记一次flink不做checkpoint的问题

记一次flink不做checkpoint的问题

Posted by Sun.Zhu on April 13, 2020

问题现象:Flink UI界面查看checkpoint的metrics发现一直没有做checkpoint,仔细排查发现有部分subtask的状态是finished。 下图是测试环境复现问题 在这里插入图片描述 问题原因:仔细排查代码后发现source是消费kafka的数据,配置的并行度大于kafka的partition数,导致有部分subtask空闲,然后状态变为finished。后来查看了checkpoint过程的源码得以佐证。 在CheckpointCoordinator类的triggerCheckpoint方法中有如下代码段

// check if all tasks that we need to trigger are running.
		// if not, abort the checkpoint
		Execution[] executions = new Execution[tasksToTrigger.length];
		for (int i = 0; i < tasksToTrigger.length; i++) {
			Execution ee = tasksToTrigger[i].getCurrentExecutionAttempt();
			if (ee == null) {
				LOG.info("Checkpoint triggering task {} of job {} is not being executed at the moment. Aborting checkpoint.",
						tasksToTrigger[i].getTaskNameWithSubtaskIndex(),
						job);
				throw new CheckpointException(CheckpointFailureReason.NOT_ALL_REQUIRED_TASKS_RUNNING);
			} else if (ee.getState() == ExecutionState.RUNNING) {
				executions[i] = ee;
			} else {
				LOG.info("Checkpoint triggering task {} of job {} is not in state {} but {} instead. Aborting checkpoint.",
						tasksToTrigger[i].getTaskNameWithSubtaskIndex(),
						job,
						ExecutionState.RUNNING,
						ee.getState());
				throw new CheckpointException(CheckpointFailureReason.NOT_ALL_REQUIRED_TASKS_RUNNING);
			}

ee.getState() == ExecutionState.RUNNING判断execution的状态是否为running,否则不做checkpoint

问题结论:在消费kafka的数据时,source的并发度不能超过kafka的partition数,可以小于partition,但是部分subtask就会消费多个partition的数据,导致吞吐达不到最大,理想状态是source并发度等于partition数。