Spark 中 BroadCast 导致的内存溢出(SparkFatalException)

背景

本文基于

Spark 3.1.1    
open-jdk-1.8.0.352

目前在排查 Spark 任务的时候，遇到了一个很奇怪的问题，在此记录一下。

现象描述

一个 Spark Application, Driver端的内存为 5GB,一直以来都是能正常调度运行，突然有一天，报错了：

Caused by: org.apache.spark.sql.catalyst.errors.package$TreeNodeException: execute, tree:
Exchange hashpartitioning(user_lable_id#530L, 500), ENSURE_REQUIREMENTS, [id=#1564]
+- *(16) Project [xxx]
   +- *(16) BroadcastHashJoin 
      ...
            +- *(14) ColumnarToRow
             +- FileScan parquet xxx

	at org.apache.spark.sql.catalyst.errors.package$.attachTree(package.scala:56)
	at org.apache.spark.sql.execution.exchange.ShuffleExchangeExec.doExecute(ShuffleExchangeExec.scala:169)
	at org.apache.spark.sql.execution.SparkPlan.$anonfun$execute$1(SparkPlan.scala:180)
	at org.apache.spark.sql.execution.SparkPlan.$anonfun$executeQuery$1(SparkPlan.scala:218)
	at org.apache.spark.rdd.RDDOperationScope$.withScope(RDDOperationScope.scala:151)
	at org.apache.spark.sql.execution.SparkPlan.executeQuery(SparkPlan.scala:215)
	at org.apache.spark.sql.execution.SparkPlan.execute(SparkPlan.scala:176)
	at org.apache.spark.sql.execution.InputAdapter.inputRDD(WholeStageCodegenExec.scala:525)
	at org.apache.spark.sql.execution.InputRDDCodegen.inputRDDs(WholeStageCodegenExec.scala:453)
	at org.apache.spark.sql.execution.InputRDDCodegen.inputRDDs$(WholeStageCodegenExec.scala:452)
	at org.apache.spark.sql.execution.InputAdapter.inputRDDs(WholeStageCodegenExec.scala:496)
	at org.apache.spark.sql.execution.SortExec.inputRDDs(SortExec.scala:132)
	at org.apache.spark.sql.execution.WholeStageCodegenExec.doExecute(WholeStageCodegenExec.scala:746)
	at org.apache.spark.sql.execution.SparkPlan.$anonfun$execute$1(SparkPlan.scala:180)
	at org.apache.spark.sql.execution.SparkPlan.$anonfun$executeQuery$1(SparkPlan.scala:218)
	at org.apache.spark.rdd.RDDOperationScope$.withScope(RDDOperationScope.scala:151)
	at org.apache.spark.sql.execution.SparkPlan.executeQuery(SparkPlan.scala:215)
	at org.apache.spark.sql.execution.SparkPlan.execute(SparkPlan.scala:176)
	at org.apache.spark.sql.execution.InputAdapter.doExecute(WholeStageCodegenExec.scala:511)
	at org.apache.spark.sql.execution.SparkPlan.$anonfun$execute$1(SparkPlan.scala:180)
	at org.apache.spark.sql.execution.SparkPlan.$anonfun$executeQuery$1(SparkPlan.scala:218)
	at org.apache.spark.rdd.RDDOperationScope$.withScope(RDDOperationScope.scala:151)
	at org.apache.spark.sql.execution.SparkPlan.executeQuery(SparkPlan.scala:215)
	at org.apache.spark.sql.execution.SparkPlan.execute(SparkPlan.scala:176)
	at org.apache.spark.sql.execution.joins.SortMergeJoinExec.inputRDDs(SortMergeJoinExec.scala:378)
	at org.apache.spark.sql.execution.ProjectExec.inputRDDs(basicPhysicalOperators.scala:50)
	at org.apache.spark.sql.execution.WholeStageCodegenExec.doExecute(WholeStageCodegenExec.scala:746)
	at org.apache.spark.sql.execution.SparkPlan.$anonfun$execute$1(SparkPlan.scala:180)
	at org.apache.spark.sql.execution.SparkPlan.$anonfun$executeQuery$1(SparkPlan.scala:218)
	at org.apache.spark.rdd.RDDOperationScope$.withScope(RDDOperationScope.scala:151)
	at org.apache.spark.sql.execution.SparkPlan.executeQuery(SparkPlan.scala:215)
	at org.apache.spark.sql.execution.SparkPlan.execute(SparkPlan.scala:176)
	at org.apache.spark.sql.execution.exchange.ShuffleExchangeExec.inputRDD$lzycompute(ShuffleExchangeExec.scala:123)
	at org.apache.spark.sql.execution.exchange.ShuffleExchangeExec.inputRDD(ShuffleExchangeExec.scala:123)
