spark集群搭建整理之解决亿级人群标签问题
Posted
tags:
篇首语：本文由小常识网(cha138.com)小编为大家整理，主要介绍了spark集群搭建整理之解决亿级人群标签问题相关的知识，希望对你有一定的参考价值。
最近在做一个人群标签的项目，也就是根据客户的一些交易行为自动给客户打标签，而这些标签更有利于我们做商品推荐，目前打上标签的数据已达5亿+，
用户量大概1亿+，项目需求就是根据各种组合条件寻找标签和人群信息。
举个例子：
集合A： ( 购买过“牙膏“的人交易金额在10-500元并且交易次数在5次的客户并且平均订单价在20 -200元) 。
集合B： （购买过“牙刷”的人交易金额在5-50 并且交易次数在3次的客户并且平均订单价在10-30元）。
求：<1> 获取集合A 交 集合B 客户数 和 客户的具体信息，希望时间最好不要超过15s。
上面这种问题如果你用mysql做的话，基本上是算不出来的，时间上更无法满足项目需求。
一：寻找解决方案
如果你用最小的工作量解决这个问题的话，可以搭建一个分布式的Elasticsearch集群，查询中相关的Nick,AvgPrice,TradeCount,TradeAmont字段可以用
keyword模式存储，避免出现fieldData字段无法查询的问题，虽然ES大体上可以解决这个问题，但是熟悉ES的朋友应该知道，它的各种查询都是我们通过json
的格式去定制，虽然可以使用少量的script脚本，但是灵活度相比spark来说的话太弱基了，用scala函数式语言定制那是多么的方便，第二个是es在group by的
桶分页特别不好实现，也很麻烦，社区里面有一些 sql on elasticsearch 的框架，大家可以看看：https://github.com/NLPchina/elasticsearch-sql，只支持一
些简单的sql查询，不过像having这样的关键词是不支持的，跟sparksql是没法比的，基于以上原因，决定用spark试试看。
二：环境搭建
搭建spark集群，需要hadoop + spark + java + scala，搭建之前一定要注意各自版本的对应关系，否则遇到各种奇葩的错误让你好受哈！！！不信去官网看
看： https://spark.apache.org/downloads.html 。
这里我采用的组合是：
hadoop-2.7.6.tar.gz
jdk-8u144-linux-x64.tar.gz
scala-2.11.0.tgz
spark-2.2.1-bin-hadoop2.7.tgz
jdk-8u144-linux-x64.tar.gz
mysql-connector-java-5.1.46.jar
sqoop-1.4.7.bin__hadoop-2.6.0.tar.gz
使用3台虚拟机：一台【namenode +resourcemanager + spark master node】 + 二台 【datanode + nodemanager + spark work data】
192.168.2.227 hadoop-spark-master
192.168.2.119 hadoop-spark-salve1
192.168.2.232 hadoop-spark-salve2
1. 先配置三台机器的免ssh登录。
[root@localhost ~]# ssh-keygen -t rsa -P \'\' Generating public/private rsa key pair. Enter file in which to save the key (/root/.ssh/id_rsa): /root/.ssh/id_rsa already exists. Overwrite (y/n)? y Your identification has been saved in /root/.ssh/id_rsa. Your public key has been saved in /root/.ssh/id_rsa.pub. The key fingerprint is: 0f:4e:26:4a:ce:7d:08:b0:7e:13:82:c6:10:77:a2:5d root@localhost.localdomain The key\'s randomart image is: +--[ RSA 2048]----+ |. o E | | = + | |o o | |o. o | |.oo + . S | |.. = = * o | | . * o o . | | . . . | | | +-----------------+ [root@localhost ~]# ls /root/.ssh authorized_keys id_rsa id_rsa.pub known_hosts [root@localhost ~]#
2. 然后将公钥文件 id_rsa.pub copy到另外两台机器，这样就可以实现hadoop-spark-master 免密登录到另外两台
slave上去了。
scp /root/.ssh/id_rsa.pub root@192.168.2.119:/root/.ssh/authorized_keys scp /root/.ssh/id_rsa.pub root@192.168.2.232:/root/.ssh/authorized_keys cat /root/.ssh/id_rsa.pub >> /root/.ssh/authorized_keys
3. 在三台机器上增加如下的host映射。
[root@hadoop-spark-master ~]# cat /etc/hosts 127.0.0.1 localhost localhost.localdomain localhost4 localhost4.localdomain4 ::1 localhost localhost.localdomain localhost6 localhost6.localdomain6 192.168.2.227 hadoop-spark-master 192.168.2.119 hadoop-spark-salve1 192.168.2.232 hadoop-spark-salve2
4. 然后就是把我列举的那些 tar.gz 下载下来之后，在/etc/profile中配置如下，然后copy到另外两台salves机器上。
[root@hadoop-spark-master ~]# tail -10 /etc/profile export JAVA_HOME=/usr/myapp/jdk8 export NODE_HOME=/usr/myapp/node export SPARK_HOME=/usr/myapp/spark export SCALA_HOME=/usr/myapp/scala export HADOOP_HOME=/usr/myapp/hadoop export HADOOP_CONF_DIR=/usr/myapp/hadoop/etc/hadoop export LD_LIBRARY_PATH=/usr/myapp/hadoop/lib/native:$LD_LIBRARY_PATH export SQOOP=/usr/myapp/sqoop export NODE=/usr/myapp/node export PATH=$NODE/bin:$SQOOP/bin:$SCALA_HOME/bin:$HADOOP_HOME/bin:$HADOOP/sbin$SPARK_HOME/bin:$NODE_HOME/bin:$JAVA_HOME/bin:$PATH
