上一份工作主要负责大数据平台的建设,在这个过程中积累了一些Hadoop生态组件的搭建和使用笔记,由于时间关系,不打算去修改其中的错别字和排版问题,直接释出原始笔记。
前置条件
我所在的集群有三台服务其,对应的host分别为master,slave1,slave2。hadoop服务的安装分部为
机器host | 组件情况 |
---|---|
master | namenode、datanode、journalnode、resourcemanager、nodemanager、jobhistoryserver |
slave1 | namenode、datanode、journalnode、resourcemanager、nodemanager |
slave2 | datanode、journalnode、nodemanager |
kerberos相关
首先我们要安装好kerberos,kerberos的安装搭建参考链接
https://www.cnblogs.com/niceshot/p/13216455.html
给hadoop各组件创建kerberos账号
进入kerberos的admin.local后,依次执行下述命令
//组件web服务的princial
addprinc -randkey HTTP/master@TEST.COM
addprinc -randkey HTTP/slave1@TEST.COM
addprinc -randkey HTTP/slave2@TEST.COM
//namenode的princial
addprinc -randkey nn/master@TEST.COM
addprinc -randkey nn/slave1@TEST.COM
//datanode的princial
addprinc -randkey dn/master@TEST.COM
addprinc -randkey dn/slave1@TEST.COM
addprinc -randkey dn/slave2@TEST.COM
//journalnode的princial
addprinc -randkey jn/master@TEST.COM
addprinc -randkey jn/slave1@TEST.COM
addprinc -randkey jn/slave2@TEST.COM
//resourcemanager 的princial
addprinc -randkey rm/master@TEST.COM
addprinc -randkey rm/slave1@TEST.COM
//nodemanager的principal
addprinc -randkey nm/master@TEST.COM
addprinc -randkey nm/slave1@TEST.COM
addprinc -randkey nm/slave2@TEST.COM
//job hisotry server的princial
addprinc -randkey jhs/master@TEST.COM
将这些账号做成keytab
同样是在admin.local中,将上述账号认证信息做成keytab
ktadd -k /opt/keytab_store/http.service.keytab HTTP/master@TEST.COM
ktadd -k /opt/keytab_store/http.service.keytab HTTP/slave1@TEST.COM
ktadd -k /opt/keytab_store/http.service.keytab HTTP/slave2@TEST.COM
ktadd -k /opt/keytab_store/nn.service.keytab nn/master@TEST.COM
ktadd -k /opt/keytab_store/nn.service.keytab nn/slave1@TEST.COM
ktadd -k /opt/keytab_store/dn.service.keytab dn/master@TEST.COM
ktadd -k /opt/keytab_store/dn.service.keytab dn/slave1@TEST.COM
ktadd -k /opt/keytab_store/dn.service.keytab dn/slave2@TEST.COM
ktadd -k /opt/keytab_store/jn.service.keytab jn/master@TEST.COM
ktadd -k /opt/keytab_store/jn.service.keytab jn/slave1@TEST.COM
ktadd -k /opt/keytab_store/jn.service.keytab jn/slave2@TEST.COM
ktadd -k /opt/keytab_store/rm.service.keytab rm/master@TEST.COM
ktadd -k /opt/keytab_store/rm.service.keytab rm/slave1@TEST.COM
ktadd -k /opt/keytab_store/nm.service.keytab nm/master@TEST.COM
ktadd -k /opt/keytab_store/nm.service.keytab nm/slave1@TEST.COM
ktadd -k /opt/keytab_store/nm.service.keytab nm/slave2@TEST.COM
ktadd -k /opt/keytab_store/jhs.service.keytab jhs/master@TEST.COM
多个账号可以做到一个keytab中去,上述的命令做了多个文件,不同组件角色的单独放到了一个keytab文件中。其实内部网络,可以把所有的hadoop相关组件做成一个大的keytab文件,降低配置复杂性。
将上述的keytab文件,分发到集群所有机器
core-site.xml
关键配置
<property>
<name>hadoop.security.authentication</name>
<value>kerberos</value>
</property>
<property>
<name>hadoop.security.authorization</name>
<value>true</value>
</property>
<property>
<name>hadoop.security.auth_to_local</name>
<value>
