如果是Standalone Cluster模式运行可以不依赖于Hadoop，直接下载对应版本即可。如果有时需要基于YARN来提交运行，则需要选择对应Hadoop版本的Flink安装部署。访问官网的下载页面Flink Downloads，可以看到如果是Flink 1.8版本与捆绑的Hadoop版本有Hadoop 2.4.1、Hadoop 2.6.5、Hadoop 2.7.5、Hadoop 2.8.3，将下载的对应捆绑的Hadoop jar包放到 $FLINK_HOME/lib 下即可，但如果Hadoop版本不同时也是要考虑对源码指定Hadoop版本进行编译获取分发包进行部署安装。同时如果想使用Blink新功能也是需要对源码进行编译。

1 从源码构建Flink(译自官方文档)
2 编译部署
- 2.1 环境要求
- 2.2 获取源码
- 2.3 查看版本和分支
- 2.4 切换到对应的版本或分支
- 2.5 查看当前所处的分支
- 2.6 开始编译
  - 2.6.1 指定Hadoop版本进行编译
  - 2.6.2 基于供应商特定版（如Cloudera、Hortonworks等）
  - 2.6.3 Blink
- 2.7 复制分发包到安装节点
- 2.8 配置环境变量
- 2.9 standalone模式启动
- 2.10 YARN模式
  - 2.10.1 yarn-session模式
  - 2.10.2 yarn-cluster模式
- 2.11 Blink

我们先阅读以下官方文档：Flink Development / Building Flink from Source

官方文档的目录
Build Flink (构建 Flink)
Dependency Shading (遮蔽依赖)
Hadoop Versions (Hadoop 版本)
Packaging Hadoop into the Flink distribution (将Hadoop 打包到Flink发行版中)
Vendor-specific Versions (供应商特定版本)
Scala Versions (Scala版本)
Encrypted File Systems (加密文件系统)

1 从源码构建Flink(官方文档)

1.1 Build Flink (构建 Flink)

为了构建 Flink，您需要源代码。下载发行版的源代码或克隆git存储库。

此外，您还需要 Maven 3 和 JDK（Java Development Kit）。 Flink至少需要 Java 8 才能构建。

注意：Maven 3.3.x 可以构建 Flink，但不会正确地遮蔽某些依赖项。 Maven 3.2.5正确创建了库。要构建单元测试，请使用 Java 8u51 或更高版本来防止使用 PowerMock 运行程序的单元测试失败。

要从git克隆，请输入：

git clone https://github.com/apache/flink.git

构建Flink的最简单方法是运行：

mvn clean install -DskipTests

这指示 Maven（mvn）首先清除所有已构建的（clean），然后创建一个新的 Flink 二进制文件（install）。

要加快构建速度，您可以跳过测试，QA插件和JavaDocs：

mvn clean install -DskipTests -Dfast

默认构建为 Hadoop 2 添加了特定于 Flink的JAR，以允许将 Flink 与 HDFS 和 YARN 一起使用。

扫描二维码关注公众号，回复： 9584862 查看本文章

1.2 Dependency Shading (依赖 shading)

Flink 隐藏了它使用的一些库，以避免与使用这些库的不同版本的用户程序的版本冲突。 Shaded 库包括Google Guava，Asm，Apache Curator，Apache HTTP Components，Netty等。

最近在 Maven中更改了依赖关系 shading 机制，并要求用户根据 Maven 版本略微不同地构建Flink：

Maven 3.0.x，3.1.x和3.2.x在Flink代码库的根目录中调用mvn clean install -DskipTests就足够了。
Maven 3.3.x构建必须分两步完成：首先在 base 目录中，然后在分发项目中：

mvn clean install -DskipTests
cd flink-dist
mvn clean install

注意：要检查Maven版本，请运行mvn --version。

1.3 Hadoop Versions (Hadoop 版本)

信息大多数用户不需要手动执行此操作。下载页面包含常见Hadoop版本的二进制包。

Flink 依赖于 HDFS 和 YARN，它们都是来自 Apache Hadoop 的依赖项。存在许多不同版本的 Hadoop（来自上游项目和不同的Hadoop发行版）。如果使用错误的版本组合，则可能发生异常。

Hadoop仅从2.4.0版本开始支持。您还可以指定要构建的特定Hadoop版本：

mvn clean install -DskipTests -Dhadoop.version=2.6.1

1.3.1 Packaging Hadoop into the Flink distribution (将Hadoop 打包到Flink发行版中)

如果要构建一个在 lib 文件夹中预先打包 shaded Hadoop 的 Flink 发行版，则可以使用 include-hadoop 配置文件来执行此操作。您将如上所述构建 Flink，但包括配置文件：

mvn clean install -DskipTests -Pinclude-hadoop

1.3.2 Vendor-specific Versions (供应商特定版本)

要查看支持的供应商版本列表，请查看https://mvnrepository.com/artifact/org.apache.hadoop/hadoop-hdfs?repo=cloudera
要针对特定于供应商的 Hadoop 版本构建 Flink，请执行以下命令：

mvn clean install -DskipTests -Pvendor-repos -Dhadoop.version=2.6.0-cdh5.16.1

-Pvendor-repos激活 Maven 构建配置文件，其中包括 Cloudera，Hortonworks 或 MapR 等流行的 Hadoop 供应商的存储库。

