books/flink-in-action-8.1.md
Flink 相关的组件和作业的稳定性通常是比较关键的，所以得需要对它们进行监控，如果有异常，则需要及时告警通知。本章先会教会教会大家如何利用现有 Flink UI 上面的信息去发现和排查问题，会指明一些比较重要和我们非常关心的指标，通过这些指标我们能够立马定位到问题的根本原因。接着笔者会教大家如何去利用现有的 Metrics Reporter 去构建一个 Flink 的监控系统，它可以收集到所有作业的监控指标，并会存储这些监控指标数据，最后还会有一个监控大盘做数据可视化，通过这个大盘可以方便排查问题。
当将 Flink JobManager、TaskManager 都运行起来了，并且也部署了不少 Flink Job，那么它到底是否还在运行、运行的状态如何、资源 TaskManager 和 Slot 的个数是否足够、Job 内部是否出现异常、计算速度是否跟得上数据生产的速度 等这些问题其实对我们来说是比较关注的，所以就很迫切的需要一个监控系统帮我们把整个 Flink 集群的运行状态给展示出来。通过监控系统我们能够很好的知道 Flink 内部的整个运行状态，然后才能够根据项目生产环境遇到的问题 ‘对症下药’。下面分别来讲下 JobManager、TaskManager、Flink Job 的监控以及最关心的一些监控指标。
我们知道 JobManager 是 Flink 集群的中控节点，类似于 Apache Storm 的 Nimbus 以及 Apache Spark 的 Driver 的角色。它负责作业的调度、作业 Jar 包的管理、Checkpoint 的协调和发起、与 TaskManager 之间的心跳检查等工作。如果 JobManager 出现问题的话，就会导致作业 UI 信息查看不了，TaskManager 和所有运行的作业都会受到一定的影响，所以这也是为啥在 7.1 节中强调 JobManager 的高可用问题。
在 Flink 自带的 UI 上 JobManager 那个 Tab 展示的其实并没有显示其对应的 Metrics，那么对于 JobManager 来说常见比较关心的监控指标有哪些呢？
因为 Flink JobManager 其实也是一个 Java 的应用程序，那么它自然也会有 Java 应用程序的指标，比如内存、CPU、GC、类加载、线程信息等。
jobmanager_Status_JVM_Memory_Direct_Count
jobmanager_Status_JVM_Memory_Direct_MemoryUsed
jobmanager_Status_JVM_Memory_Direct_TotalCapacity
jobmanager_Status_JVM_Memory_Heap_Committed
jobmanager_Status_JVM_Memory_Heap_Max
jobmanager_Status_JVM_Memory_Heap_Used
jobmanager_Status_JVM_Memory_Mapped_Count
jobmanager_Status_JVM_Memory_Mapped_MemoryUsed
jobmanager_Status_JVM_Memory_Mapped_TotalCapacity
jobmanager_Status_JVM_Memory_NonHeap_Committed
jobmanager_Status_JVM_Memory_NonHeap_Max
jobmanager_Status_JVM_Memory_NonHeap_Used
jobmanager_Status_JVM_CPU_Load
jobmanager_Status_JVM_CPU_Time
jobmanager_Status_JVM_GarbageCollector_PS_MarkSweep_Count
jobmanager_Status_JVM_GarbageCollector_PS_MarkSweep_Time
jobmanager_Status_JVM_GarbageCollector_PS_Scavenge_Count
jobmanager_Status_JVM_GarbageCollector_PS_Scavenge_Time
因为 JobManager 负责了作业的 Checkpoint 的协调和发起功能，所以 Checkpoint 相关的指标就有表示 Checkpoint 执行的时间、Checkpoint 的时间长短、完成的 Checkpoint 的次数、Checkpoint 失败的次数、Checkpoint 正在执行 Checkpoint 的个数等，其对应的指标如下：
jobmanager_job_lastCheckpointAlignmentBuffered
jobmanager_job_lastCheckpointDuration
jobmanager_job_lastCheckpointExternalPath
jobmanager_job_lastCheckpointRestoreTimestamp
jobmanager_job_lastCheckpointSize
jobmanager_job_numberOfCompletedCheckpoints
jobmanager_job_numberOfFailedCheckpoints
jobmanager_job_numberOfInProgressCheckpoints
jobmanager_job_totalNumberOfCheckpoints
另外还有比较重要的指标就是 Flink UI 上也提供的，类似于 Slot 总共个数、Slot 可使用的个数、TaskManager 的个数（通过查看该值可以知道是否有 TaskManager 发生异常重启）、正在运行的作业数量、作业运行的时间和完成的时间、作业的重启次数，对应的指标如下：
jobmanager_job_uptime
jobmanager_numRegisteredTaskManagers
jobmanager_numRunningJobs
jobmanager_taskSlotsAvailable
jobmanager_taskSlotsTotal
jobmanager_job_downtime
jobmanager_job_fullRestarts
jobmanager_job_restartingTime
....
