Flink taskmanager 在创建检查点/保存点失败后卡住（100% cpu 使用率）

Question

-- 通过将状态后端从文件系统更改为rocksdb来解决问题--

在 AWS EMR 上运行 Flink 1.9。 Flink 应用程序使用 kinesis 流作为输入数据，使用另一个 kinesis 流作为输出。 最近检查点大小已增长到 1 GB（由于更多数据）。 有时，在尝试获取检查点的过程中 - 应用程序开始利用整个处理器资源（一天发生几次）

指标：

LA (emr ec2 core node with job/task managers)

Run Loop Time - kinesis consumer

Records Per Fetch - kinesis consumer

Task manager GC

jobmanager 日志

{"level":"INFO","timestamp":"2020-08-25 04:55:27,399","thread":"Checkpoint Timer","file":"CheckpointCoordinator.java","line":"617","message":"Triggering checkpoint 1232 @ 1598331327244 for job 0039825bafae26bc34db88e037a1dae3."}

{"level":"INFO","timestamp":"2020-08-25 04:58:24,509","thread":"flink-akka.actor.default-dispatcher-7010","file":"ResourceManager.java","line":"1144","message":"The heartbeat of TaskManager with id container_1597960565773_0003_01_000002 timed out."}

{"level":"INFO","timestamp":"2020-08-25 04:58:24,510","thread":"flink-akka.actor.default-dispatcher-7010","file":"ResourceManager.java","line":"805","message":"Closing TaskExecutor connection container_1597960565773_0003_01_000002 because: The heartbeat of TaskManager with id container_1597960565773_0003_01_000002  timed out."}
{"level":"INFO","timestamp":"2020-08-25 04:58:24,514","thread":"flink-akka.actor.default-dispatcher-7015","file":"Execution.java","line":"1493","message":"Sink: kinesis-events-sink (1/1) (573401e241fe0a0ac0a8a54c81c4eefd) switched from RUNNING to FAILED."}
java.util.concurrent.TimeoutException: The heartbeat of TaskManager with id container_1597960565773_0003_01_000002  timed out.
        at org.apache.flink.runtime.resourcemanager.ResourceManager$TaskManagerHeartbeatListener.notifyHeartbeatTimeout(ResourceManager.java:1146)
        at org.apache.flink.runtime.heartbeat.HeartbeatMonitorImpl.run(HeartbeatMonitorImpl.java:109)
        at java.util.concurrent.Executors$RunnableAdapter.call(Executors.java:511)
        at java.util.concurrent.FutureTask.run(FutureTask.java:266)
        at org.apache.flink.runtime.rpc.akka.AkkaRpcActor.handleRunAsync(AkkaRpcActor.java:397)
        at org.apache.flink.runtime.rpc.akka.AkkaRpcActor.handleRpcMessage(AkkaRpcActor.java:190)
        at org.apache.flink.runtime.rpc.akka.FencedAkkaRpcActor.handleRpcMessage(FencedAkkaRpcActor.java:74)
        at org.apache.flink.runtime.rpc.akka.AkkaRpcActor.handleMessage(AkkaRpcActor.java:152)
        at akka.japi.pf.UnitCaseStatement.apply(CaseStatements.scala:26)
        at akka.japi.pf.UnitCaseStatement.apply(CaseStatements.scala:21)
        at scala.PartialFunction$class.applyOrElse(PartialFunction.scala:123)
        at akka.japi.pf.UnitCaseStatement.applyOrElse(CaseStatements.scala:21)
        at scala.PartialFunction$OrElse.applyOrElse(PartialFunction.scala:170)
        at scala.PartialFunction$OrElse.applyOrElse(PartialFunction.scala:171)
        at scala.PartialFunction$OrElse.applyOrElse(PartialFunction.scala:171)
        at akka.actor.Actor$class.aroundReceive(Actor.scala:517)
        at akka.actor.AbstractActor.aroundReceive(AbstractActor.scala:225)
        at akka.actor.ActorCell.receiveMessage(ActorCell.scala:592)
