[jira] [Commented] (FLINK-18357) ContinuousFileReaderOperator checkpoint timeout when files and directories get larger

["Stephan Ewen (Jira)" <ji...@apache.org>]

    [ https://issues.apache.org/jira/browse/FLINK-18357?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17139526#comment-17139526 ] 

Stephan Ewen commented on FLINK-18357:
--------------------------------------

I think this should be solved in Flink 1.12 by porting the file source to FLIP-27.
The solution in the current source is rather complex.

> ContinuousFileReaderOperator checkpoint timeout when files and directories get larger
> -------------------------------------------------------------------------------------
>
>                 Key: FLINK-18357
>                 URL: https://issues.apache.org/jira/browse/FLINK-18357
>             Project: Flink
>          Issue Type: Improvement
>          Components: API / DataStream
>            Reporter: Jark Wu
>            Priority: Major
>
> This is reported from user-zh, I tranlsated it into this issue: http://apache-flink.147419.n8.nabble.com/env-readFile-td4017.html
> {{env.readFile(format,path, FileProcessingMode.PROCESS_CONTINUOUSLY, 60000)}}
> It monitors a directory of A, we will generate a new directory under A every day, such as:
> A/20200101/
> A/20200102/
> A/20200103/
> .......
> .......
> However, as time goes on, it has to monitor 200 directories and 500 files in each directory until June. Then everytime, the checkpoint has to synchronize offsets of 200*500 files which is very large and checkpoint gets timeout. 
> ========================================================
> My though is that is it possible to have a configuration, to mark a sub-directory as idle after a inactive interval, then we don't need to checkpoint all the file offsets under it. 



--
This message was sent by Atlassian Jira
(v8.3.4#803005)