1.4 Scala Versions (Scala版本)

信息纯粹使用Java API 和库的用户可以忽略此部分。

Flink 具有用 Scala 编写的 API，库和运行时模块。 Scala API 和库的用户可能必须将 Flink 的 Scala 版本与其项目的 Scala 版本匹配（因为Scala不是严格向后兼容的）。

从版本 1.7 开始，Flink使用Scala版本2.11和2.12构建。

1.5 Encrypted File Systems (加密文件系统)

如果您的主目录已加密，您可能会遇到java.io.IOException：File name too long exception。某些加密文件系统（如 Ubuntu 使用的 encfs）不允许长文件名，这是导致此错误的原因。

解决方法是添加：

<args>
    <arg>-Xmax-classfile-name</arg>
    <arg>128</arg>
</args>

在导致错误的模块的 pom.xml 文件的编译器配置中。例如，如果错误出现在 flink-yarn 模块中，则应在 scala-maven-plugin 的标记下添加上述代码。有关更多信息，请参阅此问题。

2 编译部署

以下会对三种情况分别进行编译，指定Hadoop版本、Cloudera Hadoop版、Blink进行编译，以及Cloudera版本出现问题的解决方法，并对编译的发行包快速部署。最后对Standalone和YARN两种方式进行小的测试。

2.1 环境要求

JDK 1.8
Maven 3 （查看版本 mvn --version）
- Maven 3.0.x，3.1.x和3.2.x 只需调用mvn clean install -DskipTestsFlink代码库的根目录即可。
- Maven 3.3.x 构建必须分两步完成：首先在基本目录中，然后在分发项目中：
```
mvn clean install -DskipTests
cd flink-dist
mvn clean install
```
Scala （默认Scala使用的是2.11）。如果是纯粹使用Java API和库的用户可以忽略Scala的版本问题。

2.2 获取源码

获取源码

git clone https://github.com/apache/flink.git
#进入到源码根目录
cd flink

2.3 查看版本和分支

# 查看tag版本
git tag

#查看远程服务上的分支
git branch -r

2.4 切换到对应的版本或分支

#切换到 flink 1.8 发行版
git checkout tags/release-1.8.1

#切换到Blink分支
git checkout origin/blink

2.5 查看当前所处的分支

git branch

2.6 开始编译

hadoop版本选择可以访问mvn仓库地址：https://mvnrepository.com/artifact/org.apache.hadoop/hadoop-hdfs。

编译过程需要一段时间，耐心等待，顺利的话是可以直接编译通过。如果是依赖包的问题，可以到Maven仓库中把对应的依赖清除之后再重新编译。

2.6.1 指定Hadoop版本进行编译

例如这里选择的Hadoop版本为 3.1.2，命令中添加 -Pinclude-hadoop 是为了将Hadoop的二进制包直接放入Flink发行版的lib里。如果是Standalone模式可以可以去掉这个参数，当需要往YARN上提交时，只需要将编译的源码包下的flink-shaded-hadoop/flink-shaded-hadoop2-uber/target/flink-shaded-hadoop2-uber-3.1.2-1.8.1.jar拷贝到$FLINK_HOME/lib/下即可。这里一定要使用flink-shaded-hadoop2-uber下的包，如果使用flink-shaded-hadoop2会缺少类。

#-T2C表示一个CPU核心启动两个线程进行编译，可以加快源码编译的速度。
mvn -T2C clean install -DskipTests -Dfast -Pinclude-hadoop -Pvendor-repos -Dhadoop.version=3.1.2

2.6.2 基于供应商特定版（如Cloudera、Hortonworks等）

如果YARN集群是CDH方式搭建，需要指定Cloudera供应商版本的Hadoop，这里使用3.0.0-cdh6.3.0 版本进行编译。

mvn -T2C clean install -DskipTests -Dfast -Pinclude-hadoop -Pvendor-repos -Dhadoop.version=3.0.0-cdh6.3.0