        at akka.actor.ActorCell.invoke(ActorCell.scala:561)
        at akka.dispatch.Mailbox.processMailbox(Mailbox.scala:258)
        at akka.dispatch.Mailbox.run(Mailbox.scala:225)
        at akka.dispatch.Mailbox.exec(Mailbox.scala:235)
        at akka.dispatch.forkjoin.ForkJoinTask.doExec(ForkJoinTask.java:260)
        at akka.dispatch.forkjoin.ForkJoinPool$WorkQueue.runTask(ForkJoinPool.java:1339)
        at akka.dispatch.forkjoin.ForkJoinPool.runWorker(ForkJoinPool.java:1979)
        at akka.dispatch.forkjoin.ForkJoinWorkerThread.run(ForkJoinWorkerThread.java:107)
{"level":"INFO","timestamp":"2020-08-25 04:58:24,515","thread":"flink-akka.actor.default-dispatcher-7015","file":"ExecutionGraph.java","line":"1324","message":"Job JOB_NAME_HIDDEN (0039825bafae26bc34db88e037a1dae3) switched from state RUNNING to FAILING."}
java.util.concurrent.TimeoutException: The heartbeat of TaskManager with id container_1597960565773_0003_01_000002  timed out.
        at org.apache.flink.runtime.resourcemanager.ResourceManager$TaskManagerHeartbeatListener.notifyHeartbeatTimeout(ResourceManager.java:1146)
        at org.apache.flink.runtime.heartbeat.HeartbeatMonitorImpl.run(HeartbeatMonitorImpl.java:109)
        at java.util.concurrent.Executors$RunnableAdapter.call(Executors.java:511)
        at java.util.concurrent.FutureTask.run(FutureTask.java:266)
        at org.apache.flink.runtime.rpc.akka.AkkaRpcActor.handleRunAsync(AkkaRpcActor.java:397)
        at org.apache.flink.runtime.rpc.akka.AkkaRpcActor.handleRpcMessage(AkkaRpcActor.java:190)
        at org.apache.flink.runtime.rpc.akka.FencedAkkaRpcActor.handleRpcMessage(FencedAkkaRpcActor.java:74)
        at org.apache.flink.runtime.rpc.akka.AkkaRpcActor.handleMessage(AkkaRpcActor.java:152)
        at akka.japi.pf.UnitCaseStatement.apply(CaseStatements.scala:26)
        at akka.japi.pf.UnitCaseStatement.apply(CaseStatements.scala:21)
        at scala.PartialFunction$class.applyOrElse(PartialFunction.scala:123)
        at akka.japi.pf.UnitCaseStatement.applyOrElse(CaseStatements.scala:21)
        at scala.PartialFunction$OrElse.applyOrElse(PartialFunction.scala:170)
        at scala.PartialFunction$OrElse.applyOrElse(PartialFunction.scala:171)
        at scala.PartialFunction$OrElse.applyOrElse(PartialFunction.scala:171)
        at akka.actor.Actor$class.aroundReceive(Actor.scala:517)
        at akka.actor.AbstractActor.aroundReceive(AbstractActor.scala:225)
        at akka.actor.ActorCell.receiveMessage(ActorCell.scala:592)
        at akka.actor.ActorCell.invoke(ActorCell.scala:561)
        at akka.dispatch.Mailbox.processMailbox(Mailbox.scala:258)
        at akka.dispatch.Mailbox.run(Mailbox.scala:225)
        at akka.dispatch.Mailbox.exec(Mailbox.scala:235)
        at akka.dispatch.forkjoin.ForkJoinTask.doExec(ForkJoinTask.java:260)
        at akka.dispatch.forkjoin.ForkJoinPool$WorkQueue.runTask(ForkJoinPool.java:1339)
        at akka.dispatch.forkjoin.ForkJoinPool.runWorker(ForkJoinPool.java:1979)
        at akka.dispatch.forkjoin.ForkJoinWorkerThread.run(ForkJoinWorkerThread.java:107)

