Apache flink PyFlink矢量化UDF抛出NullPointerException_Apache Flink_Flink Sql_Pyflink

Apache flink PyFlink矢量化UDF抛出NullPointerException

apache-flink

Apache flink PyFlink矢量化UDF抛出NullPointerException,apache-flink,flink-sql,pyflink,Apache Flink,Flink Sql,Pyflink,我有一个ML模型，它接受两个numpy.ndarray-用户和项目-并返回一个numpy.ndarray预测。在普通Python代码中，我将执行以下操作：模型=负载\模型 df=加载数据数据帧包括4列，即用户id、电影id、分级和时间戳用户=df.user\u id.values 项目=df.movie\u id.values 预测=模型用户、项目我正在考虑将此代码移植到Flink中，以利用其分布式特性。我的假设是：通过在多个Flink节点上分配预测工作负载，我应该能够更快地运行整个预测

我有一个ML模型，它接受两个numpy.ndarray-用户和项目-并返回一个numpy.ndarray预测。在普通Python代码中，我将执行以下操作：

模型=负载\模型 df=加载数据数据帧包括4列，即用户id、电影id、分级和时间戳用户=df.user\u id.values 项目=df.movie\u id.values 预测=模型用户、项目我正在考虑将此代码移植到Flink中，以利用其分布式特性。我的假设是：通过在多个Flink节点上分配预测工作负载，我应该能够更快地运行整个预测

所以我写了一份PyFlink的工作。注意：我实现了一个名为predict的UDF来运行预测

> python batch_prediction.py
1> +I(-2.3802385330200195)
1> +I(20.000154495239258)
1> +I(-30.704544067382812)
1> +I(-11.602800369262695)
1> +I(10.998968124389648)
...

batch_prediction.py 模型=负载\模型设置=EnvironmentSettings.new_instance.use_blink_planner.build exec_env=StreamExecutionEnvironment.get_execution_environment t_env=StreamTableEnvironment.createexec_env，环境设置=设置源_DDL= 创建表源用户id INT，电影(国际)，评级TINYINT，事件\u ms BIGINT 具有 “连接器”=“文件系统”， “格式”=“csv”， 'csv.字段分隔符'='\t'， “路径”=“ml-100k/u1.测试” 接收器DDL= 创建表接收器双倍预测具有 “连接器”=“打印” t_env.execute_sqlSOURCE_DDL t_env.execute_sqlSINK_DDL t_env.execute_sql 插入接收器，从源选择PREDICTuser\u id、movie\u id 等待这是UDF

批次预测表（续） @udfresult_type=DataTypes.DOUBLE def用户，项目：返回模型[用户]，[项目]。项目创建临时函数预测，预测工作进展顺利。但是，预测实际上是在源表的每一行上运行的，这是无法执行的。相反，我想将80000个用户id和电影id对分成100个批次，每个批次有800行。该作业触发modelusers，items函数100次=批处理，其中users和items都有800个元素

我找不到这样做的方法。通过查看，矢量化的用户定义函数可能会起作用

batch_prediction.py代码段我添加func_type=pandas @udfresult_type=DataTypes.DOUBLE，func_type=pandas def用户，项目： ... 不幸的是，事实并非如此

