前言

Cloudera与Hortonworks合并后，再去用CDH、CDP与HDP的Cloudera Manager与Ambari总是没安逸了。可白piao的CDH最高版本6.3.2，组件相对来说也比较古老，生产环境的CDH也陆续要脱保了。甲方爸爸不差钱可以交保护费，我等穷人嘛，白piao一时爽，一直白piao一直爽，能白piao的机会当然是不能放过。笔者没那么勤奋，懒得部署Apache组件了【需要考虑的兼容性、编译等问题很多】，图省心省力部署了一套USDP2的双集群。

硬件配置

双路e5 2696v3 256G内存主机

主板：x99f8d
CPU：e5 2696v3 *2 【36核72线程】
内存条：DDR4 ECC 32G *8 【256G】
显卡：GT730 【2G显存】
散热器：ta 120ex *2 【单风扇】
SSD：MX500 2T *2 【4T】
HDD：NAS拆的酷狼 4T
电源：GX1000 【1000W】
机箱：614PC 【标配2风扇+套装3风扇】
显示器：13.3寸便携 【2K】
键鼠：笔记本淘汰的一套 【USB口】
网卡：USB无线网卡

花销1w左右，省着点8k也够。

虚拟机准备

制作7台虚拟机

安装包下载

USDP官网，含下载路径及操作指南

安装部署

准备路径

[root@zhiyong1 ~]# cd ..
[root@zhiyong1 /]# mkdir /data
[root@zhiyong1 /]# mkdir -p /opt/usdp-srv/

上传文件

将httpd-rpms.tar.gz 、mirror.tgz、epel.tgz放置到/data
将usdp-01-master-privatization-free-2.X.tar.gz放置到/opt/usdp-srv/

[root@zhiyong1 usdp-srv]# ll
总用量 7409020
-rw-r--r--. 1 root root 7586834393 3月   1 21:38 usdp-01-master-privatization-free-2.X.tar.gz
[root@zhiyong1 usdp-srv]# tar -zxvf usdp-01-master-privatization-free-2.X.tar.gz

其中的一些安装包：

usdp/repository/2.0.0.0/influxdb-1.8.0.tar.gz
usdp/repository/2.0.0.0/usdpmonitor-1.0.0.tar.gz
usdp/repository/2.0.0.0/alertmanager-0.21.0.tar.gz
usdp/repository/2.0.0.0/prometheus-2.18.1.tar.gz
usdp/repository/2.0.0.0/grafana-6.5.1.tar.gz
usdp/repository/2.0.0.0/node_exporter-1.0.0.tar.gz
usdp/repository/2.0.0.0/phoenix-5.0.0.tar.gz
usdp/repository/2.0.0.0/tez-0.10.0.tar.gz
usdp/repository/2.0.0.0/spark-3.0.1.tar.gz
usdp/repository/2.0.0.0/zkui-2.0.0.tar.gz
usdp/repository/2.0.0.0/elasticsearch-7.8.0.tar.gz
usdp/repository/2.0.0.0/zookeeper-3.4.13.tar.gz
usdp/repository/2.0.0.0/zeppelin-0.9.0.tar.gz
usdp/repository/2.0.0.0/tez-0.10.0-dependency.tar.gz
usdp/repository/2.0.0.0/kibana-7.8.0.tar.gz
usdp/repository/2.0.0.0/sqoop-1.4.7.tar.gz
usdp/repository/2.0.0.0/hadoop-3.1.1.tar.gz
usdp/repository/2.0.0.0/kafka-2.11-2.0.1.tar.gz
usdp/repository/2.0.0.0/flume-1.9.0.tar.gz
usdp/repository/2.0.0.0/kafkaeagle-2.0.6.tar.gz
usdp/repository/2.0.0.0/neo4j-3.5.28.tar.gz
usdp/repository/2.0.0.0/hbase-2.1.10.tar.gz
usdp/repository/2.0.0.0/hue-4.8.0.tar.gz
usdp/repository/2.0.0.0/flink-1.13.2.tar.gz
usdp/repository/2.0.0.0/ranger-2.1.0.tar.gz
usdp/repository/2.0.0.0/hive-3.1.2.tar.gz
usdp/repository/2.0.0.0/dolphinscheduler-2.0.0.tar.gz

