如何将Hadoop部署在低廉的硬件上

一、概述

Hadoop 是 Apache 软件基金会下一个开源分布式计算平台，以 HDFS（Hadoop Distributed File System）、MapReduce（Hadoop2.0 加入了 YARN，Yarn 是资源调度框架，能够细粒度的管理和调度任务，还能够支持其他的计算框架，比如 spark）为核心的 Hadoop 为用户提供了系统底层细节透明的分布式基础架构。hdfs 的高容错性、高伸缩性、高效性等优点让用户可以将 Hadoop 部署在低廉的硬件上，形成分布式系统。

HDFS YARN

二、开始部署

1）添加源

地址：

https://artifacthub.io/packages/helm/apache-hadoop-helm/hadoop

helmrepoaddapache-hadoop-helmhttps://pfisterer.github.io/apache-hadoop-helm/
helmpullapache-hadoop-helm/hadoop--version1.2.0
tar-xfhadoop-1.2.0.tgz

2）构建镜像 Dockerfile

FROMmyharbor.com/bigdata/centos:7.9.2009

RUNrm-f/etc/localtime&&ln-sv/usr/share/zoneinfo/Asia/Shanghai/etc/localtime&&echo"Asia/Shanghai">/etc/timezone

RUNexportLANG=zh_CN.UTF-8

#创建用户和用户组，跟yaml编排里的spec.template.spec.containers.securityContext.runAsUser:9999
RUNgroupadd--system--gid=9999admin&&useradd--system--home-dir/home/admin--uid=9999--gid=adminadmin

#安装sudo
RUNyum-yinstallsudo;chmod640/etc/sudoers

#给admin添加sudo权限
RUNecho"adminALL=(ALL)NOPASSWD:ALL">>/etc/sudoers

RUNyum-yinstallinstallnet-toolstelnetwget

RUNmkdir/opt/apache/

ADDjdk-8u212-linux-x64.tar.gz/opt/apache/

ENVJAVA_HOME=/opt/apache/jdk1.8.0_212
ENVPATH=$JAVA_HOME/bin:$PATH

ENVHADOOP_VERSION3.3.2
ENVHADOOP_HOME=/opt/apache/hadoop

ENVHADOOP_COMMON_HOME=${HADOOP_HOME}
HADOOP_HDFS_HOME=${HADOOP_HOME}
HADOOP_MAPRED_HOME=${HADOOP_HOME}
HADOOP_YARN_HOME=${HADOOP_HOME}
HADOOP_CONF_DIR=${HADOOP_HOME}/etc/hadoop
PATH=${PATH}:${HADOOP_HOME}/bin

#RUNcurl--silent--output/tmp/hadoop.tgzhttps://ftp-stud.hs-esslingen.de/pub/Mirrors/ftp.apache.org/dist/hadoop/common/hadoop-${HADOOP_VERSION}/hadoop-${HADOOP_VERSION}.tar.gz&&tar--directory/opt/apache-xzf/tmp/hadoop.tgz&&rm/tmp/hadoop.tgz
ADDhadoop-${HADOOP_VERSION}.tar.gz/opt/apache
RUNln-s/opt/apache/hadoop-${HADOOP_VERSION}${HADOOP_HOME}

RUNchown-Radmin:admin/opt/apache

WORKDIR$HADOOP_HOME

#Hdfsports
EXPOSE500105002050070500755009080209000

#Mapredports
EXPOSE19888

#Yarnports
EXPOSE8030803180328033804080428088

#Otherports
EXPOSE497072122

开始构建镜像

dockerbuild-tmyharbor.com/bigdata/hadoop:3.3.2.--no-cache

###参数解释
#-t：指定镜像名称
# . ：当前目录Dockerfile
#-f：指定Dockerfile路径
#--no-cache：不缓存

推送到镜像仓库

dockerpushmyharbor.com/bigdata/hadoop:3.3.2

调整目录结构

mkdirhadoop/templates/hdfshadoop/templates/yarn
mvhadoop/templates/hdfs-*hadoop/templates/hdfs/
mvhadoop/templates/yarn-*hadoop/templates/yarn/

