[ 
https://issues.apache.org/jira/browse/FLINK-40833?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=18120200#comment-18120200
 ] 

Martijn Visser commented on FLINK-40833:
----------------------------------------

This is caused by upstream issue 
https://github.com/apache/logging-log4j2/issues/4362

> JobManager and TaskManager logs are rolled at startup with log4j 2.25.2 and 
> later
> ---------------------------------------------------------------------------------
>
>                 Key: FLINK-40833
>                 URL: https://issues.apache.org/jira/browse/FLINK-40833
>             Project: Flink
>          Issue Type: Bug
>          Components: Runtime / Coordination
>    Affects Versions: 2.3.0, 2.4.0
>            Reporter: Martijn Visser
>            Assignee: Martijn Visser
>            Priority: Major
>
> Since the upgrade to log4j 2.25.3 (FLINK-39146), a JobManager or TaskManager 
> started with the default conf/log4j.properties sometimes rolls its log file 
> about half a second after startup. The startup section (JVM options, 
> configuration, Kerberos login) then sits in jobmanager.log.1, and 
> jobmanager.log starts at "Trying to start actor system".
> The first logger of a class in the RPC system's submodule classloader creates 
> a second LoggerContext, which runs OnStartupTriggeringPolicy again on the 
> open file. Since 2.25.2 log4j rounds the file's creation time to the second, 
> so a file created in the first half of the JVM's start second looks older 
> than the JVM and is rolled. Locally that happened in 31 of 100 starts with 
> 2.25.2 and 0 of 100 with 2.25.1, and not with 
> log4j2.contextSelector=org.apache.logging.log4j.core.selector.BasicContextSelector.
> This is the cause of FLINK-40109. FLINK-40070 is the same symptom in 
> DynamicParameterITCase.



--
This message was sent by Atlassian Jira
(v8.20.10#820010)

Reply via email to