5. 最后就是hadoop的几个配置文件的配置了。
《1》core-site.xml
[root@hadoop-spark-master hadoop]# cat core-site.xml <?xml version="1.0" encoding="UTF-8"?> <?xml-stylesheet type="text/xsl" href="configuration.xsl"?> <!-- Licensed under the Apache License, Version 2.0 (the "License"); you may not use this file except in compliance with the License. You may obtain a copy of the License at http://www.apache.org/licenses/LICENSE-2.0 Unless required by applicable law or agreed to in writing, software distributed under the License is distributed on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. See the License for the specific language governing permissions and limitations under the License. See accompanying LICENSE file. --> <!-- Put site-specific property overrides in this file. --> <configuration> <property> <name>hadoop.tmp.dir</name> <value>/usr/myapp/hadoop/data</value> <description>A base for other temporary directories.</description> </property> <property> <name>fs.defaultFS</name> <value>hdfs://hadoop-spark-master:9000</value> </property> </configuration>
《2》 hdfs-site.xml ：当然也可以在这里使用 dfs.datanode.data.dir 挂载多个硬盘：
[root@hadoop-spark-master hadoop]# cat hdfs-site.xml <?xml version="1.0" encoding="UTF-8"?> <?xml-stylesheet type="text/xsl" href="configuration.xsl"?> <!-- Licensed under the Apache License, Version 2.0 (the "License"); you may not use this file except in compliance with the License. You may obtain a copy of the License at http://www.apache.org/licenses/LICENSE-2.0 Unless required by applicable law or agreed to in writing, software distributed under the License is distributed on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. See the License for the specific language governing permissions and limitations under the License. See accompanying LICENSE file. --> <!-- Put site-specific property overrides in this file. --> <configuration> <property> <name>dfs.replication</name> <value>2</value> </property> </configuration>
《3》 mapred-site.xml 这个地方将mapreduce的运作寄存于yarn集群。
[root@hadoop-spark-master hadoop]# cat mapred-site.xml <?xml version="1.0"?> <?xml-stylesheet type="text/xsl" href="configuration.xsl"?> <!-- Licensed under the Apache License, Version 2.0 (the "License"); you may not use this file except in compliance with the License. You may obtain a copy of the License at http://www.apache.org/licenses/LICENSE-2.0 Unless required by applicable law or agreed to in writing, software distributed under the License is distributed on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. See the License for the specific language governing permissions and limitations under the License. See accompanying LICENSE file. --> <!-- Put site-specific property overrides in this file. --> <configuration> <property> <name>mapreduce.framework.name</name> <value>yarn</value> </property> </configuration>
《4》 yarn-site.xml 【这里要配置resoucemanager的相关地址，方便slave进行连接，否则你的集群会跑不起来的】