当使用这个版本的Hadoop会报如下错误

[INFO] ------------------------------------------------------------------------
[INFO] BUILD FAILURE
[INFO] ------------------------------------------------------------------------
[INFO] Total time:  08:18 min (Wall Clock)
[INFO] Finished at: 2019-07-27T13:18:30+08:00
[INFO] ------------------------------------------------------------------------
[ERROR] Failed to execute goal org.apache.maven.plugins:maven-compiler-plugin:3.8.0:testCompile (default-testCompile) on project flink-yarn_2.11: Compilation failure
[ERROR] /opt/flink2/flink-yarn/src/test/java/org/apache/flink/yarn/AbstractYarnClusterTest.java:[89,41] no suitable method found for newInstance(org.apache.hadoop.yarn.api.records.ApplicationId,org.apache.hadoop.yarn.api.records.ApplicationAttemptId,java.lang.String,java.lang.String,java.lang.String,java.lang.String,int,<nulltype>,org.apache.hadoop.yarn.api.records.YarnApplicationState,<nulltype>,<nulltype>,long,long,org.apache.hadoop.yarn.api.records.FinalApplicationStatus,<nulltype>,<nulltype>,float,<nulltype>,<nulltype>)
[ERROR]     method org.apache.hadoop.yarn.api.records.ApplicationReport.newInstance(org.apache.hadoop.yarn.api.records.ApplicationId,org.apache.hadoop.yarn.api.records.ApplicationAttemptId,java.lang.String,java.lang.String,java.lang.String,java.lang.String,int,org.apache.hadoop.yarn.api.records.Token,org.apache.hadoop.yarn.api.records.YarnApplicationState,java.lang.String,java.lang.String,long,long,long,org.apache.hadoop.yarn.api.records.FinalApplicationStatus,org.apache.hadoop.yarn.api.records.ApplicationResourceUsageReport,java.lang.String,float,java.lang.String,org.apache.hadoop.yarn.api.records.Token) is not applicable
[ERROR]       (actual and formal argument lists differ in length)
[ERROR]     method org.apache.hadoop.yarn.api.records.ApplicationReport.newInstance(org.apache.hadoop.yarn.api.records.ApplicationId,org.apache.hadoop.yarn.api.records.ApplicationAttemptId,java.lang.String,java.lang.String,java.lang.String,java.lang.String,int,org.apache.hadoop.yarn.api.records.Token,org.apache.hadoop.yarn.api.records.YarnApplicationState,java.lang.String,java.lang.String,long,long,org.apache.hadoop.yarn.api.records.FinalApplicationStatus,org.apache.hadoop.yarn.api.records.ApplicationResourceUsageReport,java.lang.String,float,java.lang.String,org.apache.hadoop.yarn.api.records.Token,java.util.Set<java.lang.String>,boolean,org.apache.hadoop.yarn.api.records.Priority,java.lang.String,java.lang.String) is not applicable
[ERROR]       (actual and formal argument lists differ in length)
[ERROR]     method org.apache.hadoop.yarn.api.records.ApplicationReport.newInstance(org.apache.hadoop.yarn.api.records.ApplicationId,org.apache.hadoop.yarn.api.records.ApplicationAttemptId,java.lang.String,java.lang.String,java.lang.String,java.lang.String,int,org.apache.hadoop.yarn.api.records.Token,org.apache.hadoop.yarn.api.records.YarnApplicationState,java.lang.String,java.lang.String,long,long,long,org.apache.hadoop.yarn.api.records.FinalApplicationStatus,org.apache.hadoop.yarn.api.records.ApplicationResourceUsageReport,java.lang.String,float,java.lang.String,org.apache.hadoop.yarn.api.records.Token,java.util.Set<java.lang.String>,boolean,org.apache.hadoop.yarn.api.records.Priority,java.lang.String,java.lang.String) is not applicable
[ERROR]       (actual and formal argument lists differ in length)
[ERROR]
[ERROR] -> [Help 1]
[ERROR]
[ERROR] To see the full stack trace of the errors, re-run Maven with the -e switch.
[ERROR] Re-run Maven using the -X switch to enable full debug logging.
[ERROR]
[ERROR] For more information about the errors and possible solutions, please read the following articles:
[ERROR] [Help 1] http://cwiki.apache.org/confluence/display/MAVEN/MojoFailureException
[ERROR]
[ERROR] After correcting the problems, you can resume the build with the command
[ERROR]   mvn <goals> -rf :flink-yarn_2.11

我们可以看到flink-yarn的测试项目发生了异常，在这个版本中org.apache.hadoop.yarn.api.records.ApplicationReport.newInstance方法是不可用的，因此我们需要在flink-yarn模块下的pom文件的build中添加如下插件，跳过本模块的测试代码的编译。这里在YARN集群运行时不会受到影响。

    <plugin>
        <groupId>org.apache.maven.plugins</groupId>
        <artifactId>maven-compiler-plugin</artifactId>
        <version>3.8.0</version>
        <configuration>
            <source>${java.version}</source>
            <target>${java.version}</target>
            <!-- 略过测试代码的编译 -->
            <skip>true</skip>
            <!-- The semantics of this option are reversed, see MCOMPILER-209. -->
            <useIncrementalCompilation>false</useIncrementalCompilation>
            <compilerArgs>
                 <!-- Prevents recompilation due to missing package-info.class, see MCOMPILER-205 -->
                <arg>-Xpkginfo:always</arg>
            </compilerArgs>
        </configuration>
    </plugin>

2.6.3 Blink

在2.4步骤中直接切换到origin/blink分支，然后同样的方式执行如下命令进行编译：

mvn -T2C clean package -DskipTests -Dfast -Pinclude-hadoop -Pvendor-repos -Dhadoop.version=3.1.2