RULE:[2:$1/$2@$0]([ndj]n/.*@TEST.COM)s/.*/hdfs/
RULE:[2:$1/$2@$0]([rn]m/.*@TEST.COM)s/.*/yarn/
RULE:[2:$1/$2@$0](jhs/.*@TEST.COM)s/.*/mapred/
DEFAULT
</value>
</property>
上述配置的意思是 在整个集群中费用kerberos作为安全认证和授权,
hadoop.security.auth_to_local 配置组件之间互访时被访问的服务,如何从访问的Principal中抽取出实际的用户。大致规则以第一行为例,表示将namenode, 和datanode ,journalnode的principal 映射成为hdfs的user
而最终的default是上述规则都不匹配时的默认规则,默认规则会直接从principal中提取第一个斜杠前面的信息作为user。比如test/xxhost@DOMIAN.COM 会被识别成明为test的user
HDFS
<property>
<name>dfs.block.access.token.enable</name>
<value>true</value>
</property>
<property>
<name>dfs.namenode.kerberos.principal</name>
<value>nn/_HOST@TEST.COM</value>
</property>
<property>
<name>dfs.namenode.keytab.file</name>
<value>/opt/keytab_store/nn.service.keytab</value>
</property>
<property>
<name>dfs.namenode.kerberos.internal.spnego.principal</name>
<value>${dfs.web.authentication.kerberos.principal}</value>
</property>
<property>
<name>dfs.journalnode.kerberos.principal</name>
<value>jn/_HOST@TEST.COM</value>
</property>
<property>
<name>dfs.journalnode.keytab.file</name>
<value>/opt/keytab_store/jn.service.keytab</value>
</property>
<property>
<name>dfs.journalnode.kerberos.internal.spnego.principal</name>
<value>${dfs.web.authentication.kerberos.principal}</value>
</property>
<property>
<name>dfs.datanode.kerberos.principal</name>
<value>dn/_HOST@TEST.COM</value>
</property>
<property>
<name>dfs.datanode.keytab.file</name>
<value>/opt/keytab_store/dn.service.keytab</value>
</property>
<property>
<name>dfs.web.authentication.kerberos.principal</name>
<value>HTTP/_HOST@TEST.COM</value>
</property>
<property>
<name>dfs.web.authentication.kerberos.keytab</name>
<value>/opt/keytab_store/http.service.keytab</value>
</property>
<property>
<name>dfs.http.policy</name>
<value>HTTPS_ONLY</value>
</property>
<property>
<name>dfs.data.transfer.protection</name>
<value>authentication</value>
</property>
其中大体配置是配置各组件使用的principal是什么。其中的_HOST相当于语法糖,hadoop会根据本机hostname,替换该配置,从而实现不同机器相同配置文件的目的
datanode的安全配置
由于datanode数据传输走的不是rpc,而是http。所以datanode无法使用kerberos的方式进行认证。为了解决这个问题,有两种方式的配置,来实现datanode数据传输的安全性
- JSVC
- TLS/SSL
JSVC方式的大体原理是使用JSVC工具,让datanode能够使用特权端口启动,所谓特权端口是指1024以下的端口,这种安全配置假定攻击者无法获取root权限,所以也就无法操作datanode来实现。hadoop 2.6.0以前,只能使用这种方式,配置较为复杂,不在这里赘述。hadoop 2.6.0以后引入了SASL方式,通过TLS/SSL来实现数据的安全传输,下面介绍这种方式
证书生成和安装
TLS/SSL相关原理见文档 ,这里粘贴地址
首先保证机器上已经安装好了openssl。下面是详细的配置。核心思想是,做一个私有的CA,然后通过这个私有的CA证书给所有的其它证书签名,通过将私有CA的证书安装到各机器的信任区里,实现一个各机器间的TLS/SSL通信
然后在集群中随便找一台机器,先生成CA证书,这里在Master这台机器上操作
openssl req -new -x509 -keyout ca_private.key -out ca_cert -days 9999 -subj '/C=CN/ST=chengdu/L=chengdu/O=bigdata/OU=bigdata/CN=master'
将上述的CA私钥跟更要证书拷贝到各个机器。然后再各机器上做如下操作,当然如果我们在生成证书时,用的密码完全一样也可以在一个机器上做,最后把相关的keystore和truststore分发到所有的机器。
//生成自己的公私秘钥对
keytool -keystore keystore -alias localhost -validity 9999 -genkey -keyalg RSA -keysize 2048 -dname "CN=slave2, OU=bigdata, O=bigdata, L=chengdu, ST=chengdu, C=CN"