[root@hadoop-spark-master hadoop]# cat yarn-site.xml <?xml version="1.0"?> <!-- Licensed under the Apache License, Version 2.0 (the "License"); you may not use this file except in compliance with the License. You may obtain a copy of the License at http://www.apache.org/licenses/LICENSE-2.0 Unless required by applicable law or agreed to in writing, software distributed under the License is distributed on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. See the License for the specific language governing permissions and limitations under the License. See accompanying LICENSE file. --> <configuration> <!-- Site specific YARN configuration properties --> <property> <name>yarn.nodemanager.aux-services</name> <value>mapreduce_shuffle</value> </property> <property> <name>yarn.resourcemanager.address</name> <value>hadoop-spark-master:8032</value> </property> <property> <name>yarn.resourcemanager.scheduler.address</name> <value>hadoop-spark-master:8030</value> </property> <property> <name>yarn.resourcemanager.resource-tracker.address</name> <value>hadoop-spark-master:8031</value> </property> </configuration>
《5》 修改slaves文件，里面写入的各自salve的地址。
[root@hadoop-spark-master hadoop]# cat slaves
hadoop-spark-salve1
hadoop-spark-salve2
《6》这些都配置完成之后，可以用scp把整个hadoop文件scp到两台slave机器上。
scp /usr/myapp/hadoop root@192.168.2.119:/usr/myapp/hadoop
scp /usr/myapp/hadoop root@192.168.2.232:/usr/myapp/hadoop
《7》因为hdfs是分布式文件系统，使用之前先给hdfs格式化一下，因为当前hadoop已经灌了很多数据，就不真的执行format啦！
[root@hadoop-spark-master bin]# ./hdfs namenode -format
[root@hadoop-spark-master bin]# pwd
/usr/myapp/hadoop/bin
《8》 然后分别开启 start-dfs.sh 和 start-yarn.sh ，或者干脆点直接执行 start-all.sh 也可以，不然后者已经是官方准备废弃的方式。
[root@hadoop-spark-master sbin]# ls
distribute-exclude.sh hdfs-config.sh refresh-namenodes.sh start-balancer.sh start-yarn.cmd stop-balancer.sh stop-yarn.cmd
hadoop-daemon.sh httpfs.sh slaves.sh start-dfs.cmd start-yarn.sh stop-dfs.cmd stop-yarn.sh
hadoop-daemons.sh kms.sh start-all.cmd start-dfs.sh stop-all.cmd stop-dfs.sh yarn-daemon.sh
hdfs-config.cmd mr-jobhistory-daemon.sh start-all.sh start-secure-dns.sh stop-all.sh stop-secure-dns.sh yarn-daemons.sh
《9》 记住，只要在hadoop-spark-master 节点开启 dfs 和yarn就可以了，不需要到其他机器。
[root@hadoop-spark-master sbin]# ./start-dfs.sh
Starting namenodes on [hadoop-spark-master]
hadoop-spark-master: starting namenode, logging to /usr/myapp/hadoop/logs/hadoop-root-namenode-hadoop-spark-master.out
hadoop-spark-salve2: starting datanode, logging to /usr/myapp/hadoop/logs/hadoop-root-datanode-hadoop-spark-salve2.out
hadoop-spark-salve1: starting datanode, logging to /usr/myapp/hadoop/logs/hadoop-root-datanode-hadoop-spark-salve1.out
Starting secondary namenodes [0.0.0.0]
0.0.0.0: starting secondarynamenode, logging to /usr/myapp/hadoop/logs/hadoop-root-secondarynamenode-hadoop-spark-master.out
[root@hadoop-spark-master sbin]# ./start-yarn.sh
starting yarn daemons
starting resourcemanager, logging to /usr/myapp/hadoop/logs/yarn-root-resourcemanager-hadoop-spark-master.out
hadoop-spark-salve1: starting nodemanager, logging to /usr/myapp/hadoop/logs/yarn-root-nodemanager-hadoop-spark-salve1.out
hadoop-spark-salve2: starting nodemanager, logging to /usr/myapp/hadoop/logs/yarn-root-nodemanager-hadoop-spark-salve2.out
[root@hadoop-spark-master sbin]# jps