2.7 复制分发包到安装节点

2.7.1 编译成功后提示如下

[INFO] ------------------------------------------------------------------------
[INFO] Reactor Summary:
[INFO]
[INFO] force-shading 1.8.1 ................................ SUCCESS [  1.694 s]
[INFO] flink 1.8.1 ........................................ SUCCESS [  2.068 s]
[INFO] flink-annotations 1.8.1 ............................ SUCCESS [  2.552 s]
[INFO] flink-shaded-hadoop 1.8.1 .......................... SUCCESS [  1.085 s]
[INFO] flink-shaded-hadoop2 3.0.0-cdh6.3.0-1.8.1 .......... SUCCESS [ 29.178 s]
[INFO] flink-shaded-hadoop2-uber 3.0.0-cdh6.3.0-1.8.1 ..... SUCCESS [ 58.203 s]
[INFO] flink-shaded-yarn-tests 1.8.1 ...................... SUCCESS [ 41.005 s]
[INFO] flink-shaded-curator 1.8.1 ......................... SUCCESS [  4.035 s]
[INFO] flink-metrics 1.8.1 ................................ SUCCESS [  0.781 s]
[INFO] flink-metrics-core 1.8.1 ........................... SUCCESS [  3.267 s]
[INFO] flink-test-utils-parent 1.8.1 ...................... SUCCESS [  0.831 s]
[INFO] flink-test-utils-junit 1.8.1 ....................... SUCCESS [  2.805 s]
[INFO] flink-core 1.8.1 ................................... SUCCESS [ 16.670 s]
[INFO] flink-java 1.8.1 ................................... SUCCESS [  5.010 s]
[INFO] flink-queryable-state 1.8.1 ........................ SUCCESS [  0.556 s]
[INFO] flink-queryable-state-client-java 1.8.1 ............ SUCCESS [  1.204 s]
[INFO] flink-filesystems 1.8.1 ............................ SUCCESS [  0.613 s]
[INFO] flink-hadoop-fs 1.8.1 .............................. SUCCESS [  8.472 s]
[INFO] flink-runtime 1.8.1 ................................ SUCCESS [01:15 min]
[INFO] flink-scala 1.8.1 .................................. SUCCESS [01:25 min]
[INFO] flink-mapr-fs 1.8.1 ................................ SUCCESS [  5.098 s]
[INFO] flink-filesystems :: flink-fs-hadoop-shaded 1.8.1 .. SUCCESS [  7.291 s]
[INFO] flink-s3-fs-base 1.8.1 ............................. SUCCESS [ 17.262 s]
[INFO] flink-s3-fs-hadoop 1.8.1 ........................... SUCCESS [ 20.488 s]
[INFO] flink-s3-fs-presto 1.8.1 ........................... SUCCESS [ 31.440 s]
[INFO] flink-swift-fs-hadoop 1.8.1 ........................ SUCCESS [ 39.563 s]
[INFO] flink-oss-fs-hadoop 1.8.1 .......................... SUCCESS [ 15.569 s]
[INFO] flink-optimizer 1.8.1 .............................. SUCCESS [  2.131 s]
[INFO] flink-clients 1.8.1 ................................ SUCCESS [  2.022 s]
[INFO] flink-streaming-java 1.8.1 ......................... SUCCESS [  8.685 s]
[INFO] flink-test-utils 1.8.1 ............................. SUCCESS [  7.446 s]
[INFO] flink-runtime-web 1.8.1 ............................ SUCCESS [  5.302 s]
[INFO] flink-examples 1.8.1 ............................... SUCCESS [  0.421 s]
[INFO] flink-examples-batch 1.8.1 ......................... SUCCESS [ 43.929 s]
[INFO] flink-connectors 1.8.1 ............................. SUCCESS [  0.564 s]
[INFO] flink-hadoop-compatibility 1.8.1 ................... SUCCESS [ 16.530 s]
[INFO] flink-state-backends 1.8.1 ......................... SUCCESS [  0.823 s]
[INFO] flink-statebackend-rocksdb 1.8.1 ................... SUCCESS [  3.149 s]
[INFO] flink-tests 1.8.1 .................................. SUCCESS [ 57.885 s]
[INFO] flink-streaming-scala 1.8.1 ........................ SUCCESS [01:15 min]
[INFO] flink-table 1.8.1 .................................. SUCCESS [  0.781 s]
[INFO] flink-table-common 1.8.1 ........................... SUCCESS [  0.997 s]
[INFO] flink-table-api-java 1.8.1 ......................... SUCCESS [  0.406 s]
[INFO] flink-table-api-java-bridge 1.8.1 .................. SUCCESS [  1.527 s]
[INFO] flink-libraries 1.8.1 .............................. SUCCESS [  0.542 s]
[INFO] flink-cep 1.8.1 .................................... SUCCESS [  8.363 s]
[INFO] flink-table-planner 1.8.1 .......................... SUCCESS [04:18 min]
[INFO] flink-orc 1.8.1 .................................... SUCCESS [  8.790 s]
[INFO] flink-jdbc 1.8.1 ................................... SUCCESS [  5.401 s]
[INFO] flink-hbase 1.8.1 .................................. SUCCESS [ 45.025 s]
