You are viewing a plain text version of this content. The canonical link for it is here.
Posted to issues@spark.apache.org by "Colin Ma (Jira)" <ji...@apache.org> on 2019/09/04 07:08:00 UTC
[jira] [Commented] (SPARK-28340) Noisy exceptions when tasks are
killed: "DiskBlockObjectWriter: Uncaught exception while reverting partial
writes to file: java.nio.channels.ClosedByInterruptException"
[ https://issues.apache.org/jira/browse/SPARK-28340?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16922207#comment-16922207 ]
Colin Ma commented on SPARK-28340:
----------------------------------
In general, the exception will be caught and processed in Executor.run, and there will be no stack trace about the "excepted interrupt exception", eg, TaskKilledException, InterruptedException etc.
ClosedByInterruptException should also be treated as the "excepted interrupt exception", and don't log the trace stack. It's hard to figure out all of them, but we can fix them start with reported places first.
> Noisy exceptions when tasks are killed: "DiskBlockObjectWriter: Uncaught exception while reverting partial writes to file: java.nio.channels.ClosedByInterruptException"
> ------------------------------------------------------------------------------------------------------------------------------------------------------------------------
>
> Key: SPARK-28340
> URL: https://issues.apache.org/jira/browse/SPARK-28340
> Project: Spark
> Issue Type: Improvement
> Components: Spark Core
> Affects Versions: 3.0.0
> Reporter: Josh Rosen
> Priority: Minor
>
> If a Spark task is killed while writing blocks to disk (due to intentional job kills, automated killing of redundant speculative tasks, etc) then Spark may log exceptions like
> {code:java}
> 19/07/10 21:31:08 ERROR storage.DiskBlockObjectWriter: Uncaught exception while reverting partial writes to file /<FILENAME>
> java.nio.channels.ClosedByInterruptException
> at java.nio.channels.spi.AbstractInterruptibleChannel.end(AbstractInterruptibleChannel.java:202)
> at sun.nio.ch.FileChannelImpl.truncate(FileChannelImpl.java:372)
> at org.apache.spark.storage.DiskBlockObjectWriter$$anonfun$revertPartialWritesAndClose$2.apply$mcV$sp(DiskBlockObjectWriter.scala:218)
> at org.apache.spark.util.Utils$.tryWithSafeFinally(Utils.scala:1369)
> at org.apache.spark.storage.DiskBlockObjectWriter.revertPartialWritesAndClose(DiskBlockObjectWriter.scala:214)
> at org.apache.spark.shuffle.sort.BypassMergeSortShuffleWriter.stop(BypassMergeSortShuffleWriter.java:237)
> at org.apache.spark.scheduler.ShuffleMapTask.runTask(ShuffleMapTask.scala:105)
> at org.apache.spark.scheduler.ShuffleMapTask.runTask(ShuffleMapTask.scala:55)
> at org.apache.spark.scheduler.Task.run(Task.scala:121)
> at org.apache.spark.executor.Executor$TaskRunner$$anonfun$10.apply(Executor.scala:402)
> at org.apache.spark.util.Utils$.tryWithSafeFinally(Utils.scala:1360)
> at org.apache.spark.executor.Executor$TaskRunner.run(Executor.scala:408)
> at java.util.concurrent.ThreadPoolExecutor.runWorker(ThreadPoolExecutor.java:1149)
> at java.util.concurrent.ThreadPoolExecutor$Worker.run(ThreadPoolExecutor.java:624)
> at java.lang.Thread.run(Thread.java:748){code}
> If {{BypassMergeSortShuffleWriter}} is being used then a single cancelled task can result in hundreds of these stacktraces being logged.
> Here are some StackOverflow questions asking about this:
> * [https://stackoverflow.com/questions/40027870/spark-jobserver-job-crash]
> * [https://stackoverflow.com/questions/50646953/why-is-java-nio-channels-closedbyinterruptexceptio-called-when-caling-multiple]
> * [https://stackoverflow.com/questions/41867053/java-nio-channels-closedbyinterruptexception-in-spark]
> * [https://stackoverflow.com/questions/56845041/are-closedbyinterruptexception-exceptions-expected-when-spark-speculation-kills]
> Can we prevent this exception from occurring? If not, can we treat this "expected exception" in a special manner to avoid log spam? My concern is that the presence of large numbers of spurious exceptions is confusing to users when they are inspecting Spark logs to diagnose other issues.
--
This message was sent by Atlassian Jira
(v8.3.2#803003)
---------------------------------------------------------------------
To unsubscribe, e-mail: issues-unsubscribe@spark.apache.org
For additional commands, e-mail: issues-help@spark.apache.org