5671 NameNode
5975 SecondaryNameNode
6231 ResourceManager
6503 Jps
然后到其他两台slave上可以看到dataNode都开启了。
[root@hadoop-spark-salve1 ~]# jps 5157 Jps 4728 DataNode 4938 NodeManager
[root@hadoop-spark-salve2 ~]# jps 4899 Jps 4458 DataNode 4669 NodeManager
到此hadoop就搭建完成了。
三：Spark搭建
如果仅仅是搭建spark 的 standalone模式的话，只需要在conf下修改slave文件即可，把两个work节点塞进去。
[root@hadoop-spark-master conf]# tail -5 slaves # A Spark Worker will be started on each of the machines listed below hadoop-spark-salve1 hadoop-spark-salve2 [root@hadoop-spark-master conf]# pwd /usr/myapp/spark/conf
然后还是通过scp 把整个conf文件copy过去即可，然后在sbin目录下执行start-all.sh 脚本即可。
[root@hadoop-spark-master sbin]# ./start-all.sh starting org.apache.spark.deploy.master.Master, logging to /usr/myapp/spark/logs/spark-root-org.apache.spark.deploy.master.Master-1-hadoop-spark-master.out hadoop-spark-salve1: starting org.apache.spark.deploy.worker.Worker, logging to /usr/myapp/spark/logs/spark-root-org.apache.spark.deploy.worker.Worker-1-hadoop-spark-salve1.out hadoop-spark-salve2: starting org.apache.spark.deploy.worker.Worker, logging to /usr/myapp/spark/logs/spark-root-org.apache.spark.deploy.worker.Worker-1-hadoop-spark-salve2.out [root@hadoop-spark-master sbin]# jps 6930 Master 7013 Jps 5671 NameNode 5975 SecondaryNameNode 6231 ResourceManager [root@hadoop-spark-master sbin]#
然后你会发现slave1 和 slave2 节点上多了一个work节点。
[root@hadoop-spark-salve1 ~]# jps 4728 DataNode 4938 NodeManager 5772 Jps 5646 Worker
[root@hadoop-spark-salve2 ~]# jps 5475 Jps 4458 DataNode 4669 NodeManager 5342 Worker
接下来就可以看下成果啦。
http://hadoop-spark-master:50070/dfshealth.html#tab-datanode 这个是hdfs 的监控视图，可以清楚的看到有两个DataNode。
http://hadoop-spark-master:8088/cluster/nodes 这个是yarn的一个节点监控。
http://hadoop-spark-master:8080/ 这个就是spark的计算集群。
四： 使用sqoop导入数据
基础架构搭建之后，现在就可以借助sqoop将mysql的数据导入到hadoop中，导入的格式采用parquet 列式存储格式，不过这里要注意的一点就是一定要
把mysql-connector-java-5.1.46.jar 这个驱动包丢到 sqoop的lib目录下。
[root@hadoop-spark-master lib]# ls ant-contrib-1.0b3.jar commons-logging-1.1.1.jar kite-data-mapreduce-1.1.0.jar parquet-format-2.2.0-rc1.jar ant-eclipse-1.0-jvm1.2.jar hsqldb-1.8.0.10.jar kite-hadoop-compatibility-1.1.0.jar parquet-generator-1.6.0.jar avro-1.8.1.jar jackson-annotations-2.3.1.jar mysql-connector-java-5.1.46.jar parquet-hadoop-1.6.0.jar avro-mapred-1.8.1-hadoop2.jar jackson-core-2.3.1.jar opencsv-2.3.jar parquet-jackson-1.6.0.jar commons-codec-1.4.jar jackson-core-asl-1.9.13.jar paranamer-2.7.jar slf4j-api-1.6.1.jar commons-compress-1.8.1.jar jackson-databind-2.3.1.jar parquet-avro-1.6.0.jar snappy-java-1.1.1.6.jar commons-io-1.4.jar jackson-mapper-asl-1.9.13.jar parquet-column-1.6.0.jar xz-1.5.jar commons-jexl-2.1.1.jar kite-data-core-1.1.0.jar parquet-common-1.6.0.jar commons-lang3-3.4.jar kite-data-hive-1.1.0.jar parquet-encoding-1.6.0.jar [root@hadoop-spark-master lib]# pwd /usr/myapp/sqoop/lib
接下来我们就可以导入数据了，我准备把db=zuanzhan ,table=dsp_customertag的表，大概155w的数据导入到hadoop的test路径中，因为是测试环
境没办法，文件格式为parquet列式存储。