	at org.apache.spark.sql.execution.exchange.ShuffleExchangeExec.shuffleDependency$lzycompute(ShuffleExchangeExec.scala:157)
	at org.apache.spark.sql.execution.exchange.ShuffleExchangeExec.shuffleDependency(ShuffleExchangeExec.scala:155)
	at org.apache.spark.sql.execution.exchange.ShuffleExchangeExec.$anonfun$doExecute$1(ShuffleExchangeExec.scala:172)
	at org.apache.spark.sql.catalyst.errors.package$.attachTree(package.scala:52)
	... 291 more
Caused by: java.util.concurrent.ExecutionException: org.apache.spark.util.SparkFatalException
	at java.util.concurrent.FutureTask.report(FutureTask.java:122)
	at java.util.concurrent.FutureTask.get(FutureTask.java:206)
	at org.apache.spark.sql.execution.exchange.BroadcastExchangeExec.doExecuteBroadcast(BroadcastExchangeExec.scala:199)
	at org.apache.spark.sql.execution.InputAdapter.doExecuteBroadcast(WholeStageCodegenExec.scala:515)
	at org.apache.spark.sql.execution.SparkPlan.$anonfun$executeBroadcast$1(SparkPlan.scala:193)
	at org.apache.spark.sql.execution.SparkPlan.$anonfun$executeQuery$1(SparkPlan.scala:218)
	at org.apache.spark.rdd.RDDOperationScope$.withScope(RDDOperationScope.scala:151)
	at org.apache.spark.sql.execution.SparkPlan.executeQuery(SparkPlan.scala:215)
	at org.apache.spark.sql.execution.SparkPlan.executeBroadcast(SparkPlan.scala:189)
	at org.apache.spark.sql.execution.joins.BroadcastHashJoinExec.prepareBroadcast(BroadcastHashJoinExec.scala:203)
	at org.apache.spark.sql.execution.joins.BroadcastHashJoinExec.prepareRelation(BroadcastHashJoinExec.scala:217)
	at org.apache.spark.sql.execution.joins.HashJoin.codegenOuter(HashJoin.scala:497)
	at org.apache.spark.sql.execution.joins.HashJoin.codegenOuter$(HashJoin.scala:496)
	at org.apache.spark.sql.execution.joins.BroadcastHashJoinExec.codegenOuter(BroadcastHashJoinExec.scala:40)
	at org.apache.spark.sql.execution.joins.HashJoin.doConsume(HashJoin.scala:352)
	at org.apache.spark.sql.execution.joins.HashJoin.doConsume$(HashJoin.scala:349)
	at org.apache.spark.sql.execution.joins.BroadcastHashJoinExec.doConsume(BroadcastHashJoinExec.scala:40)
	at org.apache.spark.sql.execution.CodegenSupport.consume(WholeStageCodegenExec.scala:194)
	at org.apache.spark.sql.execution.CodegenSupport.consume$(WholeStageCodegenExec.scala:149)
	at org.apache.spark.sql.execution.ProjectExec.consume(basicPhysicalOperators.scala:41)
	at org.apache.spark.sql.execution.ProjectExec.doConsume(basicPhysicalOperators.scala:87)
	at org.apache.spark.sql.execution.CodegenSupport.consume(WholeStageCodegenExec.scala:194)
	at org.apache.spark.sql.execution.CodegenSupport.consume$(WholeStageCodegenExec.scala:149)
	at org.apache.spark.sql.execution.joins.BroadcastHashJoinExec.consume(BroadcastHashJoinExec.scala:40)
	at org.apache.spark.sql.execution.joins.HashJoin.codegenOuter(HashJoin.scala:542)
	at org.apache.spark.sql.execution.joins.HashJoin.codegenOuter$(HashJoin.scala:496)
	at org.apache.spark.sql.execution.joins.BroadcastHashJoinExec.codegenOuter(BroadcastHashJoinExec.scala:40)
	at org.apache.spark.sql.execution.joins.HashJoin.doConsume(HashJoin.scala:352)
	at org.apache.spark.sql.execution.joins.HashJoin.doConsume$(HashJoin.scala:349)
	at org.apache.spark.sql.execution.joins.BroadcastHashJoinExec.doConsume(BroadcastHashJoinExec.scala:40)
	at org.apache.spark.sql.execution.CodegenSupport.consume(WholeStageCodegenExec.scala:194)
	at org.apache.spark.sql.execution.CodegenSupport.consume$(WholeStageCodegenExec.scala:149)
	at org.apache.spark.sql.execution.ProjectExec.consume(basicPhysicalOperators.scala:41)