可以大致看出版本不算太老。还算全面，当然用得上用不上另说，我可以不用，但不希望用的时候没有。。。白piao版少几个组件，Kylin、Atlas、Livy、Alluxio之类的，只好自己慢慢手动安装。Flink也老了点，体验不到1.14流批一体的欢乐。Kerberos认证暂时也没有自动化部署的方式。

准备安装

[root@zhiyong1 usdp-srv]# cd usdp/
[root@zhiyong1 usdp]# ll
总用量 4
drwxr-xr-x. 2 root root   33 12月 29 15:36 agent
drwxr-xr-x. 2 root root  136 12月 31 14:39 bin
drwxr-xr-x. 2 root root   65 12月 29 14:48 config
drwxr-xr-x. 2 root root  137 12月 29 14:46 jmx_exporter
drwxr-xr-x. 3 root root   21 12月 29 14:48 license
drwxr-xr-x. 2 root root    6 12月 29 14:48 logs
drwxr-xr-x. 2 root root   35 12月 29 14:46 recommend
drwxr-xr-x. 7 root root   73 12月 29 14:48 repair
drwxr-xr-x. 3 root root   21 12月 29 14:46 repository
drwxr-xr-x. 2 root root 4096 12月 29 15:34 scripts
drwxr-xr-x. 2 root root   34 12月 29 15:35 server
drwxr-xr-x. 2 root root   29 12月 29 14:48 sql
drwxr-xr-x. 4 root root   32 12月 29 15:30 templated
drwxr-xr-x. 2 root root    6 12月 29 14:48 verify
drwxr-xr-x. 2 root root   79 12月 31 14:49 versions

[root@zhiyong1 usdp]# cd ./repair/
[root@zhiyong1 repair]# ll
总用量 8
drwxr-xr-x. 2 root root 4096 12月 29 15:51 bin
drwxr-xr-x. 2 root root  105 12月 29 15:43 config
drwxr-xr-x. 2 root root 4096 12月 30 10:47 packages
drwxr-xr-x. 2 root root  189 12月 29 14:48 remove
drwxr-xr-x. 2 root root   23 12月 30 10:45 sbin

[root@zhiyong1 repair]# cd ./config/
[root@zhiyong1 config]# ll
总用量 12
-rwxr-xr-x. 1 root root 723 12月 29 15:44 repair-host-info-add.properties
-rwxr-xr-x. 1 root root 507 12月 29 15:44 repair-host-info.properties
-rwxr-xr-x. 1 root root 666 12月 29 15:44 repair.properties

[root@zhiyong1 config]# vim repair.properties

修改为：

# Set the YUM source host IP
yum.repo.host.ip=192.168.88.100

#The Host information for installing the NMAP service
namp.server.ip=192.168.88.100
namp.server.port=22
namp.server.password=123456

# The Host information for installing the NTP service(Master)
ntp.master.ip=192.168.88.100

# Install MySQL machine node information
mysql.ip=192.168.88.100
mysql.host.ssh.port=22
mysql.host.ssh.password=123456

# Set the MYSQL database login password
mysql.password=123456

# The total number of machines needed to be repaired.
repair.host.num=7

# The total number of added machines needed to be repaired.
repair.add.host.num=0

# Common Settings.
repair.log.dir=./logs

修改配置;