> python batch_prediction.py
...
Traceback (most recent call last):
  File "batch_prediction.py", line 55, in <module>
    "INSERT INTO sink SELECT PREDICT(user_id, movie_id) FROM source"
  File "/usr/local/anaconda3/envs/flink-ml/lib/python3.7/site-packages/pyflink/table/table_result.py", line 76, in wait
    get_method(self._j_table_result, "await")()
  File "/usr/local/anaconda3/envs/flink-ml/lib/python3.7/site-packages/py4j/java_gateway.py", line 1286, in __call__
    answer, self.gateway_client, self.target_id, self.name)
  File "/usr/local/anaconda3/envs/flink-ml/lib/python3.7/site-packages/pyflink/util/exceptions.py", line 147, in deco
    return f(*a, **kw)
  File "/usr/local/anaconda3/envs/flink-ml/lib/python3.7/site-packages/py4j/protocol.py", line 328, in get_return_value
    format(target_id, ".", name), value)
py4j.protocol.Py4JJavaError: An error occurred while calling o51.await.
: java.util.concurrent.ExecutionException: org.apache.flink.table.api.TableException: Failed to wait job finish
    at java.util.concurrent.CompletableFuture.reportGet(CompletableFuture.java:357)
    at java.util.concurrent.CompletableFuture.get(CompletableFuture.java:1908)
    at org.apache.flink.table.api.internal.TableResultImpl.awaitInternal(TableResultImpl.java:119)
    at org.apache.flink.table.api.internal.TableResultImpl.await(TableResultImpl.java:86)
    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 org.apache.flink.api.python.shaded.py4j.reflection.MethodInvoker.invoke(MethodInvoker.java:244)
    at org.apache.flink.api.python.shaded.py4j.reflection.ReflectionEngine.invoke(ReflectionEngine.java:357)
    at org.apache.flink.api.python.shaded.py4j.Gateway.invoke(Gateway.java:282)
    at org.apache.flink.api.python.shaded.py4j.commands.AbstractCommand.invokeMethod(AbstractCommand.java:132)
    at org.apache.flink.api.python.shaded.py4j.commands.CallCommand.execute(CallCommand.java:79)
    at org.apache.flink.api.python.shaded.py4j.GatewayConnection.run(GatewayConnection.java:238)
    at java.lang.Thread.run(Thread.java:748)
Caused by: org.apache.flink.table.api.TableException: Failed to wait job finish
    at org.apache.flink.table.api.internal.InsertResultIterator.hasNext(InsertResultIterator.java:59)
    at org.apache.flink.table.api.internal.TableResultImpl$CloseableRowIteratorWrapper.hasNext(TableResultImpl.java:355)
    at org.apache.flink.table.api.internal.TableResultImpl$CloseableRowIteratorWrapper.isFirstRowReady(TableResultImpl.java:368)
    at org.apache.flink.table.api.internal.TableResultImpl.lambda$awaitInternal$1(TableResultImpl.java:107)
    at java.util.concurrent.CompletableFuture$AsyncRun.run(CompletableFuture.java:1640)
    at java.util.concurrent.ThreadPoolExecutor.runWorker(ThreadPoolExecutor.java:1149)
    at java.util.concurrent.ThreadPoolExecutor$Worker.run(ThreadPoolExecutor.java:624)
    ... 1 more
Caused by: java.util.concurrent.ExecutionException: org.apache.flink.runtime.client.JobExecutionException: Job execution failed.
    at java.util.concurrent.CompletableFuture.reportGet(CompletableFuture.java:357)
    at java.util.concurrent.CompletableFuture.get(CompletableFuture.java:1908)
    at org.apache.flink.table.api.internal.InsertResultIterator.hasNext(InsertResultIterator.java:57)
    ... 7 more
Caused by: org.apache.flink.runtime.client.JobExecutionException: Job execution failed.
    at org.apache.flink.runtime.jobmaster.JobResult.toJobExecutionResult(JobResult.java:147)
    at org.apache.flink.runtime.minicluster.MiniClusterJobClient.lambda$getJobExecutionResult$2(MiniClusterJobClient.java:119)
    at java.util.concurrent.CompletableFuture.uniApply(CompletableFuture.java:616)
    at java.util.concurrent.CompletableFuture$UniApply.tryFire(CompletableFuture.java:591)
    at java.util.concurrent.CompletableFuture.postComplete(CompletableFuture.java:488)
    at java.util.concurrent.CompletableFuture.complete(CompletableFuture.java:1975)
    at org.apache.flink.runtime.rpc.akka.AkkaInvocationHandler.lambda$invokeRpc$0(AkkaInvocationHandler.java:229)
    at java.util.concurrent.CompletableFuture.uniWhenComplete(CompletableFuture.java:774)
    at java.util.concurrent.CompletableFuture$UniWhenComplete.tryFire(CompletableFuture.java:750)
    at java.util.concurrent.CompletableFuture.postComplete(CompletableFuture.java:488)
    at java.util.concurrent.CompletableFuture.complete(CompletableFuture.java:1975)
    at org.apache.flink.runtime.concurrent.FutureUtils$1.onComplete(FutureUtils.java:996)
    at akka.dispatch.OnComplete.internal(Future.scala:264)
    at akka.dispatch.OnComplete.internal(Future.scala:261)
    at akka.dispatch.japi$CallbackBridge.apply(Future.scala:191)
    at akka.dispatch.japi$CallbackBridge.apply(Future.scala:188)
    at scala.concurrent.impl.CallbackRunnable.run(Promise.scala:36)
    at org.apache.flink.runtime.concurrent.Executors$DirectExecutionContext.execute(Executors.java:74)
    at scala.concurrent.impl.CallbackRunnable.executeWithValue(Promise.scala:44)
    at scala.concurrent.impl.Promise$DefaultPromise.tryComplete(Promise.scala:252)
    at akka.pattern.PromiseActorRef.$bang(AskSupport.scala:572)
    at akka.pattern.PipeToSupport$PipeableFuture$$anonfun$pipeTo$1.applyOrElse(PipeToSupport.scala:22)
    at akka.pattern.PipeToSupport$PipeableFuture$$anonfun$pipeTo$1.applyOrElse(PipeToSupport.scala:21)
    at scala.concurrent.Future$$anonfun$andThen$1.apply(Future.scala:436)