任务管理器日志

{"level":"INFO","timestamp":"2020-08-25 04:58:17,504","thread":"kpl-daemon-0003","file":"LogInputStreamReader.java","line":"59","message":"[2020-08-25 04:58:02.469832] [0x00007394][0x00007f2673d70700] [info] [processing_statistics_logger.cc:127] (sensing-events-test) Aver
age Processing Time: -nan ms"}
{"level":"INFO","timestamp":"2020-08-25 04:58:17,504","thread":"kpl-daemon-0003","file":"LogInputStreamReader.java","line":"59","message":"[2020-08-25 04:58:17.469922] [0x00007394][0x00007f2673d70700] [info] [processing_statistics_logger.cc:109] Stage 1 Triggers: { stream
: 'sensing-events-test', manual: 0, count: 0, size: 0, matches: 0, timed: 0, UserRecords: 0, KinesisRecords: 0 }"}
{"level":"INFO","timestamp":"2020-08-25 04:58:17,504","thread":"kpl-daemon-0003","file":"LogInputStreamReader.java","line":"59","message":"[2020-08-25 04:58:17.469977] [0x00007394][0x00007f2673d70700] [info] [processing_statistics_logger.cc:112] Stage 2 Triggers: { stream
: 'sensing-events-test', manual: 0, count: 0, size: 0, matches: 0, timed: 0, KinesisRecords: 0, PutRecords: 0 }"}
{"level":"INFO","timestamp":"2020-08-25 04:58:17,504","thread":"kpl-daemon-0003","file":"LogInputStreamReader.java","line":"59","message":"[2020-08-25 04:58:17.469992] [0x00007394][0x00007f2673d70700] [info] [processing_statistics_logger.cc:127] (sensing-events-test) Aver
age Processing Time: -nan ms"}
{"level":"ERROR","timestamp":"2020-08-25 04:58:28,535","thread":"flink-akka.actor.default-dispatcher-628","file":"FatalExitExceptionHandler.java","line":"40","message":"FATAL: Thread 'flink-akka.actor.default-dispatcher-628' produced an uncaught exception. Stopping the pr
ocess..."}
java.lang.OutOfMemoryError: GC overhead limit exceeded
        at sun.reflect.UTF8.encode(UTF8.java:36)
        at sun.reflect.ClassFileAssembler.emitConstantPoolUTF8(ClassFileAssembler.java:103)
        at sun.reflect.MethodAccessorGenerator.generate(MethodAccessorGenerator.java:331)
        at sun.reflect.MethodAccessorGenerator.generateSerializationConstructor(MethodAccessorGenerator.java:112)
        at sun.reflect.ReflectionFactory.generateConstructor(ReflectionFactory.java:398)
        at sun.reflect.ReflectionFactory.newConstructorForSerialization(ReflectionFactory.java:360)
        at java.io.ObjectStreamClass.getSerializableConstructor(ObjectStreamClass.java:1588)
        at java.io.ObjectStreamClass.access$1500(ObjectStreamClass.java:79)
        at java.io.ObjectStreamClass$3.run(ObjectStreamClass.java:519)
        at java.io.ObjectStreamClass$3.run(ObjectStreamClass.java:494)
        at java.security.AccessController.doPrivileged(Native Method)
        at java.io.ObjectStreamClass.<init>(ObjectStreamClass.java:494)
        at java.io.ObjectStreamClass.lookup(ObjectStreamClass.java:391)
        at java.io.ObjectStreamClass.initNonProxy(ObjectStreamClass.java:681)
        at java.io.ObjectInputStream.readNonProxyDesc(ObjectInputStream.java:1941)
        at java.io.ObjectInputStream.readClassDesc(ObjectInputStream.java:1807)
        at java.io.ObjectInputStream.readClass(ObjectInputStream.java:1770)
        at java.io.ObjectInputStream.readObject0(ObjectInputStream.java:1595)
        at java.io.ObjectInputStream.readObject(ObjectInputStream.java:464)
        at java.io.ObjectInputStream.readObject(ObjectInputStream.java:422)
        at org.apache.flink.runtime.rpc.messages.RemoteRpcInvocation$MethodInvocation.readObject(RemoteRpcInvocation.java:204)
        at sun.reflect.NativeMethodAccessorImpl.invoke0(Native Method)
        at sun.reflect.NativeMethodAccessorImpl.invoke(NativeMethodAccessorImpl.java:62)
        at sun.reflect.DelegatingMethodAccessorImpl.invoke(DelegatingMethodAccessorImpl.java:43)
        at java.lang.reflect.Method.invoke(Method.java:498)
        at java.io.ObjectStreamClass.invokeReadObject(ObjectStreamClass.java:1184)
        at java.io.ObjectInputStream.readSerialData(ObjectInputStream.java:2234)
        at java.io.ObjectInputStream.readOrdinaryObject(ObjectInputStream.java:2125)
        at java.io.ObjectInputStream.readObject0(ObjectInputStream.java:1624)
        at java.io.ObjectInputStream.readObject(ObjectInputStream.java:464)
        at java.io.ObjectInputStream.readObject(ObjectInputStream.java:422)
        at org.apache.flink.util.InstantiationUtil.deserializeObject(InstantiationUtil.java:576)

更新。 flink-conf.yaml

state.backend.fs.checkpointdir: s3a://s3-bucket-with-checkpoints/flink-checkpoints
taskmanager.numberOfTaskSlots: 1
state.backend: filesystem
taskmanager.heap.size: 3057m
state.checkpoints.dir: s3a://s3-bucket-with-checkpoints/external-checkpoints

Flink Checkpoints

Answer 1

通过将状态后端从文件系统更改为rocksdb解决了问题

Answer 2

我认为，这可能与SlidingEventTimeWindow 有关，据我从检查点屏幕截图中了解，它是一个大小为 2 分钟的窗口，带有 2 秒的窗口幻灯片。 Flink 为每个元素所属的每个窗口创建一个副本。因此，对于滑动窗口，它会创建大约 60 个元素副本，因此状态大小是滚动窗口的 60 倍。

我猜，在检查点 flink 尝试序列化状态并且没有足够的内存，因此 GC 启动，最后你的内存用完了。