	at org.apache.spark.sql.execution.ProjectExec.doConsume(basicPhysicalOperators.scala:87)
	at org.apache.spark.sql.execution.CodegenSupport.consume(WholeStageCodegenExec.scala:194)
	at org.apache.spark.sql.execution.CodegenSupport.consume$(WholeStageCodegenExec.scala:149)
	at org.apache.spark.sql.execution.InputAdapter.consume(WholeStageCodegenExec.scala:496)
	at org.apache.spark.sql.execution.InputRDDCodegen.doProduce(WholeStageCodegenExec.scala:483)
	at org.apache.spark.sql.execution.InputRDDCodegen.doProduce$(WholeStageCodegenExec.scala:456)
	at org.apache.spark.sql.execution.InputAdapter.doProduce(WholeStageCodegenExec.scala:496)
	at org.apache.spark.sql.execution.CodegenSupport.$anonfun$produce$1(WholeStageCodegenExec.scala:95)
	at org.apache.spark.sql.execution.SparkPlan.$anonfun$executeQuery$1(SparkPlan.scala:218)
	at org.apache.spark.rdd.RDDOperationScope$.withScope(RDDOperationScope.scala:151)
	at org.apache.spark.sql.execution.SparkPlan.executeQuery(SparkPlan.scala:215)
	at org.apache.spark.sql.execution.CodegenSupport.produce(WholeStageCodegenExec.scala:90)
	at org.apache.spark.sql.execution.CodegenSupport.produce$(WholeStageCodegenExec.scala:90)
	at org.apache.spark.sql.execution.InputAdapter.produce(WholeStageCodegenExec.scala:496)
	at org.apache.spark.sql.execution.ProjectExec.doProduce(basicPhysicalOperators.scala:54)
	at org.apache.spark.sql.execution.CodegenSupport.$anonfun$produce$1(WholeStageCodegenExec.scala:95)
	at org.apache.spark.sql.execution.SparkPlan.$anonfun$executeQuery$1(SparkPlan.scala:218)
	at org.apache.spark.rdd.RDDOperationScope$.withScope(RDDOperationScope.scala:151)
	at org.apache.spark.sql.execution.SparkPlan.executeQuery(SparkPlan.scala:215)
	at org.apache.spark.sql.execution.CodegenSupport.produce(WholeStageCodegenExec.scala:90)
	at org.apache.spark.sql.execution.CodegenSupport.produce$(WholeStageCodegenExec.scala:90)
	at org.apache.spark.sql.execution.ProjectExec.produce(basicPhysicalOperators.scala:41)
	at org.apache.spark.sql.execution.joins.HashJoin.doProduce(HashJoin.scala:346)
	at org.apache.spark.sql.execution.joins.HashJoin.doProduce$(HashJoin.scala:345)
	at org.apache.spark.sql.execution.joins.BroadcastHashJoinExec.doProduce(BroadcastHashJoinExec.scala:40)
	at org.apache.spark.sql.execution.CodegenSupport.$anonfun$produce$1(WholeStageCodegenExec.scala:95)
	at org.apache.spark.sql.execution.SparkPlan.$anonfun$executeQuery$1(SparkPlan.scala:218)
	at org.apache.spark.rdd.RDDOperationScope$.withScope(RDDOperationScope.scala:151)
	at org.apache.spark.sql.execution.SparkPlan.executeQuery(SparkPlan.scala:215)
	at org.apache.spark.sql.execution.CodegenSupport.produce(WholeStageCodegenExec.scala:90)
	at org.apache.spark.sql.execution.CodegenSupport.produce$(WholeStageCodegenExec.scala:90)
	at org.apache.spark.sql.execution.joins.BroadcastHashJoinExec.produce(BroadcastHashJoinExec.scala:40)
	at org.apache.spark.sql.execution.ProjectExec.doProduce(basicPhysicalOperators.scala:54)
	at org.apache.spark.sql.execution.CodegenSupport.$anonfun$produce$1(WholeStageCodegenExec.scala:95)
	at org.apache.spark.sql.execution.SparkPlan.$anonfun$executeQuery$1(SparkPlan.scala:218)
	at org.apache.spark.rdd.RDDOperationScope$.withScope(RDDOperationScope.scala:151)
	at org.apache.spark.sql.execution.SparkPlan.executeQuery(SparkPlan.scala:215)
	at org.apache.spark.sql.execution.CodegenSupport.produce(WholeStageCodegenExec.scala:90)
	at org.apache.spark.sql.execution.CodegenSupport.produce$(WholeStageCodegenExec.scala:90)