[root@zhiyong1 config]# vim repair-host-info.properties

修改为：

# 1.Please provide the information of hosts needed to be repaired in the format specified below
# 2.usdp.ip.i(eg:i=1,2,3.....):
# 3.usdp.password.i:
# 4.usdp.ssh.port.i:
# 5.usdp.ssh.port.hostname.i:
usdp.ip.1=192.168.88.100
usdp.password.1=123456
usdp.ssh.port.1=22
usdp.ssh.port.hostname.1=zhiyong1

usdp.ip.2=192.168.88.101
usdp.password.2=123456
usdp.ssh.port.2=22
usdp.ssh.port.hostname.2=zhiyong2

usdp.ip.3=192.168.88.102
usdp.password.3=123456
usdp.ssh.port.3=22
usdp.ssh.port.hostname.3=zhiyong3

usdp.ip.4=192.168.88.103
usdp.password.4=123456
usdp.ssh.port.4=22
usdp.ssh.port.hostname.4=zhiyong4

usdp.ip.5=192.168.88.104
usdp.password.5=123456
usdp.ssh.port.5=22
usdp.ssh.port.hostname.5=zhiyong5

usdp.ip.6=192.168.88.105
usdp.password.6=123456
usdp.ssh.port.6=22
usdp.ssh.port.hostname.6=zhiyong6

usdp.ip.7=192.168.88.106
usdp.password.7=123456
usdp.ssh.port.7=22
usdp.ssh.port.hostname.7=zhiyong7

执行修复脚本

[root@zhiyong1 config]# cd /opt/usdp-srv/usdp/repair/sbin/
[root@zhiyong1 sbin]# ll
总用量 116
-rwxr-xr-x. 1 root root 117392 12月 29 15:44 repair.sh
[root@zhiyong1 sbin]# bash repair.sh initAll

等到出现：

SUCCESS: All encryption-free login links have been repaired successfully
SUCCESS: All closing firewall links have been repaired successfully
SUCCESS: All closing swap links have been repaired successfully
SUCCESS: set hostname links have been repaired successfully
SUCCESS: Set  ntp  have been repaired successfully
SUCCESS: Set  libxslt devel  have been repaired successfully
SUCCESS: Set  psmisc  have been repaired successfully
SUCCESS: Set mysql-client links have been repaired successfully
SUCCESS: Set  mysql-python  have been repaired successfully
SUCCESS: All transparent_hugepage links have been repaired successfully
SUCCESS: Set JDK links have been repaired successfully
SUCCESS: Set xdg-utils links have been repaired successfully
SUCCESS: Set redhat-lsb links have been repaired successfully
SUCCESS: Set python-devel links have been repaired successfully
SUCCESS: Set cyrus-sasl links have been repaired successfully
SUCCESS: Set python36-devel links have been repaired successfully
SUCCESS: Set gcc-c++ links have been repaired successfully
SUCCESS: Set  Cython links have been repaired successfully
SUCCESS: Set  Six links have been repaired successfully
SUCCESS: Set  websocket-client links have been repaired successfully
SUCCESS: Set  ecdsa links have been repaired successfully
SUCCESS: Set  pytest-runner links have been repaired successfully
SUCCESS: Set krb5-devel links have been repaired successfully
The USDP deployment environment of all nodes has been repaired successfully . Please proceed to the next step

说明脚本执行成功。

刷新环境变量

在7台机器都手动执行：

[root@zhiyong1 sbin]# source /etc/profile

配置MySQL

[root@zhiyong1 sbin]# vim /opt/usdp-srv/usdp/config/application-server.yml

修改这一段：

datasource:
    type: com.zaxxer.hikari.HikariDataSource
    #    driver-class-name: org.gjt.mm.mysql.Driver
    driver-class-name: com.p6spy.engine.spy.P6SpyDriver
    url: jdbc:p6spy:mysql://192.168.88.100:3306/db_udp?useUnicode=true&characterEncoding=utf-8&useSSL=false
    username: root
    password: 123456

启动集群

[root@zhiyong1 sbin]# cd /opt/usdp-srv/usdp/
[root@zhiyong1 usdp]# bin/start-udp-server.sh
BASE_PATH: /opt/usdp-srv/usdp/bin
JMX_PATH: /opt/usdp-srv/usdp/jmx_exporter
ln -s /opt/usdp-srv /data/usdp-srv
ln -s /opt/usdp-srv/srv/udp /srv/
ln -s /data/var/log/udp /var/log/
REPAIR_PATH: /opt/usdp-srv/usdp/repair
UDP_PATH: /opt/usdp-srv/usdp
REPAIR_BIN_PATH: /opt/usdp-srv/usdp/repair/bin
REPAIR_SBIN_PATH: /opt/usdp-srv/usdp/repair/sbin
PACKAGES_PATH: /opt/usdp-srv/usdp/repair/packages
nmap-6.40-19.el7.x86_64
nmap exists

UDP Server is running with: 52552
Done.

