Flink源码——1、区分运行环境

Flink判断运行环境（本地、集群）的逻辑如下：

（1）在任务的main方法中，通过 StreamExecutionEnvironment 获取运行环境

StreamExecutionEnvironment environment = StreamExecutionEnvironment.getExecutionEnvironment();

（2）生成运行环境的工厂类放在ThreadLocal中；threadLocalContextEnvironmentFactory 是StreamExecutionEnvironment类的静态属性 

	/** The ThreadLocal used to store {@link StreamExecutionEnvironmentFactory}. */
	private static final ThreadLocal<StreamExecutionEnvironmentFactory> threadLocalContextEnvironmentFactory = new ThreadLocal<>();

①当是本地IDE直接运行任务main方法时，ThreadLocal中获取到的StreamExecutionEnvironmentFactory为空，此时生成本地运行环境LocalStreamEnvironment

	public static StreamExecutionEnvironment getExecutionEnvironment() {
		return Utils.resolveFactory(threadLocalContextEnvironmentFactory, contextEnvironmentFactory)
			.map(StreamExecutionEnvironmentFactory::createExecutionEnvironment)
			.orElseGet(StreamExecutionEnvironment::createLocalEnvironment);
	}

当ThreadLocal中有StreamExecutionEnvironmentFactory时，则用其createExecutionEnvironment()方法来生成运行环境

②当集群环境时，是如何将StreamExecutionEnvironmentFactory放入到ThreadLocal中？

通过 bin/flink run ....   命令提交jar包到集群运行命令时，该脚本会调用 org.apache.flink.client.cli.CliFrontend  来运行用户程序，如下：

.......
.......
# Add HADOOP_CLASSPATH to allow the usage of Hadoop file systems
exec $JAVA_RUN $JVM_ARGS $FLINK_ENV_JAVA_OPTS "${log_setting[@]}" -classpath "`manglePathList "$CC_CLASSPATH:$INTERNAL_HADOOP_CLASSPATHS"`" org.apache.flink.client.cli.CliFrontend "$@"

在CliFrontend中依次执行以下方法 main() ->  parseParameters() -> run() -> executeProgram() 

	protected void executeProgram(final Configuration configuration, final PackagedProgram program) throws ProgramInvocationException {
		ClientUtils.executeProgram(new DefaultExecutorServiceLoader(), configuration, program, false, false);
	}

在org.apache.flink.client.ClientUtils的executeProgram()中调用 StreamContextEnvironment.setAsContext(...)，StreamContextEnvironment继承自StreamExecutionEnvironment。setAsContext()代码如下

	public static void setAsContext(
			final PipelineExecutorServiceLoader executorServiceLoader,
			final Configuration configuration,
			final ClassLoader userCodeClassLoader,
			final boolean enforceSingleJobExecution,
			final boolean suppressSysout) {
		StreamExecutionEnvironmentFactory factory = () -> new StreamContextEnvironment(
			executorServiceLoader,
			configuration,
			userCodeClassLoader,
			enforceSingleJobExecution,
			suppressSysout);
		initializeContextEnvironment(factory);
	}

创建生成运行环境的工厂类实例，在initializeContextEnvironment()方法中把实例放到StreamExecutionEnvironment类的静态属性threadLocalContextEnvironmentFactory 中 ，代码如下

	protected static void initializeContextEnvironment(StreamExecutionEnvironmentFactory ctx) {
		contextEnvironmentFactory = ctx;
		threadLocalContextEnvironmentFactory.set(contextEnvironmentFactory);
	}

这样在用户程序 StreamExecutionEnvironment.getExecutionEnvironment() 时，获取到的运行环境就是StreamContextEnvironment类的setAsContext()方法中生成的

	public static void setAsContext(
			final PipelineExecutorServiceLoader executorServiceLoader,
			final Configuration configuration,
			final ClassLoader userCodeClassLoader,
			final boolean enforceSingleJobExecution,
			final boolean suppressSysout) {
		StreamExecutionEnvironmentFactory factory = () -> new StreamContextEnvironment(
			executorServiceLoader,
			configuration,
			userCodeClassLoader,
			enforceSingleJobExecution,
			suppressSysout);
		......
	}

本地运行环境LocalStreamEnvironment 和 独立集群、flink on yarn等运行环境StreamContextEnvironment 的主要区别在于，他们的成员属性 configuration 不同。LocalStreamEnvironment 中是创建的空键值对(new Configuration())，而StreamContextEnvironment 是通过 CliFrontend 生成的 Configuration 对象。