DolphinScheduler/docs/zh_CN/后端部署文档.md

后端部署文档

基础软件安装

• Mysql (5.5+) : 必装
• JDK (1.8+) : 必装
• ZooKeeper(3.4.6) ：必装
• Hadoop(2.7.3) ：选装， 如果需要使用到资源上传功能，MapReduce任务提交则需要配置Hadoop(上传的资源文件目前保存在Hdfs上)
• Hive(1.2.1) : 选装，hive任务提交需要安装
• Spark(1.x,2.x) : 选装，Spark任务提交需要安装
• PostgreSQL(8.2.15+) : 选装，PostgreSQL PostgreSQL存储过程需要安装

 注意：EasyScheduler本身不依赖Hadoop、Hive、Spark、PostgreSQL,仅是会调用他们的Client，用于对应任务的运行。

项目编译

• 执行编译命令：

 mvn -U clean package assembly:assembly -Dmaven.test.skip=true

• 查看目录

正常编译完后，会在当前目录生成 target/escheduler-{version}/

    bin
    conf
    lib
    script
    sql
    install.sh

• 说明

bin : 基础服务启动脚本
conf : 项目配置文件
lib : 项目依赖jar包，包括各个模块jar和第三方jar
script : 集群启动、停止和服务监控启停脚本
sql : 项目依赖sql文件
install.sh : 一键部署脚本

数据库初始化

• 创建database和账号

mysql -h {host} -u {user} -p{password}
mysql> CREATE DATABASE escheduler DEFAULT CHARACTER SET utf8 DEFAULT COLLATE utf8_general_ci;
mysql> GRANT ALL PRIVILEGES ON escheduler.* TO '{user}'@'%' IDENTIFIED BY '{password}';
mysql> GRANT ALL PRIVILEGES ON escheduler.* TO '{user}'@'localhost' IDENTIFIED BY '{password}';
mysql> flush privileges;

• 创建表和导入基础数据

说明：在 target/escheduler-{version}/sql/escheduler.sql和quartz.sql

mysql -h {host} -u {user} -p{password} -D {db} < escheduler.sql

mysql -h {host} -u {user} -p{password} -D {db} < quartz.sql

创建部署用户

因为escheduler worker都是以 sudo -u {linux-user} 方式来执行作业，所以部署用户需要有 sudo 权限，而且是免密的。

vi /etc/sudoers

# 部署用户是 escheduler 账号
escheduler  ALL=(ALL)       NOPASSWD: NOPASSWD: ALL

# 并且需要注释掉 Default requiretty 一行
#Default requiretty

配置文件说明

说明：配置文件位于 target/escheduler-{version}/conf 下面 

escheduler-alert

配置邮件告警信息

• alert.properties

#以qq邮箱为例，如果是别的邮箱，请更改对应配置
#alert type is EMAIL/SMS
alert.type=EMAIL

# mail server configuration
mail.protocol=SMTP
mail.server.host=smtp.exmail.qq.com
mail.server.port=25
mail.sender=xxxxxxx@qq.com
mail.passwd=xxxxxxx

# xls file path, need manually create it before use if not exist
xls.file.path=/opt/xls

escheduler-common

通用配置文件配置，队列选择及地址配置，通用文件目录配置

• common/common.properties

#task queue implementation, default "zookeeper"
escheduler.queue.impl=zookeeper

# user data directory path, self configuration, please make sure the directory exists and have read write permissions
data.basedir.path=/tmp/escheduler

# directory path for user data download. self configuration, please make sure the directory exists and have read write permissions
data.download.basedir.path=/tmp/escheduler/download

# process execute directory. self configuration, please make sure the directory exists and have read write permissions
process.exec.basepath=/tmp/escheduler/exec

# data base dir, resource file will store to this hadoop hdfs path, self configuration, please make sure the directory exists on hdfs and have read write permissions。"/escheduler" is recommended
data.store2hdfs.basepath=/escheduler