集群启动成功。此时Server的可用内存：

[root@zhiyong1 usdp]# free -m
              total        used        free      shared  buff/cache   available
Mem:          31994        2631         216          13       29146       28813
Swap:             0           0           0

Worker的可用内存：

[root@zhiyong2 packages]# free -m
              total        used        free      shared  buff/cache   available
Mem:          15867         243        8921          11        6701       15175
Swap:             0           0           0

打开Web UI

浏览器：

http://192.168.88.100/login

即可创建用户。都写Zhiyong1。

软件版本

安装组件

点下一步安装组件：

确认并联通节点：

检查环境是否正确：

选择推荐方案A做全量配置：

可以看到：





智能推荐：

有一些默认配置：


下一步等待安装完成：

都绿了就可以完成：

可以看到集群运行状态暂时良好【用不了多久就会JVM堆栈内存溢出导致组件宕掉】：

【建议：16G是能体验的最低配置，不搞HA，只启动最基础的HDFS、HBASE、KAFKA等服务的备用集群可以这么搞，主集群32G内存是组件全开不宕的最低要求】

默认的密码

服务名称	账号	密码
Grafana	hadoop	hadoop
Hue	hadoop	hadoop
Zeppelin	hadoop	hadoop
Airflow	hadoop	hadoop
ZKUI	hadoop	hadoop
KafkaEagle	admin	123456
Ranger	admin	admin
DolphinSchedule	admin	hadoop

默认的端口

服务名称	端口号
Flink	8082
Hive	10002
Spark	4000
Tez	9999
Yarn	8088、19888、8188
Hbase	16010
HDFS	50070
AlertManager	9093
Grafana	3000
Prometheus	9090
Hue	8888
KafkaEagle	8048
Kibana	5601
Zepplin	28080
Zkui	9099
DolphinScheduler	12345
Ranger	6080

部署说明

USDP官网部署说明

测试

[root@zhiyong2 bin]# pwd
/opt/usdp-srv/srv/udp/2.0.0.0/hdfs/bin
[root@zhiyong2 bin]# ./hadoop fs -ls /zhiyong-1/user/hive
Found 1 items
drwxrwxrwx   - hadoop supergroup          0 2022-03-01 23:08 /zhiyong-1/user/hive/warehouse

可以访问到数据。

再添加一套集群

在上方可以找到切换集群的选项，然后和之前的操作一致。部署后：


当前资源占用：

不推荐128G以下内存的机器搭建Vmware双集群，撑死一个就够了。

笔者的虚拟机全部放在SSD以加快运行速度。

最后可以打快照，压缩存档。安装一次需要40分钟+。

内存溢出

组件全开时，用不了多久就会发生JVM堆内存溢出。笔者之后将zhiyong2->zhiyong4的内存提升到32G，zhiyong5->zhiyong7则只开启必要的Zookeeper、HDFS、HBase、Kafka做备用集群。稳定性有了少许提高。这2套集群还是很吃内存的。

致谢

感谢UCloud如此大方地让我等穷人可以白piao大数据集群管理工具USDP。对于新手而言，大数据组件的安装相当友好。安装部署相较CDH、CDP与HDP容易太多了。

新版本已经取消了授权操作，更是简化了操作步骤。

对于大数据开发攻城狮而言，也可以抽出更多精力放在Java/Scala/SQL开发而不是组件兼容性适配、源码编译这些琐碎的小事上。