flink源码解析之任务执行流程

1. LocalStreamEnvironment#execute
2. MiniCluster#executeJobBlocking -> MiniCluster#submitJob
3. dispatcherGateway.submitJob(jobGraph, rpcTimeout)
4. Dispatcher#submitJob --> Dispatcher#internalSubmitJob --> Dispatcher#persistAndRunJob --> Dispatcher#runJob
5. Dispatcher#createJobManagerRunner
   –>jobManagerRunnerFactory.createJobManagerRunner
   –>this::startJobManagerRunner
6. JobManagerRunner#start 成功之后回调JobManagerRunner#grantLeadership
7. JobManagerRunner#verifyJobSchedulingStatusAndStartJobManager --> JobManagerRunner#startJobMaster
8. JobMasterService#start
9. JobMaster#start --> JobMaster#startJobExecution
   startJobMasterServices
   resetAndScheduleExecutionGraph
10. JobMaster#resetAndScheduleExecutionGraph --> JobMaster#scheduleExecutionGraph
11. ExecutionGraph#scheduleForExecution --> ExecutionGraph#scheduleEager
12. org.apache.flink.runtime.executiongraph.Execution#deploy
13. taskManagerGateway.submitTask(deployment, rpcTimeout) taskManagerGateway这里是RpcTaskManagerGateway#submitTask --> TaskExecutor#submitTask
14. task.startTaskThread();开始执行。在task线程的run方法中，通过
    invokable.invoke();反射的方式来执行任务。对于WordCount例子来说，invokable就是OneInputStreamTask，调用invoke的 时候，就会调用OneInputStreamTask的父类StreamTask的invoke方法。invoke方法中调用抽象方法run方法来执行，然后进入OneInputStreamTask#run

	@Override
	protected void run() throws Exception {
		// cache processor reference on the stack, to make the code more JIT friendly
		final StreamInputProcessor<IN> inputProcessor = this.inputProcessor;

		while (running && inputProcessor.processInput()) {
			// all the work happens in the "processInput" method
		}
	}

15. StreamInputProcessor#processInput，真正的执行代码在streamOperator.processElement(record);中，然后执行相应的子类的方法，对于StreamFlatMap来说，进入processElement方法，处理用户所写的代码逻辑。

@Override
	public void processElement(StreamRecord<IN> element) throws Exception {
		collector.setTimestamp(element);
		userFunction.flatMap(element.getValue(), collector);
	}
}