# whether hdfs starts
hdfs.startup.state=true

# system env path. self configuration, please make sure the directory and file exists and have read write execute permissions
escheduler.env.path=/opt/.escheduler_env.sh
escheduler.env.py=/opt/escheduler_env.py

#resource.view.suffixs
resource.view.suffixs=txt,log,sh,conf,cfg,py,java,sql,hql,xml

# is development state? default "false"
development.state=false

SHELL任务 环境变量配置

说明：配置文件位于 target/escheduler-{version}/conf/env 下面，这个会是Worker执行任务时加载的环境

.escheduler_env.sh

export HADOOP_HOME=/opt/soft/hadoop
export HADOOP_CONF_DIR=/opt/soft/hadoop/etc/hadoop
export SPARK_HOME1=/opt/soft/spark1
export SPARK_HOME2=/opt/soft/spark2
export PYTHON_HOME=/opt/soft/python
export JAVA_HOME=/opt/soft/java
export HIVE_HOME=/opt/soft/hive
	
export PATH=$HADOOP_HOME/bin:$SPARK_HOME1/bin:$SPARK_HOME2/bin:$PYTHON_HOME/bin:$JAVA_HOME/bin:$HIVE_HOME/bin:$PATH

​

Python任务 环境变量配置

说明：配置文件位于 target/escheduler-{version}/conf/env 下面

escheduler_env.py

import os

HADOOP_HOME="/opt/soft/hadoop"
SPARK_HOME1="/opt/soft/spark1"
SPARK_HOME2="/opt/soft/spark2"
PYTHON_HOME="/opt/soft/python"
JAVA_HOME="/opt/soft/java"
HIVE_HOME="/opt/soft/hive"
PATH=os.environ['PATH']
PATH="%s/bin:%s/bin:%s/bin:%s/bin:%s/bin:%s/bin:%s"%(HIVE_HOME,HADOOP_HOME,SPARK_HOME1,SPARK_HOME2,JAVA_HOME,PYTHON_HOME,PATH)

os.putenv('PATH','%s'%PATH)	

hadoop 配置文件

• common/hadoop/hadoop.properties

# ha or single namenode,If namenode ha needs to copy core-site.xml and hdfs-site.xml to the conf directory
fs.defaultFS=hdfs://mycluster:8020

#resourcemanager ha note this need ips , this empty if single
yarn.resourcemanager.ha.rm.ids=192.168.xx.xx,192.168.xx.xx

# If it is a single resourcemanager, you only need to configure one host name. If it is resourcemanager HA, the default configuration is fine
yarn.application.status.address=http://ark1:8088/ws/v1/cluster/apps/%s

定时器配置文件

• quartz.properties

#============================================================================
# Configure Main Scheduler Properties
#============================================================================
org.quartz.scheduler.instanceName = EasyScheduler
org.quartz.scheduler.instanceId = AUTO
org.quartz.scheduler.makeSchedulerThreadDaemon = true
org.quartz.jobStore.useProperties = false

#============================================================================
# Configure ThreadPool
#============================================================================

org.quartz.threadPool.class = org.quartz.simpl.SimpleThreadPool
org.quartz.threadPool.makeThreadsDaemons = true
org.quartz.threadPool.threadCount = 25
org.quartz.threadPool.threadPriority = 5

#============================================================================
# Configure JobStore
#============================================================================
 
org.quartz.jobStore.class = org.quartz.impl.jdbcjobstore.JobStoreTX
org.quartz.jobStore.driverDelegateClass = org.quartz.impl.jdbcjobstore.StdJDBCDelegate
org.quartz.jobStore.tablePrefix = QRTZ_
org.quartz.jobStore.isClustered = true
org.quartz.jobStore.misfireThreshold = 60000
org.quartz.jobStore.clusterCheckinInterval = 5000
org.quartz.jobStore.dataSource = myDs