//将上述的CA公钥证书导入本机的信任区truststore
keytool -keystore truststore -alias CARoot -import -file ca_cert
//将上述的CA公钥导入本机的keystore中
keytool -keystore keystore -alias CARoot -import -file ca_cert
//将本机的公钥证书导出
keytool -certreq -alias localhost -keystore keystore -file local_cert
//对CA私钥,对本机的公钥证书进行签名
openssl x509 -req -CA hd_ca_cert -CAkey ca_private.key -in local_cert -out local_cert_signed -days 9999 -CAcreateserial
//将签名后的证书导入的自己的Keystore
keytool -keystore keystore -alias localhost -import -file local_cert_signed
hdfs-site.xml的重点配置
配置dfs.http.policy
的value为HTTPS_ONLY
配置dfs.data.transfer.protection
的value为authentication
、 integrity
、 privacy
任意一种。一般内部集群用authentication即可
- authentication ,只认证签名
- integrity 除了认证签名外,还验证数据是否被篡改
- privacy,数据除了上述的认证和完整性验证之外还要加密传输
ssl-client.xml 和 ssl-server.xml配置
hadoop在在跟core-site.xml同级目录下一般有ssl-client.xml.example和ssl-server.xml.example两个模板文件,我们可以直接去掉template来后作为配置文件来配置。他们是用来配置当前组件作为服务端时,自己的证书kestore位置,和作为客户端时,自己的信任证书truststore位置
ssl-client.xml配置如下
<configuration>
<property>
<name>ssl.client.truststore.location</name>
<value>/opt/ssl_store/truststore</value>
<description>Truststore to be used by clients like distcp. Must be
specified.
</description>
</property>
<property>
<name>ssl.client.truststore.password</name>
<value>123456</value>
<description>Optional. Default value is "".
</description>
</property>
<property>
<name>ssl.client.truststore.type</name>
<value>jks</value>
<description>Optional. The keystore file format, default value is "jks".
</description>
</property>
<property>
<name>ssl.client.truststore.reload.interval</name>
<value>10000</value>
<description>Truststore reload check interval, in milliseconds.
Default value is 10000 (10 seconds).
</description>
</property>
<property>
<name>ssl.client.keystore.location</name>
<value>/opt/ssl_store/keystore</value>
<description>Keystore to be used by clients like distcp. Must be
specified.
</description>
</property>
<property>
<name>ssl.client.keystore.password</name>
<value>123456</value>
<description>Optional. Default value is "".
</description>
</property>
<property>
<name>ssl.client.keystore.keypassword</name>
<value>123456</value>
<description>Optional. Default value is "".
</description>
</property>
<property>
<name>ssl.client.keystore.type</name>
<value>jks</value>
<description>Optional. The keystore file format, default value is "jks".
</description>
</property>
</configuration>
ssl-server.xml
<property>
<name>ssl.server.keystore.password</name>
<value>123456</value>
<description>Must be specified.
</description>
</property>
<property>
<name>ssl.server.keystore.keypassword</name>
<value>123456</value>
<description>Must be specified.
</description>
</property>
<property>
<name>ssl.server.keystore.type</name>
<value>jks</value>
<description>Optional. The keystore file format, default value is "jks".
