[jira] [Commented] (SPARK-21873) CachedKafkaConsumer throws NonLocalReturnControl during fetching from Kafka

["Apache Spark (JIRA)" <ji...@apache.org>]

    [ https://issues.apache.org/jira/browse/SPARK-21873?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16146901#comment-16146901 ] 

Apache Spark commented on SPARK-21873:
--------------------------------------

User 'YuvalItzchakov' has created a pull request for this issue:
https://github.com/apache/spark/pull/19059

> CachedKafkaConsumer throws NonLocalReturnControl during fetching from Kafka
> ---------------------------------------------------------------------------
>
>                 Key: SPARK-21873
>                 URL: https://issues.apache.org/jira/browse/SPARK-21873
>             Project: Spark
>          Issue Type: Bug
>          Components: Structured Streaming
>    Affects Versions: 2.1.0, 2.1.1, 2.2.0
>            Reporter: Yuval Itzchakov
>            Priority: Minor
>   Original Estimate: 0h
>  Remaining Estimate: 0h
>
> In Scala, using `return` inside a function causes a `NonLocalReturnControl` exception to be thrown and caught in order to escape the current scope.
> While profiling Structured Streaming in production, it clearly shows:
> !https://user-images.githubusercontent.com/3448320/29743366-4149ef78-8a99-11e7-94d6-f0cbb691134a.png!
> This happens during a 1 minute profiling session on a single executor. The code is:
> {code:scala}
> while (toFetchOffset != UNKNOWN_OFFSET) {
>       try {
>         return fetchData(toFetchOffset, untilOffset, pollTimeoutMs, failOnDataLoss)
>       } catch {
>         case e: OffsetOutOfRangeException =>
>           // When there is some error thrown, it's better to use a new consumer to drop all cached
>           // states in the old consumer. We don't need to worry about the performance because this
>           // is not a common path.
>           resetConsumer()
>           reportDataLoss(failOnDataLoss, s"Cannot fetch offset $toFetchOffset", e)
>           toFetchOffset = getEarliestAvailableOffsetBetween(toFetchOffset, untilOffset)
>       }
>     }
> {code}
> This happens because this method is converted to a function which is ran inside:
> {code:scala}
> private def runUninterruptiblyIfPossible[T](body: => T): T
> {code}
> We should avoid using `return` in general, and here specifically as it is a hot path for applications using Kafka.



--
This message was sent by Atlassian JIRA
(v6.4.14#64029)

---------------------------------------------------------------------
To unsubscribe, e-mail: issues-unsubscribe@spark.apache.org
For additional commands, e-mail: issues-help@spark.apache.org