#============================================================================
# Configure Datasources  
#============================================================================
 
org.quartz.dataSource.myDs.driver = com.mysql.jdbc.Driver
org.quartz.dataSource.myDs.URL = jdbc:mysql://192.168.xx.xx:3306/escheduler?characterEncoding=utf8&useSSL=false
org.quartz.dataSource.myDs.user = xx
org.quartz.dataSource.myDs.password = xx
org.quartz.dataSource.myDs.maxConnections = 10
org.quartz.dataSource.myDs.validationQuery = select 1

zookeeper 配置文件

• zookeeper.properties

#zookeeper cluster
zookeeper.quorum=192.168.xx.xx:2181,192.168.xx.xx:2181,192.168.xx.xx:2181

#escheduler root directory
zookeeper.escheduler.root=/escheduler

#zookeeper server dirctory
zookeeper.escheduler.dead.servers=/escheduler/dead-servers
zookeeper.escheduler.masters=/escheduler/masters
zookeeper.escheduler.workers=/escheduler/workers

#zookeeper lock dirctory
zookeeper.escheduler.lock.masters=/escheduler/lock/masters
zookeeper.escheduler.lock.workers=/escheduler/lock/workers

#escheduler failover directory
zookeeper.escheduler.lock.masters.failover=/escheduler/lock/failover/masters
zookeeper.escheduler.lock.workers.failover=/escheduler/lock/failover/workers

#escheduler failover directory
zookeeper.session.timeout=300
zookeeper.connection.timeout=300
zookeeper.retry.sleep=1000
zookeeper.retry.maxtime=5

escheduler-dao

dao数据源配置

• dao/data_source.properties

# base spring data source configuration
spring.datasource.type=com.alibaba.druid.pool.DruidDataSource
spring.datasource.driver-class-name=com.mysql.jdbc.Driver
spring.datasource.url=jdbc:mysql://192.168.xx.xx:3306/escheduler?characterEncoding=UTF-8
spring.datasource.username=xx
spring.datasource.password=xx

# connection configuration
spring.datasource.initialSize=5
# min connection number
spring.datasource.minIdle=5
# max connection number
spring.datasource.maxActive=50

# max wait time for get a connection in milliseconds. if configuring maxWait, fair locks are enabled by default and concurrency efficiency decreases.
# If necessary, unfair locks can be used by configuring the useUnfairLock attribute to true.
spring.datasource.maxWait=60000

# milliseconds for check to close free connections
spring.datasource.timeBetweenEvictionRunsMillis=60000

# the Destroy thread detects the connection interval and closes the physical connection in milliseconds if the connection idle time is greater than or equal to minEvictableIdleTimeMillis.
spring.datasource.timeBetweenConnectErrorMillis=60000

# the longest time a connection remains idle without being evicted, in milliseconds
spring.datasource.minEvictableIdleTimeMillis=300000

#the SQL used to check whether the connection is valid requires a query statement. If validation Query is null, testOnBorrow, testOnReturn, and testWhileIdle will not work.
spring.datasource.validationQuery=SELECT 1
#check whether the connection is valid for timeout, in seconds
spring.datasource.validationQueryTimeout=3

# when applying for a connection, if it is detected that the connection is idle longer than time Between Eviction Runs Millis,
# validation Query is performed to check whether the connection is valid
spring.datasource.testWhileIdle=true

#execute validation to check if the connection is valid when applying for a connection
spring.datasource.testOnBorrow=true
#execute validation to check if the connection is valid when the connection is returned
spring.datasource.testOnReturn=false
spring.datasource.defaultAutoCommit=true
spring.datasource.keepAlive=true

# open PSCache, specify count PSCache for every connection
spring.datasource.poolPreparedStatements=true
spring.datasource.maxPoolPreparedStatementPerConnectionSize=20

escheduler-server

master配置文件

• master.properties

# master execute thread num
master.exec.threads=100

# master execute task number in parallel
master.exec.task.number=20

# master heartbeat interval
master.heartbeat.interval=10

# master commit task retry times
master.task.commit.retryTimes=5

# master commit task interval
master.task.commit.interval=100


# only less than cpu avg load, master server can work. default value : the number of cpu cores * 2
master.max.cpuload.avg=10