	at org.apache.spark.sql.execution.ProjectExec.produce(basicPhysicalOperators.scala:41)
	at org.apache.spark.sql.execution.joins.HashJoin.doProduce(HashJoin.scala:346)
	at org.apache.spark.sql.execution.joins.HashJoin.doProduce$(HashJoin.scala:345)
	at org.apache.spark.sql.execution.joins.BroadcastHashJoinExec.doProduce(BroadcastHashJoinExec.scala:40)
	at org.apache.spark.sql.execution.CodegenSupport.$anonfun$produce$1(WholeStageCodegenExec.scala:95)
	at org.apache.spark.sql.execution.SparkPlan.$anonfun$executeQuery$1(SparkPlan.scala:218)
	at org.apache.spark.rdd.RDDOperationScope$.withScope(RDDOperationScope.scala:151)
	at org.apache.spark.sql.execution.SparkPlan.executeQuery(SparkPlan.scala:215)
	at org.apache.spark.sql.execution.CodegenSupport.produce(WholeStageCodegenExec.scala:90)
	at org.apache.spark.sql.execution.CodegenSupport.produce$(WholeStageCodegenExec.scala:90)
	at org.apache.spark.sql.execution.joins.BroadcastHashJoinExec.produce(BroadcastHashJoinExec.scala:40)
	at org.apache.spark.sql.execution.ProjectExec.doProduce(basicPhysicalOperators.scala:54)
	at org.apache.spark.sql.execution.CodegenSupport.$anonfun$produce$1(WholeStageCodegenExec.scala:95)
	at org.apache.spark.sql.execution.SparkPlan.$anonfun$executeQuery$1(SparkPlan.scala:218)
	at org.apache.spark.rdd.RDDOperationScope$.withScope(RDDOperationScope.scala:151)
	at org.apache.spark.sql.execution.SparkPlan.executeQuery(SparkPlan.scala:215)
	at org.apache.spark.sql.execution.CodegenSupport.produce(WholeStageCodegenExec.scala:90)
	at org.apache.spark.sql.execution.CodegenSupport.produce$(WholeStageCodegenExec.scala:90)
	at org.apache.spark.sql.execution.ProjectExec.produce(basicPhysicalOperators.scala:41)
	at org.apache.spark.sql.execution.WholeStageCodegenExec.doCodeGen(WholeStageCodegenExec.scala:655)
	at org.apache.spark.sql.execution.WholeStageCodegenExec.doExecute(WholeStageCodegenExec.scala:718)
	at org.apache.spark.sql.execution.SparkPlan.$anonfun$execute$1(SparkPlan.scala:180)
	at org.apache.spark.sql.execution.SparkPlan.$anonfun$executeQuery$1(SparkPlan.scala:218)
	at org.apache.spark.rdd.RDDOperationScope$.withScope(RDDOperationScope.scala:151)
	at org.apache.spark.sql.execution.SparkPlan.executeQuery(SparkPlan.scala:215)
	at org.apache.spark.sql.execution.SparkPlan.execute(SparkPlan.scala:176)
	at org.apache.spark.sql.execution.exchange.ShuffleExchangeExec.inputRDD$lzycompute(ShuffleExchangeExec.scala:123)
	at org.apache.spark.sql.execution.exchange.ShuffleExchangeExec.inputRDD(ShuffleExchangeExec.scala:123)
	at org.apache.spark.sql.execution.exchange.ShuffleExchangeExec.shuffleDependency$lzycompute(ShuffleExchangeExec.scala:157)
	at org.apache.spark.sql.execution.exchange.ShuffleExchangeExec.shuffleDependency(ShuffleExchangeExec.scala:155)
	at org.apache.spark.sql.execution.exchange.ShuffleExchangeExec.$anonfun$doExecute$1(ShuffleExchangeExec.scala:172)
	at org.apache.spark.sql.catalyst.errors.package$.attachTree(package.scala:52)
	... 328 more
Caused by: org.apache.spark.util.SparkFatalException
	at org.apache.spark.sql.execution.exchange.BroadcastExchangeExec.$anonfun$relationFuture$1(BroadcastExchangeExec.scala:173)
	at org.apache.spark.sql.execution.SQLExecution$.$anonfun$withThreadLocalCaptured$1(SQLExecution.scala:190)
	at java.util.concurrent.FutureTask.run(FutureTask.java:266)
	at java.util.concurrent.ThreadPoolExecutor.runWorker(ThreadPoolExecutor.java:1149)
	at java.util.concurrent.ThreadPoolExecutor$Worker.run(ThreadPoolExecutor.java:624)
	at java.lang.Thread.run(Thread.java:750)