[root@hadoop-spark-master lib]# [root@hadoop-spark-master bin]# sqoop import --connect jdbc:mysql://192.168.2.166:3306/zuanzhan --username admin --password 123456 --table dsp_customertag --m 1 --target-dir test --as-parquetfile bash: [root@hadoop-spark-master: command not found... [root@hadoop-spark-master lib]# sqoop import --connect jdbc:mysql://192.168.2.166:3306/zuanzhan --username admin --password 123456 --table dsp_customertag --m 1 --target-dir test --as-parquetfile Warning: /usr/myapp/sqoop/bin/../../hbase does not exist! HBase imports will fail. Please set $HBASE_HOME to the root of your HBase installation. Warning: /usr/myapp/sqoop/bin/../../hcatalog does not exist! HCatalog jobs will fail. Please set $HCAT_HOME to the root of your HCatalog installation. Warning: /usr/myapp/sqoop/bin/../../accumulo does not exist! Accumulo imports will fail. Please set $ACCUMULO_HOME to the root of your Accumulo installation. Warning: /usr/myapp/sqoop/bin/../../zookeeper does not exist! Accumulo imports will fail. Please set $ZOOKEEPER_HOME to the root of your Zookeeper installation. 18/05/29 00:19:40 INFO sqoop.Sqoop: Running Sqoop version: 1.4.7 18/05/29 00:19:40 WARN tool.BaseSqoopTool: Setting your password on the command-line is insecure. Consider using -P instead. 18/05/29 00:19:40 INFO manager.MySQLManager: Preparing to use a MySQL streaming resultset. 18/05/29 00:19:40 INFO tool.CodeGenTool: Beginning code generation 18/05/29 00:19:40 INFO tool.CodeGenTool: Will generate java class as codegen_dsp_customertag 18/05/29 00:19:41 INFO manager.SqlManager: Executing SQL statement: SELECT t.* FROM `dsp_customertag` AS t LIMIT 1 18/05/29 00:19:42 INFO manager.SqlManager: Executing SQL statement: SELECT t.* FROM `dsp_customertag` AS t LIMIT 1 18/05/29 00:19:42 INFO orm.CompilationManager: HADOOP_MAPRED_HOME is /usr/myapp/hadoop Note: /tmp/sqoop-root/compile/0020f679e735b365bf96dabecb1611de/codegen_dsp_customertag.java uses or overrides a deprecated API. Note: Recompile with -Xlint:deprecation for details. 18/05/29 00:19:48 INFO orm.CompilationManager: Writing jar file: /tmp/sqoop-root/compile/0020f679e735b365bf96dabecb1611de/codegen_dsp_customertag.jar 18/05/29 00:19:48 WARN manager.MySQLManager: It looks like you are importing from mysql. 18/05/29 00:19:48 WARN manager.MySQLManager: This transfer can be faster! Use the --direct 18/05/29 00:19:48 WARN manager.MySQLManager: option to exercise a MySQL-specific fast path. 18/05/29 00:19:48 INFO manager.MySQLManager: Setting zero DATETIME behavior to convertToNull (mysql) 18/05/29 00:19:48 WARN manager.CatalogQueryManager: The table dsp_customertag contains a multi-column primary key. Sqoop will default to the column CustomerTagId only for this job. 18/05/29 00:19:48 WARN manager.CatalogQueryManager: The table dsp_customertag contains a multi-column primary key. Sqoop will default to the column CustomerTagId only for this job. 18/05/29 00:19:48 INFO mapreduce.ImportJobBase: Beginning import of dsp_customertag 18/05/29 00:19:48 INFO Configuration.deprecation: mapred.jar is deprecated. Instead, use mapreduce.job.jar 18/05/29 00:19:50 INFO manager.SqlManager: Executing SQL statement: SELECT t.* FROM `dsp_customertag` AS t LIMIT 1 18/05/29 00:19:50 INFO manager.SqlManager: Executing SQL statement: SELECT t.