[
https://issues.apache.org/jira/browse/FLINK-27910?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17550823#comment-17550823
]
Gen Luo commented on FLINK-27910:
---------------------------------
This is a mistake while migrating the FileSink to the new sink API. In the new
sink API createWriter and restoreWriter have been separated into two methods,
while originally creating a writer is by calling restoreWriter with an empty
state collection. We mistook the the meaning of createWriter and only created a
writer in it.
A PR has been provided to fix this bug, which changes the createWriter as the
original behavior.
> FileSink not registered the timer to enforce rolling policy if started from
> scratch
> -----------------------------------------------------------------------------------
>
> Key: FLINK-27910
> URL: https://issues.apache.org/jira/browse/FLINK-27910
> Project: Flink
> Issue Type: Bug
> Components: Connectors / FileSystem
> Affects Versions: 1.15.0, 1.16.0
> Reporter: Yun Gao
> Priority: Critical
> Labels: pull-request-available
>
> The current FileWriter only register the timer in initializeState, which is
> now only called on restoring. Thus if the job is started from scratch, the
> timer would fail to be registered and cause the rolling policy not work.
--
This message was sent by Atlassian Jira
(v8.20.7#820007)