注意：处于安全考虑，本文隐藏了具体的物理执行计划
对于一个在大数据行业摸爬滚打了多年的老手来说，第一眼肯定是跟着堆栈信息进行排查，
理所当然的就是会找到BroadcastExchangeExec这个类，但是就算把代码全看一遍也不会有所发现。

蓦然回首

这个问题折腾了我大约2个小时，错误发生的上下文都看了不止十遍了，还是没找到一丝头绪，可能是上帝的旨意，在离错误不到50行的地方，刚好是一个页面的距离，发现了以下错误：

53.024: [Full GC (Ergonomics) [PSYoungGen: 802227K->698101K(1191424K)] [ParOldGen: 3085945K->3085781K(3495424K)] 3888173K->3783883K(4686848K), [Metaspace: 135862K->135862K(1185792K)], 0.9651630 secs] [Times: user=25.51 sys=0.39, real=0.96 secs] 
53.990: [Full GC (Allocation Failure) [PSYoungGen: 698101K->698047K(1191424K)] [ParOldGen: 3085781K->3079721K(3495424K)] 3783883K->3777769K(4686848K), [Metaspace: 135862K->134900K(1185792K)], 0.6236139 secs] [Times: user=14.05 sys=0.28, real=0.63 secs] 
java.lang.OutOfMemoryError: Java heap space
Dumping heap to panda_dump ...
Heap dump file created [3938522340 bytes in 5.708 secs]

真是 众人寻他千百度，蓦然回首， 没想到是 OOM 问题。

结论

在查找错误的时候，还是得在错误的上下文中多翻几页。