:gp-0:gpadmin-[INFO]:-Segment Instance Status Report
20210310:02:50:05:002506 gpstate:gp-0:gpadmin-[INFO]:-----------------------------------------------------
20210310:02:50:05:002506 gpstate:gp-0:gpadmin-[INFO]:- Segment Info
20210310:02:50:05:002506 gpstate:gp-0:gpadmin-[INFO]:- Hostname = gp-1.gp
20210310:02:50:05:002506 gpstate:gp-0:gpadmin-[INFO]:- Address = gp-1.gp
20210310:02:50:05:002506 gpstate:gp-0:gpadmin-[INFO]:- Datadir = /home/gpadmin/data/gpseg0
20210310:02:50:05:002506 gpstate:gp-0:gpadmin-[INFO]:- Port = 10000
20210310:02:50:05:002506 gpstate:gp-0:gpadmin-[INFO]:- Status
20210310:02:50:05:002506 gpstate:gp-0:gpadmin-[INFO]:- PID = 842
20210310:02:50:05:002506 gpstate:gp-0:gpadmin-[INFO]:- Configuration reports status as = Up
20210310:02:50:05:002506 gpstate:gp-0:gpadmin-[INFO]:- Database status = Up
20210310:02:50:05:002506 gpstate:gp-0:gpadmin-[INFO]:-----------------------------------------------------
20210310:02:50:05:002506 gpstate:gp-0:gpadmin-[INFO]:- Segment Info
20210310:02:50:05:002506 gpstate:gp-0:gpadmin-[INFO]:- Hostname = gp-1.gp
20210310:02:50:05:002506 gpstate:gp-0:gpadmin-[INFO]:- Address = gp-1.gp
20210310:02:50:05:002506 gpstate:gp-0:gpadmin-[INFO]:- Datadir = /home/gpadmin/data/gpseg1
20210310:02:50:05:002506 gpstate:gp-0:gpadmin-[INFO]:- Port = 10001
20210310:02:50:05:002506 gpstate:gp-0:gpadmin-[INFO]:- Status
20210310:02:50:05:002506 gpstate:gp-0:gpadmin-[INFO]:- PID = 843
20210310:02:50:05:002506 gpstate:gp-0:gpadmin-[INFO]:- Configuration reports status as = Up
20210310:02:50:05:002506 gpstate:gp-0:gpadmin-[INFO]:- Database status = Up
20210310:02:50:05:002506 gpstate:gp-0:gpadmin-[INFO]:-----------------------------------------------------
20210310:02:50:05:002506 gpstate:gp-0:gpadmin-[INFO]:- Segment Info
20210310:02:50:05:002506 gpstate:gp-0:gpadmin-[INFO]:- Hostname = gp-2.gp
20210310:02:50:05:002506 gpstate:gp-0:gpadmin-[INFO]:- Address = gp-2.gp
20210310:02:50:05:002506 gpstate:gp-0:gpadmin-[INFO]:- Datadir = /home/gpadmin/data/gpseg2
20210310:02:50:05:002506 gpstate:gp-0:gpadmin-[INFO]:- Port = 10000
20210310:02:50:05:002506 gpstate:gp-0:gpadmin-[INFO]:- Status
20210310:02:50:05:002506 gpstate:gp-0:gpadmin-[INFO]:- PID = 842
20210310:02:50:05:002506 gpstate:gp-0:gpadmin-[INFO]:- Configuration reports status as = Up
20210310:02:50:05:002506 gpstate:gp-0:gpadmin-[INFO]:- Database status = Up
20210310:02:50:05:002506 gpstate:gp-0:gpadmin-[INFO]:-----------------------------------------------------
20210310:02:50:05:002506 gpstate:gp-0:gpadmin-[INFO]:- Segment Info