[INFO] flink-hcatalog 1.8.1 ............................... SUCCESS [ 20.901 s]
[INFO] flink-metrics-jmx 1.8.1 ............................ SUCCESS [  3.721 s]
[INFO] flink-connector-kafka-base 1.8.1 ................... SUCCESS [ 19.666 s]
[INFO] flink-connector-kafka-0.9 1.8.1 .................... SUCCESS [  7.921 s]
[INFO] flink-connector-kafka-0.10 1.8.1 ................... SUCCESS [  6.931 s]
[INFO] flink-connector-kafka-0.11 1.8.1 ................... SUCCESS [  5.260 s]
[INFO] flink-formats 1.8.1 ................................ SUCCESS [  0.639 s]
[INFO] flink-json 1.8.1 ................................... SUCCESS [  5.258 s]
[INFO] flink-connector-elasticsearch-base 1.8.1 ........... SUCCESS [ 11.332 s]
[INFO] flink-connector-elasticsearch 1.8.1 ................ SUCCESS [ 50.150 s]
[INFO] flink-connector-elasticsearch2 1.8.1 ............... SUCCESS [ 57.804 s]
[INFO] flink-connector-elasticsearch5 1.8.1 ............... SUCCESS [01:01 min]
[INFO] flink-connector-elasticsearch6 1.8.1 ............... SUCCESS [ 22.513 s]
[INFO] flink-connector-rabbitmq 1.8.1 ..................... SUCCESS [  1.885 s]
[INFO] flink-connector-twitter 1.8.1 ...................... SUCCESS [  4.839 s]
[INFO] flink-connector-nifi 1.8.1 ......................... SUCCESS [  2.129 s]
[INFO] flink-connector-cassandra 1.8.1 .................... SUCCESS [ 16.521 s]
[INFO] flink-avro 1.8.1 ................................... SUCCESS [ 10.093 s]
[INFO] flink-connector-filesystem 1.8.1 ................... SUCCESS [  6.873 s]
[INFO] flink-connector-kafka 1.8.1 ........................ SUCCESS [  8.335 s]
[INFO] flink-sql-connector-elasticsearch6 1.8.1 ........... SUCCESS [ 28.862 s]
[INFO] flink-sql-connector-kafka-0.9 1.8.1 ................ SUCCESS [  2.863 s]
[INFO] flink-sql-connector-kafka-0.10 1.8.1 ............... SUCCESS [  2.771 s]
[INFO] flink-sql-connector-kafka-0.11 1.8.1 ............... SUCCESS [  3.670 s]
[INFO] flink-sql-connector-kafka 1.8.1 .................... SUCCESS [  6.913 s]
[INFO] flink-connector-kafka-0.8 1.8.1 .................... SUCCESS [  5.247 s]
[INFO] flink-avro-confluent-registry 1.8.1 ................ SUCCESS [ 10.463 s]
[INFO] flink-parquet 1.8.1 ................................ SUCCESS [  5.704 s]
[INFO] flink-sequence-file 1.8.1 .......................... SUCCESS [  1.195 s]
[INFO] flink-csv 1.8.1 .................................... SUCCESS [  5.268 s]
[INFO] flink-examples-streaming 1.8.1 ..................... SUCCESS [ 48.896 s]
[INFO] flink-table-api-scala 1.8.1 ........................ SUCCESS [  0.343 s]
[INFO] flink-table-api-scala-bridge 1.8.1 ................. SUCCESS [  1.233 s]
[INFO] flink-examples-table 1.8.1 ......................... SUCCESS [ 47.093 s]
[INFO] flink-examples-build-helper 1.8.1 .................. SUCCESS [  0.287 s]
[INFO] flink-examples-streaming-twitter 1.8.1 ............. SUCCESS [  1.148 s]
[INFO] flink-examples-streaming-state-machine 1.8.1 ....... SUCCESS [  0.889 s]
[INFO] flink-container 1.8.1 .............................. SUCCESS [  1.882 s]
[INFO] flink-queryable-state-runtime 1.8.1 ................ SUCCESS [  4.483 s]
[INFO] flink-end-to-end-tests 1.8.1 ....................... SUCCESS [  0.829 s]
[INFO] flink-cli-test 1.8.1 ............................... SUCCESS [  1.406 s]
[INFO] flink-parent-child-classloading-test-program 1.8.1 . SUCCESS [  1.587 s]
[INFO] flink-parent-child-classloading-test-lib-package 1.8.1 SUCCESS [  1.895 s]
[INFO] flink-dataset-allround-test 1.8.1 .................. SUCCESS [  0.294 s]
[INFO] flink-datastream-allround-test 1.8.1 ............... SUCCESS [  5.243 s]
[INFO] flink-stream-sql-test 1.8.1 ........................ SUCCESS [  4.549 s]
[INFO] flink-bucketing-sink-test 1.8.1 .................... SUCCESS [ 12.400 s]
[INFO] flink-distributed-cache-via-blob 1.8.1 ............. SUCCESS [  1.512 s]
[INFO] flink-high-parallelism-iterations-test 1.8.1 ....... SUCCESS [ 10.873 s]
[INFO] flink-stream-stateful-job-upgrade-test 1.8.1 ....... SUCCESS [  5.087 s]
[INFO] flink-queryable-state-test 1.8.1 ................... SUCCESS [  3.373 s]
[INFO] flink-local-recovery-and-allocation-test 1.8.1 ..... SUCCESS [  0.536 s]