2）修改配置

hadoop/values.yaml

image:
repository:myharbor.com/bigdata/hadoop
tag:3.3.2
pullPolicy:IfNotPresent

...

persistence:
nameNode:
enabled:true
storageClass:"hadoop-nn-local-storage"
accessMode:ReadWriteOnce
size:10Gi
local:
-name:hadoop-nn-0
host:"local-168-182-110"
path:"/opt/bigdata/servers/hadoop/nn/data/data1"

dataNode:
enabled:true
storageClass:"hadoop-dn-local-storage"
accessMode:ReadWriteOnce
size:20Gi
local:
-name:hadoop-dn-0
host:"local-168-182-110"
path:"/opt/bigdata/servers/hadoop/dn/data/data1"
-name:hadoop-dn-1
host:"local-168-182-110"
path:"/opt/bigdata/servers/hadoop/dn/data/data2"
-name:hadoop-dn-2
host:"local-168-182-110"
path:"/opt/bigdata/servers/hadoop/dn/data/data3"
-name:hadoop-dn-3
host:"local-168-182-111"
path:"/opt/bigdata/servers/hadoop/dn/data/data1"
-name:hadoop-dn-4
host:"local-168-182-111"
path:"/opt/bigdata/servers/hadoop/dn/data/data2"
-name:hadoop-dn-5
host:"local-168-182-111"
path:"/opt/bigdata/servers/hadoop/dn/data/data3"
-name:hadoop-dn-6
host:"local-168-182-112"
path:"/opt/bigdata/servers/hadoop/dn/data/data1"
-name:hadoop-dn-7
host:"local-168-182-112"
path:"/opt/bigdata/servers/hadoop/dn/data/data2"
-name:hadoop-dn-8
host:"local-168-182-112"
path:"/opt/bigdata/servers/hadoop/dn/data/data3"

...

service:
nameNode:
type:NodePort
ports:
dfs:9000
webhdfs:9870
nodePorts:
dfs:30900
webhdfs:30870
dataNode:
type:NodePort
ports:
dfs:9000
webhdfs:9864
nodePorts:
dfs:30901
webhdfs:30864
resourceManager:
type:NodePort
ports:
web:8088
nodePorts:
web:30088
...

securityContext:
runAsUser:9999
privileged:true

hadoop/templates/hdfs/hdfs-nn-pv.yaml

{{-range.Values.persistence.nameNode.local}}
---
apiVersion:v1
kind:PersistentVolume
metadata:
name:{{.name}}
labels:
name:{{.name}}
spec:
storageClassName:{{$.Values.persistence.nameNode.storageClass}}
capacity:
storage:{{$.Values.persistence.nameNode.size}}
accessModes:
-ReadWriteOnce
local:
path:{{.path}}
nodeAffinity:
required:
nodeSelectorTerms:
-matchExpressions:
-key:kubernetes.io/hostname
operator:In
values:
-{{.host}}
---
{{-end}}

hadoop/templates/hdfs/hdfs-dn-pv.yaml

{{-range.Values.persistence.dataNode.local}}
---
apiVersion:v1
kind:PersistentVolume
metadata:
name:{{.name}}
labels:
name:{{.name}}
spec:
storageClassName:{{$.Values.persistence.dataNode.storageClass}}
capacity:
storage:{{$.Values.persistence.dataNode.size}}
accessModes:
-ReadWriteOnce
local:
path:{{.path}}
nodeAffinity:
required:
nodeSelectorTerms:
-matchExpressions:
-key:kubernetes.io/hostname
operator:In
values:
-{{.host}}
---
{{-end}}