# only larger than reserved memory, master server can work. default value : physical memory * 1/10, unit is G.
master.reserved.memory=1

worker配置文件

• worker.properties

# worker execute thread num
worker.exec.threads=100

# worker heartbeat interval
worker.heartbeat.interval=10

# submit the number of tasks at a time
worker.fetch.task.num = 10


# only less than cpu avg load, worker server can work. default value : the number of cpu cores * 2
worker.max.cpuload.avg=10

# only larger than reserved memory, worker server can work. default value : physical memory * 1/6, unit is G.
worker.reserved.memory=1

escheduler-api

web配置文件

• application.properties

# server port
server.port=12345

# session config
server.session.timeout=7200

server.context-path=/escheduler/

# file size limit for upload
spring.http.multipart.max-file-size=1024MB
spring.http.multipart.max-request-size=1024MB

# post content
server.max-http-post-size=5000000

伪分布式部署

1，创建部署用户

​ 如上 创建部署用户

2，根据实际需求来创建HDFS根路径

​ 根据 common/common.properties 中 hdfs.startup.state 的配置来判断是否启动HDFS，如果启动，则需要创建HDFS根路径，并将 owner 修改为部署用户，否则忽略此步骤

3，项目编译

​ 如上进行 项目编译

4，修改配置文件

​ 根据 配置文件说明 修改配置文件和 环境变量 文件

5，创建目录并将环境变量文件复制到指定目录

• 创建 common/common.properties 下的data.basedir.path、data.download.basedir.path和process.exec.basepath路径

• 将**.escheduler_env.sh** 和 escheduler_env.py 两个环境变量文件复制到 common/common.properties配置的escheduler.env.path 和 escheduler.env.py 的目录下，并将 owner 修改为部署用户

6，启停服务

• 启停Master

sh ./bin/escheduler-daemon.sh start master-server
sh ./bin/escheduler-daemon.sh stop master-server

• 启停Worker

sh ./bin/escheduler-daemon.sh start worker-server
sh ./bin/escheduler-daemon.sh stop worker-server

• 启停Api

sh ./bin/escheduler-daemon.sh start api-server
sh ./bin/escheduler-daemon.sh stop api-server

• 启停Logger

sh ./bin/escheduler-daemon.sh start logger-server
sh ./bin/escheduler-daemon.sh stop logger-server

• 启停Alert

sh ./bin/escheduler-daemon.sh start alert-server
sh ./bin/escheduler-daemon.sh stop alert-server

分布式部署

1，创建部署用户

• 在需要部署调度的机器上如上 创建部署用户
• 将 主机器 和各个其它机器SSH打通

2，根据实际需求来创建HDFS根路径

​ 根据 common/common.properties 中 hdfs.startup.state 的配置来判断是否启动HDFS，如果启动，则需要创建HDFS根路径，并将 owner 修改为部署用户，否则忽略此步骤

3，项目编译

​ 如上进行 项目编译

4，将环境变量文件复制到指定目录

​ 将**.escheduler_env.sh** 和 escheduler_env.py 两个环境变量文件复制到 common/common.properties配置的escheduler.env.path 和 escheduler.env.py 的目录下，并将 owner 修改为部署用户

5，修改 install.sh

​ 修改 install.sh 中变量的值，替换成自身业务所需的值

6，一键部署

• 安装 pip install kazoo

• 使用部署用户 sh install.sh 一键部署

  • 注意：scp_hosts.sh 里 tar -zxvf $workDir/../escheduler-1.0.0.tar.gz -C $installPath 中的版本号(1.0.0)需要执行前手动替换成对应的版本号

服务监控

monitor_server.py 脚本是监听，master和worker服务挂掉重启的脚本

注意：在全部服务都启动之后启动

nohup python -u monitor_server.py > nohup.out 2>&1 &

日志查看

日志统一存放于指定文件夹内

 logs/
    ├── escheduler-alert-server.log
    ├── escheduler-master-server.log
    |—— escheduler-worker-server.log
    |—— escheduler-api-server.log
    |—— escheduler-logger-server.log