Spark Gradle jar 包分离方法

想不想把spark项目打包更快一点？如果把几百M的依赖包打成一个包是很花时间的，少则1分钟，多则几分钟，这些时间还不如留着跟前台妹子聊会天呢，经过几百遍的测试，平时要花1分20秒才能打成的jar包，现在只需要9秒，没错你没听错就是9秒，咳咳....，好了不吹牛了回归正文吧。

使用方法

build.gradle

group 'com.dounine'
version '1.0.0-SNAPSHOT'

apply plugin: 'application'
apply plugin: 'java'
apply plugin: 'scala'

sourceCompatibility = 1.8

repositories {
    mavenLocal()
    mavenCentral()
    maven { url "http://repo.hortonworks.com/content/repositories/releases/" }
}

configurations {
    provided
}

task copyJars(type: Copy) {
    from configurations.runtime
    into new File('build/libs/lib')
}
compileJava.dependsOn copyJars

def env = hasProperty("pro") ? "RELEASE" : "SNAPSHOT"

sourceSets {
    main {
        compileClasspath += configurations.provided
        resources {
            srcDirs = ["src/main/resources", "src/main/profile/$env"]
        }
    }
}

jar {
    String buildDir = project.buildDir
    manifest {
        attributes 'Implementation-Title': 'HbaseDataInsert',
                'Implementation-Version': version,
                'Main-Class': 'com.dounine.hbase.HbaseHdfsWrite',
                'Class-Path': new File(buildDir+'/libs/lib').list().collect { "lib/${it}" }.join(" ")
    }
}

dependencies {
    compile 'org.scala-lang:scala-library:2.11.12'
    compile group: 'com.alibaba', name: 'fastjson', version: '1.2.54'
    compile group: 'org.apache.spark', name: 'spark-sql_2.11', version: '2.3.1.3.0.1.0-187'
    compile group: 'org.apache.hbase', name: 'hbase-it', version: '2.0.0.3.0.1.0-187'

}

核心配置

apply plugin: 'application'
configurations {
    provided
}

task copyJars(type: Copy) {
    from configurations.runtime
    into new File('build/libs/lib')
}
compileJava.dependsOn copyJars
jar {
    String buildDir = project.buildDir
    manifest {
        attributes 'Implementation-Title': 'HbaseDataInsert',
                'Implementation-Version': version,
                'Main-Class': 'com.dounine.hbase.HbaseHdfsWrite',
                'Class-Path': new File(buildDir+'/libs/lib').list().collect { "lib/${it}" }.join(" ")
    }
}

打出来的依赖包会放在build/libs/lib里面
那应该如何提交spark任务呢?请看下面

任务提交

spark-submit --master yarn  \ 
--class com.dounine.hbase.HbaseHdfsWrite   \
--conf spark.yarn.tokens.hbase.enabled=true \
--executor-memory 1g  --driver-memory 1g  \
--keytab /etc/security/keytabs/hbase.headless.keytab  \
--jars /root/bd/hbase/build/libs/lib/a.jar,/root/bd/hbase/build/libs/lib/b.jar,......    \
--principal hbase-bd@EXAMPLE.COM build/libs/hbase-1.0.0-SNAPSHOT.jar hdfs:///tmp/logdir2

有好多个点的地方就是要大家手动补全的jar包路径啦
好啦，开个玩笑，这么玩还不如回到当初呢

最终解决方案

使用通配符

spark-submit --master yarn  \ 
--class com.dounine.hbase.HbaseHdfsWrite   \
--conf spark.yarn.tokens.hbase.enabled=true \
--executor-memory 1g  --driver-memory 1g  \
--keytab /etc/security/keytabs/hbase.headless.keytab  \
--jars "/root/bd/hbase/build/libs/lib/*"    \
--principal hbase-bd@EXAMPLE.COM build/libs/hbase-1.0.0-SNAPSHOT.jar hdfs:///tmp/logdir2

其实搞了那么久就是多出了一个双引号而已