[INFO] flink-elasticsearch1-test 1.8.1 .................... SUCCESS [  6.442 s]
[INFO] flink-elasticsearch2-test 1.8.1 .................... SUCCESS [  5.664 s]
[INFO] flink-elasticsearch5-test 1.8.1 .................... SUCCESS [  6.762 s]
[INFO] flink-elasticsearch6-test 1.8.1 .................... SUCCESS [ 18.431 s]
[INFO] flink-quickstart 1.8.1 ............................. SUCCESS [  1.859 s]
[INFO] flink-quickstart-java 1.8.1 ........................ SUCCESS [  2.254 s]
[INFO] flink-quickstart-scala 1.8.1 ....................... SUCCESS [  2.254 s]
[INFO] flink-quickstart-test 1.8.1 ........................ SUCCESS [  0.616 s]
[INFO] flink-confluent-schema-registry 1.8.1 .............. SUCCESS [ 10.980 s]
[INFO] flink-stream-state-ttl-test 1.8.1 .................. SUCCESS [ 20.440 s]
[INFO] flink-sql-client-test 1.8.1 ........................ SUCCESS [  3.704 s]
[INFO] flink-streaming-file-sink-test 1.8.1 ............... SUCCESS [  1.489 s]
[INFO] flink-state-evolution-test 1.8.1 ................... SUCCESS [  3.776 s]
[INFO] flink-e2e-test-utils 1.8.1 ......................... SUCCESS [ 14.975 s]
[INFO] flink-streaming-python 1.8.1 ....................... SUCCESS [ 14.727 s]
[INFO] flink-mesos 1.8.1 .................................. SUCCESS [ 57.098 s]
[INFO] flink-yarn 1.8.1 ................................... SUCCESS [  4.752 s]
[INFO] flink-gelly 1.8.1 .................................. SUCCESS [  8.741 s]
[INFO] flink-gelly-scala 1.8.1 ............................ SUCCESS [ 36.059 s]
[INFO] flink-gelly-examples 1.8.1 ......................... SUCCESS [ 27.027 s]
[INFO] flink-metrics-dropwizard 1.8.1 ..................... SUCCESS [  0.581 s]
[INFO] flink-metrics-graphite 1.8.1 ....................... SUCCESS [  0.421 s]
[INFO] flink-metrics-influxdb 1.8.1 ....................... SUCCESS [  2.458 s]
[INFO] flink-metrics-prometheus 1.8.1 ..................... SUCCESS [  1.203 s]
[INFO] flink-metrics-statsd 1.8.1 ......................... SUCCESS [  0.516 s]
[INFO] flink-metrics-datadog 1.8.1 ........................ SUCCESS [  1.316 s]
[INFO] flink-metrics-slf4j 1.8.1 .......................... SUCCESS [  0.499 s]
[INFO] flink-python 1.8.1 ................................. SUCCESS [  4.480 s]
[INFO] flink-cep-scala 1.8.1 .............................. SUCCESS [ 48.001 s]
[INFO] flink-ml 1.8.1 ..................................... SUCCESS [01:33 min]
[INFO] flink-ml-uber 1.8.1 ................................ SUCCESS [  5.972 s]
[INFO] flink-table-uber 1.8.1 ............................. SUCCESS [  6.276 s]
[INFO] flink-sql-client 1.8.1 ............................. SUCCESS [ 14.243 s]
[INFO] flink-scala-shell 1.8.1 ............................ SUCCESS [01:00 min]
[INFO] flink-dist 1.8.1 ................................... SUCCESS [ 22.946 s]
[INFO] flink-end-to-end-tests-common 1.8.1 ................ SUCCESS [  1.660 s]
[INFO] flink-metrics-availability-test 1.8.1 .............. SUCCESS [  0.446 s]
[INFO] flink-metrics-reporter-prometheus-test 1.8.1 ....... SUCCESS [  0.415 s]
[INFO] flink-heavy-deployment-stress-test 1.8.1 ........... SUCCESS [ 39.937 s]
[INFO] flink-streaming-kafka-test-base 1.8.1 .............. SUCCESS [  1.775 s]
[INFO] flink-streaming-kafka-test 1.8.1 ................... SUCCESS [ 33.772 s]
[INFO] flink-streaming-kafka011-test 1.8.1 ................ SUCCESS [ 26.880 s]
[INFO] flink-streaming-kafka010-test 1.8.1 ................ SUCCESS [ 27.664 s]
[INFO] flink-contrib 1.8.1 ................................ SUCCESS [  0.619 s]
[INFO] flink-connector-wikiedits 1.8.1 .................... SUCCESS [  1.773 s]
[INFO] flink-yarn-tests 1.8.1 ............................. SUCCESS [01:13 min]
[INFO] flink-fs-tests 1.8.1 ............................... SUCCESS [  3.835 s]
[INFO] flink-docs 1.8.1 ................................... SUCCESS [  1.896 s]
[INFO] ------------------------------------------------------------------------
[INFO] BUILD SUCCESS
[INFO] ------------------------------------------------------------------------
[INFO] Total time:  13:18 min (Wall Clock)
[INFO] Finished at: 2019-07-28T11:13:20+08:00
[INFO] ------------------------------------------------------------------------