</description>
</property>
<property>
<name>ssl.server.exclude.cipher.list</name>
<value>TLS_ECDHE_RSA_WITH_RC4_128_SHA,SSL_DHE_RSA_EXPORT_WITH_DES40_CBC_SHA,
SSL_RSA_WITH_DES_CBC_SHA,SSL_DHE_RSA_WITH_DES_CBC_SHA,
SSL_RSA_EXPORT_WITH_RC4_40_MD5,SSL_RSA_EXPORT_WITH_DES40_CBC_SHA,
SSL_RSA_WITH_RC4_128_MD5</value>
<description>Optional. The weak security cipher suites that you want excluded
from SSL communication.</description>
</property>
</configuration>
上述配置的123456是我们在做证书时使用的密码
yarn
整体配置
<property>
<name>yarn.resourcemanager.principal</name>
<value>rm/_HOST@TEST.COM</value>
</property>
<property>
<name>yarn.resourcemanager.keytab</name>
<value>/opt/keytab_store/rm.service.keytab</value>
</property>
<property>
<name>yarn.nodemanager.principal</name>
<value>nm/_HOST@TEST.COM</value>
</property>
<property>
<name>yarn.nodemanager.keytab</name>
<value>/opt/keytab_store/nm.service.keytab</value>
</property>
<property>
<!--安全集群必须使用下面的LinuxContainerExecutor-->
<name>yarn.nodemanager.container-executor.class</name>
<value>org.apache.hadoop.yarn.server.nodemanager.LinuxContainerExecutor</value>
</property>
<property>
<name>yarn.nodemanager.linux-container-executor.group</name>
<value>hadoop</value>
</property>
<property>
<name>yarn.nodemanager.linux-container-executor.path</name>
<value>/opt/hadoop-3.1.3/bin/container-executor</value>
</property>
container-executor
build LinuxContainerExecutor
上述yarn.nodemanager.linux-container-executor.path指定了LinuxContainerExecutor对应的可执行文件container-executor的路径。
hadoop发行包在bin路径下,一般就已经有这个文件了。
这个文件执行需要一个配置,container-executor.cfg 。其默认加载的是$HADOOP_HOME/etc/hadoop/container-executor.cfg这个路径的配置文件。
但由于这个路径本身又有hadoop的其它配置文件,而container-executor又要求container-executor.cfg所在路径所有层级权限都只能root访问。这会导致我们其其它组件启动出现各种奇奇古怪的问题。
所以我们需要另外指定container-executor.cfg文件的位置。但问题是container-executor这个二进制文件在构建时,已经写死了文件路径。如果我们需要重指定配置文件路径,需要重新打包container-executor。构建步骤为
- 首先下载同版本的hadoop源码
- 进入到源码包的路径
hadoop-yarn-project/hadoop-yarn/hadoop-yarn-server/hadoop-yarn-server-nodemanager
- 使用命令
mvn package -DskipTests=true -Dcontainer-executor.conf.dir=/etc/hadoop/
构建,container-executor.conf.dir参数即指定新的container-executor.cfg文件路径 - 构建完成后,在构建路径下的
target/native/target/usr/local/bin
路径即可找到新构建的container-executor,将其拷贝到$HADOOP_HOME/bin下,替换原来的程序即可
配置container-executor.cfg
在/etc/hadoop/中,创建container-executor.cfg,其配置内容如下
yarn.nodemanager.linux-container-executor.group=hadoop
banned.users=hdfs,yarn,mapred,bin
min.user.id=1000
allowed.system.users=
feature.tc.enabled=false
注意配置每行不要有空格,yarn.nodemanager.linux-container-executor.group这个配置值同yarn-site.xml中的一致
总结权限配置需要配置的项
文件权限修改
chown root:hadoop /opt/hadoop-3.1.3/bin/container-executor
chmod 6050 /opt/hadoop-3.1.3/bin/container-executor
chown root:hadoop /etc/hadoop/container-executor.cfg
chmod 400 /etc/hadoop/container-executor.cfg
假设在yarn-site.xml的中yarn.nodemanager.local-dirs 配置 路径为/home/var/data/hadoop/nodemanager/data
yarn.nodemanager.log-dirs配置路径为 /home/var/data/hadoop/nodemanager/log,还需要做以下权限配置
chown yarn:hadoop /home/var/data/hadoop/nodemanager/data
chown yarn:hadoop /home/var/data/hadoop/nodemanager/log
chmod 755 /home/var/data/hadoop/nodemanager/data
chmod 755 /home/var/data/hadoop/nodemanager/log
mapreduce
<property>
<name>mapreduce.jobhistory.keytab</name>
<value>/opt/keytab_store/jhs.service.keytab</value>
</property>
<property>
<name>mapreduce.jobhistory.principal</name>
<value>jhs/_HOST@TEST.COM</value>
</property>
启动
配置完后,按原来的方式启动即可。只是由于hdfs开起了SSL/TLS ,其原来的9870端口,变成了9871, 且需要通过https访问。比如我们这地址为:https://master:9871
参考资料
https://hadoop.apache.org/docs/stable/hadoop-project-dist/hadoop-common/SecureMode.html
https://secfree.github.io/blog/2015/07/01/sasl-data-transfer-protocol.html
https://blog.csdn.net/picway/article/details/74299086
https://developer.aliyun.com/article/245586
https://makeling.github.io/bigdata/dcb921f7.html
https://makeling.github.io/bigdata/39395030.html
http://secfree.github.io/blog/2015/06/25/yarn-container-executor-traps.html
欢迎关注我的个人公众号"西北偏北UP",记录代码人生,行业思考,科技评论