    at scala.concurrent.Future$$anonfun$andThen$1.apply(Future.scala:435)
    at scala.concurrent.impl.CallbackRunnable.run(Promise.scala:36)
    at akka.dispatch.BatchingExecutor$AbstractBatch.processBatch(BatchingExecutor.scala:55)
    at akka.dispatch.BatchingExecutor$BlockableBatch$$anonfun$run$1.apply$mcV$sp(BatchingExecutor.scala:91)
    at akka.dispatch.BatchingExecutor$BlockableBatch$$anonfun$run$1.apply(BatchingExecutor.scala:91)
    at akka.dispatch.BatchingExecutor$BlockableBatch$$anonfun$run$1.apply(BatchingExecutor.scala:91)
    at scala.concurrent.BlockContext$.withBlockContext(BlockContext.scala:72)
    at akka.dispatch.BatchingExecutor$BlockableBatch.run(BatchingExecutor.scala:90)
    at akka.dispatch.TaskInvocation.run(AbstractDispatcher.scala:40)
    at akka.dispatch.ForkJoinExecutorConfigurator$AkkaForkJoinTask.exec(ForkJoinExecutorConfigurator.scala:44)
    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)
Caused by: org.apache.flink.runtime.JobException: Recovery is suppressed by NoRestartBackoffTimeStrategy
    at org.apache.flink.runtime.executiongraph.failover.flip1.ExecutionFailureHandler.handleFailure(ExecutionFailureHandler.java:116)
    at org.apache.flink.runtime.executiongraph.failover.flip1.ExecutionFailureHandler.getFailureHandlingResult(ExecutionFailureHandler.java:78)
    at org.apache.flink.runtime.scheduler.DefaultScheduler.handleTaskFailure(DefaultScheduler.java:224)
    at org.apache.flink.runtime.scheduler.DefaultScheduler.maybeHandleTaskFailure(DefaultScheduler.java:217)
    at org.apache.flink.runtime.scheduler.DefaultScheduler.updateTaskExecutionStateInternal(DefaultScheduler.java:208)
    at org.apache.flink.runtime.scheduler.SchedulerBase.updateTaskExecutionState(SchedulerBase.java:610)
    at org.apache.flink.runtime.scheduler.SchedulerNG.updateTaskExecutionState(SchedulerNG.java:89)
    at org.apache.flink.runtime.jobmaster.JobMaster.updateTaskExecutionState(JobMaster.java:419)
    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 org.apache.flink.runtime.rpc.akka.AkkaRpcActor.handleRpcInvocation(AkkaRpcActor.java:286)
    at org.apache.flink.runtime.rpc.akka.AkkaRpcActor.handleRpcMessage(AkkaRpcActor.java:201)
    at org.apache.flink.runtime.rpc.akka.FencedAkkaRpcActor.handleRpcMessage(FencedAkkaRpcActor.java:74)
    at org.apache.flink.runtime.rpc.akka.AkkaRpcActor.handleMessage(AkkaRpcActor.java:154)
    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)
    ... 4 more
Caused by: org.apache.flink.streaming.runtime.tasks.AsynchronousException: Caught exception while processing timer.
    at org.apache.flink.streaming.runtime.tasks.StreamTask$StreamTaskAsyncExceptionHandler.handleAsyncException(StreamTask.java:1108)
    at org.apache.flink.streaming.runtime.tasks.StreamTask.handleAsyncException(StreamTask.java:1082)
    at org.apache.flink.streaming.runtime.tasks.StreamTask.invokeProcessingTimeCallback(StreamTask.java:1213)
    at org.apache.flink.streaming.runtime.tasks.StreamTask.lambda$null$17(StreamTask.java:1202)
    at org.apache.flink.streaming.runtime.tasks.StreamTaskActionExecutor$SynchronizedStreamTaskActionExecutor.runThrowing(StreamTaskActionExecutor.java:92)
    at org.apache.flink.streaming.runtime.tasks.mailbox.Mail.run(Mail.java:78)
    at org.apache.flink.streaming.runtime.tasks.mailbox.MailboxExecutorImpl.tryYield(MailboxExecutorImpl.java:91)
    at org.apache.flink.streaming.runtime.tasks.StreamOperatorWrapper.quiesceTimeServiceAndCloseOperator(StreamOperatorWrapper.java:155)
    at org.apache.flink.streaming.runtime.tasks.StreamOperatorWrapper.close(StreamOperatorWrapper.java:130)
    at org.apache.flink.streaming.runtime.tasks.OperatorChain.closeOperators(OperatorChain.java:412)
    at org.apache.flink.streaming.runtime.tasks.StreamTask.afterInvoke(StreamTask.java:585)
    at org.apache.flink.streaming.runtime.tasks.StreamTask.invoke(StreamTask.java:547)
    at org.apache.flink.runtime.taskmanager.Task.doRun(Task.java:722)
    at org.apache.flink.runtime.taskmanager.Task.run(Task.java:547)
    at java.lang.Thread.run(Thread.java:748)
Caused by: TimerException{java.lang.RuntimeException: Failed to close remote bundle}
    ... 13 more
Caused by: java.lang.RuntimeException: Failed to close remote bundle
    at org.apache.flink.streaming.api.runners.python.beam.BeamPythonFunctionRunner.finishBundle(BeamPythonFunctionRunner.java:371)
    at org.apache.flink.streaming.api.runners.python.beam.BeamPythonFunctionRunner.flush(BeamPythonFunctionRunner.java:325)
    at org.apache.flink.streaming.api.operators.python.AbstractPythonFunctionOperator.invokeFinishBundle(AbstractPythonFunctionOperator.java:291)
    at org.apache.flink.table.runtime.operators.python.scalar.arrow.RowDataArrowPythonScalarFunctionOperator.invokeFinishBundle(RowDataArrowPythonScalarFunctionOperator.java:77)
    at org.apache.flink.streaming.api.operators.python.AbstractPythonFunctionOperator.checkInvokeFinishBundleByTime(AbstractPythonFunctionOperator.java:285)
    at org.apache.flink.streaming.api.operators.python.AbstractPythonFunctionOperator.lambda$open$0(AbstractPythonFunctionOperator.java:134)
    at org.apache.flink.streaming.runtime.tasks.StreamTask.invokeProcessingTimeCallback(StreamTask.java:1211)
    ... 12 more
Caused by: java.lang.NullPointerException
    at org.apache.flink.streaming.api.runners.python.beam.BeamPythonFunctionRunner.finishBundle(BeamPythonFunctionRunner.java:369)
    ... 18 more