2.7.2 编译成功后目录如下

Su-yuexi:~ yoreyuan$# ls -l
total 832
lrwxrwxrwx  1 root root     61 Jul 28 11:04 build-target -> /opt/flink/flink-dist/target/flink-1.8.1-bin/flink-1.8.1
drwxr-xr-x 18 root root   4096 Jul 28 10:16 docs
drwxr-xr-x  4 root root   4096 Jul 28 10:54 flink-annotations
drwxr-xr-x  4 root root   4096 Jul 28 10:57 flink-clients
drwxr-xr-x 30 root root   4096 Jul 28 10:54 flink-connectors
drwxr-xr-x  6 root root   4096 Jul 28 10:58 flink-container
drwxr-xr-x  5 root root   4096 Jul 28 10:54 flink-contrib
drwxr-xr-x  4 root root   4096 Jul 28 10:54 flink-core
drwxr-xr-x  4 root root   4096 Jul 28 11:04 flink-dist
drwxr-xr-x  4 root root   4096 Jul 28 10:58 flink-docs
drwxr-xr-x 36 root root   4096 Jul 28 10:54 flink-end-to-end-tests
drwxr-xr-x  7 root root   4096 Jul 28 10:55 flink-examples
drwxr-xr-x 11 root root   4096 Jul 28 10:54 flink-filesystems
drwxr-xr-x  9 root root   4096 Jul 28 10:54 flink-formats
drwxr-xr-x  4 root root   4096 Jul 28 11:03 flink-fs-tests
drwxr-xr-x  4 root root   4096 Jul 28 10:55 flink-java
drwxr-xr-x  6 root root   4096 Jul 28 10:16 flink-jepsen
drwxr-xr-x 12 root root   4096 Jul 28 10:54 flink-libraries
drwxr-xr-x  4 root root   4096 Jul 28 10:58 flink-mesos
drwxr-xr-x 12 root root   4096 Jul 28 10:54 flink-metrics
drwxr-xr-x  4 root root   4096 Jul 28 10:57 flink-optimizer
drwxr-xr-x  5 root root   4096 Jul 28 10:54 flink-queryable-state
drwxr-xr-x  5 root root   4096 Jul 28 10:54 flink-quickstart
drwxr-xr-x  4 root root   4096 Jul 28 10:55 flink-runtime
drwxr-xr-x  5 root root   4096 Jul 28 10:58 flink-runtime-web
drwxr-xr-x  4 root root   4096 Jul 28 10:57 flink-scala
drwxr-xr-x  5 root root   4096 Jul 28 11:03 flink-scala-shell
drwxr-xr-x  4 root root   4096 Jul 28 10:54 flink-shaded-curator
drwxr-xr-x  6 root root   4096 Jul 28 10:54 flink-shaded-hadoop
drwxr-xr-x  4 root root   4096 Jul 28 10:54 flink-state-backends
drwxr-xr-x  4 root root   4096 Jul 28 10:57 flink-streaming-java
drwxr-xr-x  4 root root   4096 Jul 28 10:59 flink-streaming-scala
drwxr-xr-x 11 root root   4096 Jul 28 10:54 flink-table
drwxr-xr-x  4 root root   4096 Jul 28 10:59 flink-tests
drwxr-xr-x  5 root root   4096 Jul 28 10:54 flink-test-utils-parent
drwxr-xr-x  4 root root   4096 Jul 28 10:58 flink-yarn
drwxr-xr-x  4 root root   4096 Jul 28 11:04 flink-yarn-tests
-rw-r--r--  1 root root  11357 Jul 28 09:57 LICENSE
drwxr-xr-x  2 root root   4096 Jul 28 10:16 licenses
drwxr-xr-x  2 root root   4096 Jul 28 10:16 licenses-binary
-rw-r--r--  1 root root   2165 Jul 28 09:57 NOTICE
-rw-r--r--  1 root root 596009 Jul 28 10:16 NOTICE-binary
-rw-r--r--  1 root root  59071 Jul 28 10:16 pom.xml
-rw-r--r--  1 root root   4531 Jul 28 10:16 README.md
drwxr-xr-x  3 root root   4096 Jul 28 10:54 target
drwxr-xr-x  6 root root   4096 Jul 28 10:16 tools