taskmanager_Status_JVM_CPU_Load
taskmanager_Status_JVM_CPU_Time
taskmanager_Status_JVM_ClassLoader_ClassesLoaded
taskmanager_Status_JVM_ClassLoader_ClassesUnloaded
taskmanager_Status_JVM_GarbageCollector_G1_Old_Generation_Count
taskmanager_Status_JVM_GarbageCollector_G1_Old_Generation_Time
taskmanager_Status_JVM_GarbageCollector_G1_Young_Generation_Count
taskmanager_Status_JVM_GarbageCollector_G1_Young_Generation_Time
taskmanager_Status_JVM_Memory_Direct_Count
taskmanager_Status_JVM_Memory_Direct_MemoryUsed
taskmanager_Status_JVM_Memory_Direct_TotalCapacity
taskmanager_Status_JVM_Memory_Heap_Committed
taskmanager_Status_JVM_Memory_Heap_Max
taskmanager_Status_JVM_Memory_Heap_Used
taskmanager_Status_JVM_Memory_Mapped_Count
taskmanager_Status_JVM_Memory_Mapped_MemoryUsed
taskmanager_Status_JVM_Memory_Mapped_TotalCapacity
taskmanager_Status_JVM_Memory_NonHeap_Committed
taskmanager_Status_JVM_Memory_NonHeap_Max
taskmanager_Status_JVM_Memory_NonHeap_Used
taskmanager_Status_JVM_Threads_Count
taskmanager_Status_Network_AvailableMemorySegments
taskmanager_Status_Network_TotalMemorySegments
taskmanager_Status_Shuffle_Netty_AvailableMemorySegments
taskmanager_Status_Shuffle_Netty_TotalMemorySegments
...
taskmanager_job_task_Shuffle_Netty_Input_Buffers_outPoolUsage
taskmanager_job_task_Shuffle_Netty_Input_Buffers_outputQueueLength
taskmanager_job_task_Shuffle_Netty_Output_Buffers_inPoolUsage
taskmanager_job_task_Shuffle_Netty_Output_Buffers_inputExclusiveBuffersUsage
taskmanager_job_task_Shuffle_Netty_Output_Buffers_inputFloatingBuffersUsage
taskmanager_job_task_Shuffle_Netty_Output_Buffers_inputQueueLength
taskmanager_job_task_Shuffle_Netty_Output_numBuffersInLocal
taskmanager_job_task_Shuffle_Netty_Output_numBuffersInLocalPerSecond
taskmanager_job_task_Shuffle_Netty_Output_numBuffersInRemote
taskmanager_job_task_Shuffle_Netty_Output_numBuffersInRemotePerSecond
taskmanager_job_task_Shuffle_Netty_Output_numBytesInLocal
taskmanager_job_task_Shuffle_Netty_Output_numBytesInLocalPerSecond
taskmanager_job_task_Shuffle_Netty_Output_numBytesInRemote
taskmanager_job_task_Shuffle_Netty_Output_numBytesInRemotePerSecond
taskmanager_job_task_buffers_inPoolUsage
taskmanager_job_task_buffers_inputExclusiveBuffersUsage
taskmanager_job_task_buffers_inputFloatingBuffersUsage
taskmanager_job_task_buffers_inputQueueLength
taskmanager_job_task_buffers_outPoolUsage
taskmanager_job_task_buffers_outputQueueLength
taskmanager_job_task_checkpointAlignmentTime
taskmanager_job_task_currentInputWatermark
taskmanager_job_task_numBuffersInLocal
taskmanager_job_task_numBuffersInLocalPerSecond
taskmanager_job_task_numBuffersInRemote
taskmanager_job_task_numBuffersInRemotePerSecond
taskmanager_job_task_numBuffersOut
taskmanager_job_task_numBuffersOutPerSecond
taskmanager_job_task_numBytesIn
taskmanager_job_task_numBytesInLocal
taskmanager_job_task_numBytesInLocalPerSecond
taskmanager_job_task_numBytesInPerSecond
taskmanager_job_task_numBytesInRemote
taskmanager_job_task_numBytesInRemotePerSecond
taskmanager_job_task_numBytesOut
taskmanager_job_task_numBytesOutPerSecond
taskmanager_job_task_numRecordsIn
taskmanager_job_task_numRecordsInPerSecond
taskmanager_job_task_numRecordsOut
taskmanager_job_task_numRecordsOutPerSecond
taskmanager_job_task_operator_currentInputWatermark
taskmanager_job_task_operator_currentOutputWatermark
taskmanager_job_task_operator_numLateRecordsDropped
taskmanager_job_task_operator_numRecordsIn
taskmanager_job_task_operator_numRecordsInPerSecond
taskmanager_job_task_operator_numRecordsOut
taskmanager_job_task_operator_numRecordsOutPerSecond
加入知识星球可以看到上面文章：https://t.zsxq.com/f66iAMz