数据质量模块是大数据平台中必不可少的一个功能组件,(以下简称Griffin)是一个开源的大数据数据质量解决方案,它支持批处理和流模式两种数据质量检测方式,可以从不同维度(比如离线任务执行完毕后检查源端和目标端的数据数量是否一致、源表的数据空值数量等)度量数据资产,从而提升数据的准确度、可信度。
在Griffin的架构中,主要分为Define、Measure和Analyze三个部分,如下图所示:
各部分的职责如下:
- Define:主要负责定义数据质量统计的维度,比如数据质量统计的时间跨度、统计的目标(源端和目标端的数据数量是否一致,数据源里某一字段的非空的数量、不重复值的数量、最大值、最小值、top5的值数量等)
- Measure:主要负责执行统计任务,生成统计结果
- Analyze:主要负责保存与展示统计结果
基于以上功能,我们大数据平台计划引入Griffin作为数据质量解决方案,实现数据一致性检查、空值统计等功能。以下是安装步骤总结:
安装部署
依赖准备
- JDK (1.8 or later versions)
- MySQL(version 5.6及以上)
- Hadoop (2.6.0 or later)
- Hive (version 2.x)
- Spark (version 2.2.1)
- Livy(livy-0.5.0-incubating)
- ElasticSearch (5.0 or later versions)
初始化
初始化操作具体请参考,由于我的测试环境中Hadoop集群、Hive集群已搭好,故这里省略Hadoop、Hive安装步骤,只保留拷贝配置文件、配置Hadoop配置文件目录步骤。
1、MySQL:
在MySQL中创建数据库quartz,然后执行脚本初始化表信息:
mysql -u-p < Init_quartz_mysql_innodb.sql
2、Hadoop和Hive:
从Hadoop服务器拷贝配置文件到Livy服务器上,这里假设将配置文件放在/usr/data/conf目录下。
在Hadoop服务器上创建/home/spark_conf目录,并将Hive的配置文件hive-site.xml上传到该目录下:
#创建/home/spark_conf目录hadoop fs -mkdir -p /home/spark_conf#上传hive-site.xmlhadoop fs -put hive-site.xml /home/spark_conf/
3、设置环境变量:
#!/bin/bashexport JAVA_HOME=/data/jdk1.8.0_192#spark目录export SPARK_HOME=/usr/data/spark-2.1.1-bin-2.6.3#livy命令目录export LIVY_HOME=/usr/data/livy/bin#hadoop配置文件目录export HADOOP_CONF_DIR=/usr/data/conf
4、Livy配置:
更新livy/conf下的livy.conf配置文件:
livy.server.host = 127.0.0.1livy.spark.master = yarnlivy.spark.deployMode = clusterlivy.repl.enable-hive-context = true
启动livy:
livy-server start
5、Elasticsearch配置:
在ES里创建griffin索引:
curl -XPUT http://es:9200/griffin -d '{ "aliases": {}, "mappings": { "accuracy": { "properties": { "name": { "fields": { "keyword": { "ignore_above": 256, "type": "keyword" } }, "type": "text" }, "tmst": { "type": "date" } } } }, "settings": { "index": { "number_of_replicas": "2", "number_of_shards": "5" } }}'
源码打包部署
在这里我使用源码编译打包的方式来部署Griffin,Griffin的源码地址是:,这里我使用的源码tag是griffin-0.4.0,下载完成在idea中导入并展开源码的结构图如下:
Griffin的源码结构很清晰,主要包括griffin-doc、measure、service和ui四个模块,其中griffin-doc负责存放Griffin的文档,measure负责与spark交互,执行统计任务,service使用spring boot作为服务实现,负责给ui模块提供交互所需的restful api,保存统计任务,展示统计结果。
源码导入构建完毕后,需要修改配置文件,具体修改的配置文件如下:
1、service/src/main/resources/application.properties:
# Apache Griffin应用名称spring.application.name=griffin_service# MySQL数据库配置信息spring.datasource.url=jdbc:mysql://10.104.20.126:3306/griffin_quartz?useSSL=falsespring.datasource.username=xnuserspring.datasource.password=Xn20!@n0oLkspring.jpa.generate-ddl=truespring.datasource.driver-class-name=com.mysql.jdbc.Driverspring.jpa.show-sql=true# Hive metastore配置信息hive.metastore.uris=thrift://namenodetest01.bi:9083hive.metastore.dbname=defaulthive.hmshandler.retry.attempts=15hive.hmshandler.retry.interval=2000ms# Hive cache timecache.evict.hive.fixedRate.in.milliseconds=900000# Kafka schema registry,按需配置kafka.schema.registry.url=http://namenodetest01.bi:8081# Update job instance state at regular intervalsjobInstance.fixedDelay.in.milliseconds=60000# Expired time of job instance which is 7 days that is 604800000 milliseconds.Time unit only supports millisecondsjobInstance.expired.milliseconds=604800000# schedule predicate job every 5 minutes and repeat 12 times at most#interval time unit s:second m:minute h:hour d:day,only support these four unitspredicate.job.interval=5mpredicate.job.repeat.count=12# external properties directory locationexternal.config.location=# external BATCH or STREAMING envexternal.env.location=# login strategy ("default" or "ldap")login.strategy=default# ldap,登录策略为ldap时配置ldap.url=ldap://hostname:portldap.email=@example.comldap.searchBase=DC=org,DC=exampleldap.searchPattern=(sAMAccountName={0})# hdfs default namefs.defaultFS=# elasticsearch配置elasticsearch.host=griffindq02-test1-rgtj1-tj1elasticsearch.port=9200elasticsearch.scheme=http# elasticsearch.user = user# elasticsearch.password = password# livy配置livy.uri=http://10.104.110.116:8998/batches# yarn url配置yarn.uri=http://10.104.110.116:8088# griffin event listenerinternal.event.listeners=GriffinJobEventHook
2、service/src/main/resources/quartz.properties
## Licensed to the Apache Software Foundation (ASF) under one# or more contributor license agreements. See the NOTICE file# distributed with this work for additional information# regarding copyright ownership. The ASF licenses this file# to you 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.#org.quartz.scheduler.instanceName=spring-boot-quartzorg.quartz.scheduler.instanceId=AUTOorg.quartz.threadPool.threadCount=5org.quartz.jobStore.class=org.quartz.impl.jdbcjobstore.JobStoreTX# If you use postgresql as your database,set this property value to org.quartz.impl.jdbcjobstore.PostgreSQLDelegate# If you use mysql as your database,set this property value to org.quartz.impl.jdbcjobstore.StdJDBCDelegate# If you use h2 as your database, it's ok to set this property value to StdJDBCDelegate, PostgreSQLDelegate or othersorg.quartz.jobStore.driverDelegateClass=org.quartz.impl.jdbcjobstore.StdJDBCDelegateorg.quartz.jobStore.useProperties=trueorg.quartz.jobStore.misfireThreshold=60000org.quartz.jobStore.tablePrefix=QRTZ_org.quartz.jobStore.isClustered=trueorg.quartz.jobStore.clusterCheckinInterval=20000
3、service/src/main/resources/sparkProperties.json:
{ "file": "hdfs:///griffin/griffin-measure.jar", "className": "org.apache.griffin.measure.Application", "name": "griffin", "queue": "default", "numExecutors": 2, "executorCores": 1, "driverMemory": "1g", "executorMemory": "1g", "conf": { "spark.yarn.dist.files": "hdfs:///home/spark_conf/hive-site.xml" }, "files": [ ]}
4、service/src/main/resources/env/env_batch.json:
{ "spark": { "log.level": "INFO" }, "sinks": [ { "type": "CONSOLE", "config": { "max.log.lines": 10 } }, { "type": "HDFS", "config": { "path": "hdfs://namenodetest01.bi.10101111.com:9001/griffin/persist", "max.persist.lines": 10000, "max.lines.per.file": 10000 } }, { "type": "ELASTICSEARCH", "config": { "method": "post", "api": "http://10.104.110.119:9200/griffin/accuracy", "connection.timeout": "1m", "retry": 10 } } ], "griffin.checkpoint": []}
配置文件修改好后,在idea里的terminal里执行如下maven命令进行编译打包:
mvn -Dmaven.test.skip=true clean install
命令执行完成后,会在service和measure模块的target目录下分别看到service-0.4.0.jar和measure-0.4.0.jar两个jar,将这两个jar分别拷贝到服务器目录下。这两个jar的使用方式如下:
1、使用如下命令将measure-0.4.0.jar这个jar上传到HDFS的/griffin文件目录里:
#改变jar名称mv measure-0.4.0.jar griffin-measure.jar#上传griffin-measure.jar到HDFS文件目录里hadoop fs -put measure-0.4.0.jar /griffin/
这样做的目的主要是因为spark在yarn集群上执行任务时,需要到HDFS的/griffin目录下加载griffin-measure.jar,避免发生类org.apache.griffin.measure.Application找不到的错误。
2、运行service-0.4.0.jar,启动Griffin管理后台:
nohup java -jar service-0.4.0.jar>service.out 2>&1 &
几秒钟后,我们可以访问Apache Griffin的默认UI(默认情况下,spring boot的端口是8080)。
http://IP:8080
UI操作文档链接:。通过UI操作界面,我们可以创建自己的统计任务,部分结果展示界面如下:
功能体验
1、在hive里创建表demo_src和demo_tgt:
--create hive tables here. hql script--Note: replace hdfs location with your own pathCREATE EXTERNAL TABLE `demo_src`( `id` bigint, `age` int, `desc` string) PARTITIONED BY ( `dt` string, `hour` string)ROW FORMAT DELIMITED FIELDS TERMINATED BY '|'LOCATION 'hdfs:///griffin/data/batch/demo_src';--Note: replace hdfs location with your own pathCREATE EXTERNAL TABLE `demo_tgt`( `id` bigint, `age` int, `desc` string) PARTITIONED BY ( `dt` string, `hour` string)ROW FORMAT DELIMITED FIELDS TERMINATED BY '|'LOCATION 'hdfs:///griffin/data/batch/demo_tgt';
2、生成测试数据:
从地址下载所有文件到Hadoop服务器上,然后使用如下命令执行gen-hive-data.sh脚本:
nohup ./gen-hive-data.sh>gen.out 2>&1 &
注意观察gen.out日志文件,如果有错误,视情况进行调整。这里我的测试环境Hadoop和Hive安装在同一台服务器上,因此直接运行脚本。
3、通过UI界面创建统计任务,具体按照 一步步操作。
相关命令
# 启动service.jarnohup java -jar service-0.4.0.jar>service.out 2>&1 & # 运行测试数据生成脚本nohup ./gen-hive-data.sh>gen.out 2>&1 &# 查询后台任务jobs -l# 查看livy日志tail -f /usr/data/livy/logs/livy-root-server.out# 启动livy-serverlivy-server start# 停止livy-serverlivy-server stop
踩坑记录
1、gen-hive-data.sh脚本生成数据失败,报no such file or directory错误。
错误原因:HDFS中的/griffin/data/batch/demo_src/和/griffin/data/batch/demo_tgt/目录下"dt=时间"目录不存在,如dt=20190113。
解决办法:给脚本中增加hadoop fs -mkdir创建目录操作,修改完后如下:
#!/bin/bash#create tablehive -f create-table.hqlecho "create table done"#current hoursudo ./gen_demo_data.shcur_date=`date +%Y%m%d%H`dt=${cur_date:0:8}hour=${cur_date:8:2}partition_date="dt='$dt',hour='$hour'"sed s/PARTITION_DATE/$partition_date/ ./insert-data.hql.template > insert-data.hqlhive -f insert-data.hqlsrc_done_path=/griffin/data/batch/demo_src/dt=${dt}/hour=${hour}/_DONEtgt_done_path=/griffin/data/batch/demo_tgt/dt=${dt}/hour=${hour}/_DONEhadoop fs -mkdir -p /griffin/data/batch/demo_src/dt=${dt}/hour=${hour}hadoop fs -mkdir -p /griffin/data/batch/demo_tgt/dt=${dt}/hour=${hour}hadoop fs -touchz ${src_done_path}hadoop fs -touchz ${tgt_done_path}echo "insert data [$partition_date] done"#last hoursudo ./gen_demo_data.shcur_date=`date -d '1 hour ago' +%Y%m%d%H`dt=${cur_date:0:8}hour=${cur_date:8:2}partition_date="dt='$dt',hour='$hour'"sed s/PARTITION_DATE/$partition_date/ ./insert-data.hql.template > insert-data.hqlhive -f insert-data.hqlsrc_done_path=/griffin/data/batch/demo_src/dt=${dt}/hour=${hour}/_DONEtgt_done_path=/griffin/data/batch/demo_tgt/dt=${dt}/hour=${hour}/_DONEhadoop fs -mkdir -p /griffin/data/batch/demo_src/dt=${dt}/hour=${hour}hadoop fs -mkdir -p /griffin/data/batch/demo_tgt/dt=${dt}/hour=${hour}hadoop fs -touchz ${src_done_path}hadoop fs -touchz ${tgt_done_path}echo "insert data [$partition_date] done"#next hoursset +ewhile truedo sudo ./gen_demo_data.sh cur_date=`date +%Y%m%d%H` next_date=`date -d "+1hour" '+%Y%m%d%H'` dt=${next_date:0:8} hour=${next_date:8:2} partition_date="dt='$dt',hour='$hour'" sed s/PARTITION_DATE/$partition_date/ ./insert-data.hql.template > insert-data.hql hive -f insert-data.hql src_done_path=/griffin/data/batch/demo_src/dt=${dt}/hour=${hour}/_DONE tgt_done_path=/griffin/data/batch/demo_tgt/dt=${dt}/hour=${hour}/_DONE hadoop fs -mkdir -p /griffin/data/batch/demo_src/dt=${dt}/hour=${hour} hadoop fs -mkdir -p /griffin/data/batch/demo_tgt/dt=${dt}/hour=${hour} hadoop fs -touchz ${src_done_path} hadoop fs -touchz ${tgt_done_path} echo "insert data [$partition_date] done" sleep 3600doneset -e
2、HDFS的/griffin/persist目录下没有统计结果文件,检查该目录的权限,设置合适的权限即可。
3、ES中的metric数据为空,有两种可能:
- service/src/main/resources/env/env_batch.json里的ES配置信息不正确
- 执行spark任务的yarn服务器上没有配置ES服务器的hostname,连接异常
4、启动service-0.4.0.jar之后,访问不到UI界面,查看启动日志无异常。检查打包时是不是执行的mvn package命令,将该命令替换成mvn -Dmaven.test.skip=true clean install命令重新打包启动即可。