2.7.3 Flink的发行包

由上面我们可以看到build-target软连指向的就是我们的部署文件，因为在编译的时候我们已经将Hadoop编译到这个部署文件中了，所以这里直接将编译后的发行包cp或者scp到安装节点即可。

# 以软连接方式创建到安装目录
cp -r  build-target ../opt/flink-1.8.1

# 或者，拷贝源分发文件到安装目录
cp -r /opt/flink/flink-dist/target/flink-1.8.1-bin/flink-1.8.1 /opt/flink-1.8.1

# 或者，scp到安装目录
scp -r /opt/flink/flink-dist/target/flink-1.8.1-bin/flink-1.8.1 root@chd3:/opt/flink-1.8.1

2.8 配置环境变量

vim ~/.bash_profile

添加如下配置，保存并退出，并生效. ~/.bash_profile

# Flink
export FLINK_HOME=/opt/flink-1.8.1
export PATH=$PATH:$FLINK_HOME/bin

2.9 standalone模式启动

2.9.1 启动

$FLINK_HOME/bin/start-cluster.sh

2.9.2 访问Flink UI

浏览器访问页面 http://localhost:8081/#/overview

2.9.3 测试

$FLINK_HOME/bin/flink run $FLINK_HOME/examples/batch/WordCount.jar

2.9.4 停止

 $FLINK_HOME/bin/stop-cluster.sh

2.10 YARN模式

YARN模式有两种方式：

yarn-session：在yarn中初始化一个flink集群，开辟指定的资源，以后提交任务都向这里提交。这个flink集群会常驻在yarn集群中，除非手工停止。
yarn-cluster(推荐)：每次提交都会创建一个新的flink集群，任务之间互相独立，互不影响，方便管理。任务执行完成之后创建的集群也会消失。

2.10.1 yarn-session模式

2.10.1.1 启动

先在YARN上申请Flink 集群资源：

# 查看参数的说明
$FLINK_HOME/bin/yarn-session.sh -h

#启动yarn-session，指定JobManage和TaskManager资源
$FLINK_HOME/bin/yarn-session.sh -jm 1024m -tm 4096m

2.10.1.2 访问Flink UI

首先访问YARN页面，然后可以看到启动了ApplicationMaster，点解可以看到Flink UI

2.10.1.3 测试

#现在一份测试数据
wget -O LICENSE-2.0.txt http://www.apache.org/licenses/LICENSE-2.0.txt
#上传到HDFS
hadoop fs -copyFromLocal LICENSE-2.0.txt hdfs:///home/flink
# Wordcount。在极少数问题的情况下，您还可以使用-m参数传递JobManager地址。JobManager地址在YARN控制台中可见。
$FLINK_HOME/bin/flink run $FLINK_HOME/examples/batch/WordCount.jar --input hdfs:///home/flink/LICENSE-2.0.txt --output hdfs:///home/flink/wordcount-result.txt

提交之后可以在YARN的ApplicationMaster打开的Flink UI连接看到运行的job状态，同时在hdfs:///home/flink/wordcount-result.txt可以看到运行的结果

2.10.1.4 关闭

如果关闭，可以使用YARN命令：

# 查看YARN上运行的Application
# yarn application -list
yarn app -list

# 关闭
yarn app -kill ${Application-Id}

2.10.2 yarn-cluster模式

2.10.2.1 启动并运行Job

$FLINK_HOME/bin/flink run -m yarn-cluster ./examples/batch/WordCount.jar

2.10.2.2 访问状态信息

这一步要快，提交之后可以在YARN的ApplicationMaster打开Flink UI，可以看到运行的job状态。当这个Job运行完毕之后，会马上释放资源，YARN页面上可以看到提交的Application已经FINISHED，Flink UI页面也会自动关闭。

2.11 Blink

将通过 2.6.3 步骤编译的发行包按照前面 Flink 方式配置和安装，注意端口，避免冲突。我这里修改了rest.port: 9081。运行一个

$BLINK_HOME/bin/flink run $BLINK_HOME/examples/batch/WordCount.jar

控制台会打印出结果，访问Flink UI如下图：

同样也可以yarn-session模式启动：

$BLINK_HOME/bin/yarn-session.sh -jm 1024 -tm 4096
#提交Wordcount
$BLINK_HOME/bin/flink run $BLINK_HOME/examples/batch/WordCount.jar --input hdfs:///home/flink/LICENSE-2.0.txt --output hdfs:///home/flink/wordcount-result2.txt

yarn-cluster模式提交Job

$BLINK_HOME/bin/flink run -m yarn-cluster ./examples/batch/WordCount.jar

YoreYuan

发布了40 篇原创文章 · 获赞 97 · 访问量 15万+

私信关注

从源码编译Apache Flink快速部署和测试(包含指定Hadoop及Cloudera Hadoop、Blink)

目录