修改 hdfs service

mvhadoop/templates/hdfs/hdfs-nn-svc.yamlhadoop/templates/hdfs/hdfs-nn-svc-headless.yaml
mvhadoop/templates/hdfs/hdfs-dn-svc.yamlhadoop/templates/hdfs/hdfs-dn-svc-headless.yaml
#注意修改名称，不要重复

hadoop/templates/hdfs/hdfs-nn-svc.yaml

#AheadlessservicetocreateDNSrecords
apiVersion:v1
kind:Service
metadata:
name:{{include"hadoop.fullname".}}-hdfs-nn
labels:
app.kubernetes.io/name:{{include"hadoop.name".}}
helm.sh/chart:{{include"hadoop.chart".}}
app.kubernetes.io/instance:{{.Release.Name}}
app.kubernetes.io/component:hdfs-nn
spec:
ports:
-name:dfs
port:{{.Values.service.nameNode.ports.dfs}}
protocol:TCP
nodePort:{{.Values.service.nameNode.nodePorts.dfs}}
-name:webhdfs
port:{{.Values.service.nameNode.ports.webhdfs}}
nodePort:{{.Values.service.nameNode.nodePorts.webhdfs}}
type:{{.Values.service.nameNode.type}}
selector:
app.kubernetes.io/name:{{include"hadoop.name".}}
app.kubernetes.io/instance:{{.Release.Name}}
app.kubernetes.io/component:hdfs-nn

hadoop/templates/hdfs/hdfs-dn-svc.yaml

#AheadlessservicetocreateDNSrecords
apiVersion:v1
kind:Service
metadata:
name:{{include"hadoop.fullname".}}-hdfs-dn
labels:
app.kubernetes.io/name:{{include"hadoop.name".}}
helm.sh/chart:{{include"hadoop.chart".}}
app.kubernetes.io/instance:{{.Release.Name}}
app.kubernetes.io/component:hdfs-nn
spec:
ports:
-name:dfs
port:{{.Values.service.dataNode.ports.dfs}}
protocol:TCP
nodePort:{{.Values.service.dataNode.nodePorts.dfs}}
-name:webhdfs
port:{{.Values.service.dataNode.ports.webhdfs}}
nodePort:{{.Values.service.dataNode.nodePorts.webhdfs}}
type:{{.Values.service.dataNode.type}}
selector:
app.kubernetes.io/name:{{include"hadoop.name".}}
app.kubernetes.io/instance:{{.Release.Name}}
app.kubernetes.io/component:hdfs-dn

修改 yarn service

mvhadoop/templates/yarn/yarn-nm-svc.yamlhadoop/templates/yarn/yarn-nm-svc-headless.yaml
mvhadoop/templates/yarn/yarn-rm-svc.yamlhadoop/templates/yarn/yarn-rm-svc-headless.yaml
mvhadoop/templates/yarn/yarn-ui-svc.yamlhadoop/templates/yarn/yarn-rm-svc.yaml
#注意修改名称，不要重复

hadoop/templates/yarn/yarn-rm-svc.yaml

#Servicetoaccesstheyarnwebui
apiVersion:v1
kind:Service
metadata:
name:{{include"hadoop.fullname".}}-yarn-rm
labels:
app.kubernetes.io/name:{{include"hadoop.name".}}
helm.sh/chart:{{include"hadoop.chart".}}
app.kubernetes.io/instance:{{.Release.Name}}
app.kubernetes.io/component:yarn-rm
spec:
ports:
-port:{{.Values.service.resourceManager.ports.web}}
name:web
nodePort:{{.Values.service.resourceManager.nodePorts.web}}
type:{{.Values.service.resourceManager.type}}
selector:
app.kubernetes.io/name:{{include"hadoop.name".}}
app.kubernetes.io/instance:{{.Release.Name}}
app.kubernetes.io/component:yarn-rm

修改控制器

在所有控制中新增如下内容：

containers:
...
securityContext:
runAsUser:{{.Values.securityContext.runAsUser}}
privileged:{{.Values.securityContext.privileged}}

hadoop/templates/hadoop-configmap.yaml

###1、将/root换成/opt/apache
###2、TMP_URL="http://{{include"hadoop.fullname".}}-yarn-rm-headless:8088/ws/v1/cluster/info"