* FROM `dsp_customertag` AS t LIMIT 1 18/05/29 00:19:51 WARN spi.Registration: Not loading URI patterns in org.kitesdk.data.spi.hive.Loader 18/05/29 00:19:53 INFO Configuration.deprecation: mapred.map.tasks is deprecated. Instead, use mapreduce.job.maps 18/05/29 00:19:53 INFO client.RMProxy: Connecting to ResourceManager at hadoop-spark-master/192.168.2.227:8032 18/05/29 00:19:57 INFO db.DBInputFormat: Using read commited transaction isolation 18/05/29 00:19:57 INFO mapreduce.JobSubmitter: number of splits:1 18/05/29 00:19:57 INFO mapreduce.JobSubmitter: Submitting tokens for job: job_1527575811851_0001 18/05/29 00:19:57 INFO impl.YarnClientImpl: Submitted application application_1527575811851_0001 18/05/29 00:19:58 INFO mapreduce.Job: The url to track the job: http://hadoop-spark-master:8088/proxy/application_1527575811851_0001/ 18/05/29 00:19:58 INFO mapreduce.Job: Running job: job_1527575811851_0001 18/05/29 00:20:07 INFO mapreduce.Job: Job job_1527575811851_0001 running in uber mode : false 18/05/29 00:20:07 INFO mapreduce.Job: map 0% reduce 0% 18/05/29 00:20:26 INFO mapreduce.Job: map 100% reduce 0% 18/05/29 00:20:26 INFO mapreduce.Job: Job job_1527575811851_0001 completed successfully 18/05/29 00:20:26 INFO mapreduce.Job: Counters: 30 File System Counters FILE: Number of bytes read=0 FILE: Number of bytes written=142261 FILE: Number of read operations=0 FILE: Number of large read operations=0 FILE: Number of write operations=0 HDFS: Number of bytes read=8616 HDFS: Number of bytes written=28954674 HDFS: Number of read operations=50 HDFS: Number of large read operations=0 HDFS: Number of write operations=10 Job Counters Launched map tasks=1 Other local map tasks=1 Total time spent by all maps in occupied slots (ms)=16729 Total time spent by all reduces in occupied slots (ms)=0 Total time spent by all map tasks (ms)=16729 Total vcore-milliseconds taken by all map tasks=16729 Total megabyte-milliseconds taken by all map tasks=17130496 Map-Reduce Framework Map input records=1556209 Map output records=1556209 Input split bytes=87 Spilled Records=0 Failed Shuffles=0 Merged Map outputs=0 GC time elapsed (ms)=1147 CPU time spent (ms)=16710 Physical memory (bytes) snapshot=283635712 Virtual memory (bytes) snapshot=2148511744 Total committed heap usage (bytes)=150994944 File Input Format Counters Bytes Read=0 File Output Format Counters Bytes Written=0 18/05/29 00:20:26 INFO mapreduce.ImportJobBase: Transferred 27.6133 MB in 32.896 seconds (859.5585 KB/sec) 18/05/29 00:20:26 INFO mapreduce.ImportJobBase: Retrieved 1556209 records.
然后可以在UI中看到有这么一个parquet文件。
五：使用python对spark进行操作
之前使用scala对spark进行操作，使用maven进行打包，用起来不大方便，采用python还是很方便的，大家先要下载一个pyspark的安装包，一定要和spark
的版本对应起来。 pypy官网：https://pypi.org/project/pyspark/2.2.1/
你可以在master机器和开发机上直接安装 pyspark 2.2.1 模板，这样master机上执行就不需要通过pyspark-shell提交给spark集群了，下面我使用清华大学的
临时镜像下载的，毕竟官网的pip install不要太慢。
pip install -i https://pypi.tuna.tsinghua.edu.cn/simple pyspark==2.2.1
下面就是app.py脚本，采用spark sql 的模式。
# coding=utf-8 import time; import sys; from pyspark.sql import SparkSession; from pyspark.conf import SparkConf # reload(sys); # sys.setdefaultencoding(\'utf8\'); logFile = "hdfs://hadoop-spark-master:9000/user/root以上是关于spark集群搭建整理之解决亿级人群标签问题的主要内容，如果未能解决你的问题，请参考以下文章