You are viewing a plain text version of this content. The canonical link for it is here.
Posted to issues@spark.apache.org by "Apache Spark (Jira)" <ji...@apache.org> on 2022/06/03 03:02:00 UTC

[jira] [Commented] (SPARK-39366) BlockInfoManager should not release write locks on task end

    [ https://issues.apache.org/jira/browse/SPARK-39366?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17545728#comment-17545728 ] 

Apache Spark commented on SPARK-39366:
--------------------------------------

User 'hvanhovell' has created a pull request for this issue:
https://github.com/apache/spark/pull/36751

> BlockInfoManager should not release write locks on task end
> -----------------------------------------------------------
>
>                 Key: SPARK-39366
>                 URL: https://issues.apache.org/jira/browse/SPARK-39366
>             Project: Spark
>          Issue Type: Bug
>          Components: SQL
>    Affects Versions: 3.2.1
>            Reporter: Herman van Hövell
>            Assignee: Herman van Hövell
>            Priority: Major
>
> The BlockInfoManager releases all locks held by a task when the task is done. It also release write locks, the problem with that is that a thread (other than the main task thread) might still be modifying the block. By releasing it the block now seems readable, and a reader might observe a block in a partial or non-existent state.
> This is fortunately not the massive problem as it appears to be because the BlockManager (only place where we write blocks) is well behaved and always puts the block in a consistent state. This means the errors caused by this are only transient.
> Given the fact that the write code is well behaved we don't need to release the write locks on task end. We should remove that.



--
This message was sent by Atlassian Jira
(v8.20.7#820007)

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