3）开始安装

#创建存储目录
mkdir-p/opt/bigdata/servers/hadoop/{nn,dn}/data/data{1..3}

helminstallhadoop./hadoop-nhadoop--create-namespace

NOTES

NAME:hadoop
LASTDEPLOYED:SatSep2417552022
NAMESPACE:hadoop
STATUS:deployed
REVISION:1
TESTSUITE:None
NOTES:
1.YoucancheckthestatusofHDFSbyrunningthiscommand:
kubectlexec-nhadoop-ithadoop-hadoop-hdfs-nn-0--/opt/hadoop/bin/hdfsdfsadmin-report

2.Youcanlisttheyarnnodesbyrunningthiscommand:
kubectlexec-nhadoop-ithadoop-hadoop-yarn-rm-0--/opt/hadoop/bin/yarnnode-list

3.Createaport-forwardtotheyarnresourcemanagerUI:
kubectlport-forward-nhadoophadoop-hadoop-yarn-rm-08088:8088

Thenopentheuiinyourbrowser:

openhttp://localhost:8088

4.Youcanrunincludedhadooptestslikethis:
kubectlexec-nhadoop-ithadoop-hadoop-yarn-nm-0--/opt/hadoop/bin/hadoopjar/opt/hadoop/share/hadoop/mapreduce/hadoop-mapreduce-client-jobclient-3.3.2-tests.jarTestDFSIO-write-nrFiles5-fileSize128MB-resFile/tmp/TestDFSIOwrite.txt

5.Youcanlistthemapreducejobslikethis:
kubectlexec-nhadoop-ithadoop-hadoop-yarn-rm-0--/opt/hadoop/bin/mapredjob-list

6.Thischartcanalsobeusedwiththezeppelinchart
helminstall--namespacehadoop--sethadoop.useConfigMap=true,hadoop.configMapName=hadoop-hadoopstable/zeppelin

7.Youcanscalethenumberofyarnnodeslikethis:
helmupgradehadoop--setyarn.nodeManager.replicas=4stable/hadoop

Makesuretoupdatethevalues.yamlifyouwanttomakethispermanent.

查看

kubectlgetpods,svc-nhadoop-owide

hdfs web：

http://192.168.182.110:30870/

yarn web：

http://192.168.182.110:30088/

5）测试验证

HDFS 测试验证

kubectlexec-ithadoop-hadoop-hdfs-nn-0-nhadoop--bash
[root@local-168-182-110hadoop]#kubectlexec-ithadoop-hadoop-hdfs-nn-0-nhadoop--bash
bash-4.2$
bash-4.2$
bash-4.2$hdfsdfs-mkdir/tmp
bash-4.2$hdfsdfs-ls/
Found1items
drwxr-xr-x-adminsupergroup02022-09-2417:56/tmp
bash-4.2$echo"testhadoop">test.txt
bash-4.2$hdfsdfs-puttest.txt/tmp/
bash-4.2$hdfsdfs-ls/tmp/
Found1items
-rw-r--r--3adminsupergroup122022-09-2417:57/tmp/test.txt
bash-4.2$hdfsdfs-cat/tmp/
cat:`/tmp':Isadirectory
bash-4.2$hdfsdfs-cat/tmp/test.txt
testhadoop
bash-4.2$

Yarn 的测试验证等后面讲到 hive on k8s 再来测试验证。

6）卸载

helmuninstallhadoop-nhadoop

kubectldeletepod-nhadoop`kubectlgetpod-nhadoop|awk'NR>1{print$1}'`--force
kubectlpatchnshadoop-p'{"metadata":{"finalizers":null}}'
kubectldeletenshadoop--force

这里也提供 git 下载地址，有需要的小伙伴可以下载部署玩玩：

https://gitee.com/hadoop-bigdata/hadoop-on-k8s

在 k8s 集群中 yarn 会慢慢被弱化，直接使用 k8s 资源调度，而不再使用 yarn 去调度资源了，这里只是部署了单点，仅限于测试环境使用，下一篇文章会讲 Hadoop 高可用 on k8s 实现，请小伙伴耐心等待，有任何疑问欢迎给我留言~