20210310:02:50:05:002506 gpstate:gp-0:gpadmin-[INFO]:- Hostname = gp-2.gp
20210310:02:50:05:002506 gpstate:gp-0:gpadmin-[INFO]:- Address = gp-2.gp
20210310:02:50:05:002506 gpstate:gp-0:gpadmin-[INFO]:- Datadir = /home/gpadmin/data/gpseg3
20210310:02:50:05:002506 gpstate:gp-0:gpadmin-[INFO]:- Port = 10001
20210310:02:50:05:002506 gpstate:gp-0:gpadmin-[INFO]:- Status
20210310:02:50:05:002506 gpstate:gp-0:gpadmin-[INFO]:- PID = 843
20210310:02:50:05:002506 gpstate:gp-0:gpadmin-[INFO]:- Configuration reports status as = Up
20210310:02:50:05:002506 gpstate:gp-0:gpadmin-[INFO]:- Database status = Up
恢复数据
部署到k8s中最大的问题就是当pod重建时，容器被“打回原形”，但好在我们将gp的数据保存在宿主机上，但这时不能用上面 init.sh 脚本初始化了，这样会把之前的数据覆盖掉。因此需要对镜像再做修改。
按照前面的方法，启动一个容器，进去添加一个新的脚本 repair.sh ，专门用来修复数据的。
# 用上一个版本的镜像启动
$ docker run -it registry.cn-hangzhou.aliyuncs.com/9c/greenplum-k8s:6.14.18
在镜像中添加：repair.sh
#!/bin/bash
# whoami: gpadmin
# only run on master
host=`hostname`
if [ $host = "gp-0" ];then
source /usr/local/greenplum-db/greenplum_path.sh
# 注意下prepare.sh脚本，这里面有个删除master、data等目录的操作，需要注释掉
artifact/prepare.sh -s 2 -n 2
#gpinitsystem -a -c gpinitsystem_config
source env.sh
#artifact/postinstall.sh
# 解决pod重建后，挂载目录权限问题导致启动失败
sudo chmod 0700 -R $HOME/master/*
for HOST in `cat $HOME/hostfile`; do
ssh gpadmin@${HOST} "sudo chmod 0700 -R $HOME/data/*"
ssh gpadmin@${HOST} "sudo chmod 0700 -R $HOME/master/*"
ssh gpadmin@${HOST} "sudo chmod 0700 -R $HOME/mirror/*"
done
gpstart
fi
提交容器生成一个新的镜像
$ docker commit -a "wangjc" -m "Greenplum on Kubernetes" 135b86314ce1 registry.cn-hangzhou.aliyuncs.com/9c/greenplum-k8s:6.14.18
$ docker push registry.cn-hangzhou.aliyuncs.com/9c/greenplum-k8s:6.14.18
使用新镜像部署gp，并新建数据库添加数据，这时将gp-0删除，这里要注意的是：删除pod会导致svc失效，需要重新创建下。
$ kubectl delete pod gp-0
$ kubectl delete -f svc.yaml && kubectl create -f svc.yaml
#连接到gp库的master，进去执行修复操作
$ ssh -p 8222 gpadmin@192.168.x.x
#执行恢复脚本， 执行到最后需要手工输入一个：y
$ ./repair.sh
支持镜像节点
默认镜像节点功能是关闭的，还是按之前的方式进入容器中修改后重新提交一个新的容器。
- 修改 vi artifact/prepare.sh 文件添加如下代码
MIRDATASTR=""
for i in $(seq 1 $SEG_NUMPERHOST); do
MIRDATASTR="$MIRDATASTR $PREFIX/mirror"
done
# 增加一个MIRDATASTR替换
sed "s/%%PORT_BASE%%/$PORT_BASE/g; s|%%MIRDATASTR%%|$MIRDATASTR|g; s|%%PREFIX%%|$PREFIX|g;
- 修改 artifact/gpinitsystem_config_template 文件将所有以MIRROR开头的属性注释打开，并修改镜像目录如下
declare -a MIRROR_DATA_DIRECTORY=(%%MIRDATASTR%%)
- 修改init.sh 文件，修改3个段服务器，每个服务器分4个段
artifact/prepare.sh -s 3 -n 4
- 提交容器生成一个新的镜像
$ docker commit -a "wangjc" -m "Greenplum on Kubernetes" 135b86314ce1 registry.cn-hangzhou.aliyuncs.com/9c/greenplum-k8s:6.14.18
$ docker push registry.cn-hangzhou.aliyuncs.com/9c/greenplum-k8s:6.14.18
