You are viewing a plain text version of this content. The canonical link for it is here.

Posted to issues@spark.apache.org by "Jose Torres (JIRA)" <ji...@apache.org> on 2018/01/04 22:15:00 UTC

[jira] [Updated] (SPARK-22963) Make failure recovery global and automatic for continuous processing.

     [ https://issues.apache.org/jira/browse/SPARK-22963?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ]

Jose Torres updated SPARK-22963:
--------------------------------
    Summary: Make failure recovery global and automatic for continuous processing.  (was: Clean up continuous processing failure recovery)

> Make failure recovery global and automatic for continuous processing.
> ---------------------------------------------------------------------
>
>                 Key: SPARK-22963
>                 URL: https://issues.apache.org/jira/browse/SPARK-22963
>             Project: Spark
>          Issue Type: Improvement
>          Components: Structured Streaming
>    Affects Versions: 2.3.0
>            Reporter: Jose Torres
>
> Spark native task restarts don't work well for continuous processing. They will process all data from the task's original start offset - even data which has already been committed. This is not semantically incorrect under at least once semantics, but it's awkward and bad.
> Fortunately, they're also not necessary; the central coordinator can restart every task from the checkpointed offsets without losing much. So we should make that happen on task failures instead.



--
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