阿帕奇·弗林克社区对典福的信任。看

对于Pandas UDF，每个输入参数的输入类型为Pandas.Series，结果类型也应为Pandas.Series。此外，结果的长度应与输入的长度相同。您能检查一下您的熊猫UDF实施是否如此吗

然后我决定为我的UDF添加一个pytest单元测试来验证输入和输出类型。以下是如何：

作为pd进口熊猫从udf_def导入预测 def测试预测： f=预测。_func 用户=pd.系列[1,2,3] 项目=pd.系列[1,4,9] preds=熔断器、项目断言isinstancepreds，pd.系列断言lenpreds==3 它揭示了实现错误。然后修复我的UDF实现以通过单元测试

从model_def导入矩阵工厂化 @udfresult_type=DataTypes.DOUBLE，func_type=pandas 定义用户、项目： n_用户，n_项目=9431682 模型=矩阵分解n个用户，n个项目 model.load\u state\u.loadmodel.pth 返回pd.Seriesmodelusers，items.detach.numpy 使用新的UDF实现，我能够运行批处理预测

> python batch_prediction.py
1> +I(-2.3802385330200195)
1> +I(20.000154495239258)
1> +I(-30.704544067382812)
1> +I(-11.602800369262695)
1> +I(10.998968124389648)
...

可以找到更新的源代码