java - flink : job won't run with higher taskmanager. heap.mb

Question

简单的工作:kafka->flatmap->reduce->map。

作业运行正常，默认值为 taskmanager.heap.mb (512Mb)。根据docs : 此值应尽可能大。由于有问题的机器有 96Gb 的 RAM，我将其设置为 75000(任意值)。

开始工作时出现此错误:

Caused by: org.apache.flink.runtime.client.JobExecutionException: Job execution failed.   
at org.apache.flink.runtime.jobmanager.JobManager$$anonfun$handleMessage$1$$anonfun$applyOrElse$5.apply$mcV$sp(JobManager.scala:563)   
at org.apache.flink.runtime.jobmanager.JobManager$$anonfun$handleMessage$1$$anonfun$applyOrElse$5.apply(JobManager.scala:509)
at org.apache.flink.runtime.jobmanager.JobManager$$anonfun$handleMessage$1$$anonfun$applyOrElse$5.apply(JobManager.scala:509)
at scala.concurrent.impl.Future$PromiseCompletingRunnable.liftedTree1$1(Future.scala:24)
at scala.concurrent.impl.Future$PromiseCompletingRunnable.run(Future.scala:24)
at akka.dispatch.TaskInvocation.run(AbstractDispatcher.scala:41)
at akka.dispatch.ForkJoinExecutorConfigurator$AkkaForkJoinTask.exec(AbstractDispatcher.scala:401)
at scala.concurrent.forkjoin.ForkJoinTask.doExec(ForkJoinTask.java:260)
at scala.concurrent.forkjoin.ForkJoinPool$WorkQueue.runTask(ForkJoinPool.java:1339)
at scala.concurrent.forkjoin.ForkJoinPool.runWorker(ForkJoinPool.java:1979)
at scala.concurrent.forkjoin.ForkJoinWorkerThread.run(ForkJoinWorkerThread.java:107)

Caused by: org.apache.flink.runtime.jobmanager.scheduler.NoResourceAvailableException: Not enough free slots available to run the job. You can decrease the operator parallelism or increase the number of slots per TaskManager in the configuration. Task to schedule: < Attempt #0 (Source: Custom Source (1/1)) @ (unassigned) - [SCHEDULED] > with groupID < 95b239d1777b2baf728645df9a1c4232 > in sharing group < SlotSharingGroup [772c9ff1cf0b6cb3a361e3352f75fcee, d4f856f13654f424d7c49d0f00f6ecca, 81bb8c4310faefe32f97ebd6baa4c04f, 95b239d1777b2baf728645df9a1c4232] >. Resources available to scheduler: Number of instances=0, total number of slots=0, available slots=0
at org.apache.flink.runtime.jobmanager.scheduler.Scheduler.scheduleTask(Scheduler.java:255)
at org.apache.flink.runtime.jobmanager.scheduler.Scheduler.scheduleImmediately(Scheduler.java:131)
at org.apache.flink.runtime.executiongraph.Execution.scheduleForExecution(Execution.java:298)
at org.apache.flink.runtime.executiongraph.ExecutionVertex.scheduleForExecution(ExecutionVertex.java:458)
at org.apache.flink.runtime.executiongraph.ExecutionJobVertex.scheduleAll(ExecutionJobVertex.java:322)
at org.apache.flink.runtime.executiongraph.ExecutionGraph.scheduleForExecution(ExecutionGraph.java:686)
at org.apache.flink.runtime.jobmanager.JobManager$$anonfun$org$apache$flink$runtime$jobmanager$JobManager$$submitJob$1.apply$mcV$sp(JobManager.scala:982)
at org.apache.flink.runtime.jobmanager.JobManager$$anonfun$org$apache$flink$runtime$jobmanager$JobManager$$submitJob$1.apply(JobManager.scala:962)
at org.apache.flink.runtime.jobmanager.JobManager$$anonfun$org$apache$flink$runtime$jobmanager$JobManager$$submitJob$1.apply(JobManager.scala:962)
... 8 more

将此参数恢复默认值(512)，作业运行正常。在 5000 时有效 -> 在 10000 时无效。

我错过了什么？

---

编辑:这比我想象的更成功。将值设置为 50000 并重新提交即可成功。在每次测试中，集群都会停止并重新启动。

Answer 1

您可能正在经历的是在 worker 向 master 注册之前提交工作。

一个 5GB 的 JVM 堆被快速初始化，TaskManager 几乎可以立即注册。对于 70GB 的堆，JVM 需要一段时间来初始化和引导。结果worker注册的比较晚，提交的job无法执行，因为缺少worker。

这也是为什么在您重新提交作业后它会起作用的原因。

如果您以“流”模式启动集群(通过 start-cluster-streaming.sh 独立运行)，JVM 的初始化速度会更快，因为至少 Flink 的内部内存会延迟初始化。