[
https://issues.apache.org/jira/browse/FLINK-26030?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17489957#comment-17489957
]
Biao Geng commented on FLINK-26030:
-----------------------------------
Yes, it should be a bug.
I believe to be consistent with the behavior in flink client side, in YARN
cluster side, we should make the {{FLINK_LIB_DIR}} in YARN container env to the
{{lib}} dir under YARN's working dir. If that is the case, I may start
investigating on how to implement it properly. Let me know if above assumption
make sense. Thanks!
> Set FLINK_LIB_DIR to 'lib' under working dir in YARN containers
> ---------------------------------------------------------------
>
> Key: FLINK-26030
> URL: https://issues.apache.org/jira/browse/FLINK-26030
> Project: Flink
> Issue Type: Bug
> Components: Deployment / YARN
> Reporter: Biao Geng
> Priority: Minor
>
> Currently, we utilize
> {{org.apache.flink.runtime.entrypoint.ClusterEntrypointUtils#tryFindUserLibDirectory}}
> to locate usrlib in both flink client and cluster side.
> This method relies on the value of environment variable {{FLINK_LIB_DIR}} to
> find the {{usrlib}}.
> It makes sense in client side since in {{bin/config.sh}}, {{FLINK_LIB_DIR}}
> will be set by default(i.e. {{FLINK_HOME/lib}} if not exists. But in YARN
> cluster's containers, when we want to reuse this method to find {{usrlib}},
> as the YARN usually starts the process using commands like
> bq. /bin/bash -c /usr/lib/jvm/java-1.8.0/bin/java -Xmx1073741824
> -Xms1073741824
> -XX:MaxMetaspaceSize=268435456org.apache.flink.yarn.entrypoint.YarnJobClusterEntrypoint
> -D jobmanager.memory.off-heap.size=134217728b -D
> jobmanager.memory.jvm-overhead.min=201326592b -D
> jobmanager.memory.jvm-metaspace.size=268435456b -D
> jobmanager.memory.heap.size=1073741824b -D
> jobmanager.memory.jvm-overhead.max=201326592b ...
> {{FLINK_LIB_DIR}} is not guaranteed to be set in such case. Current codes
> will use current working dir to locate the {{usrlib}} which is correct in
> most cases. But bad things can happen if the machine which the YARN container
> resides in has already set {{FLINK_LIB_DIR}} to a different folder. In that
> case, codes will try to find {{usrlib}} in a undesired place.
> One possible solution would be overriding the {{FLINK_LIB_DIR}} in YARN
> container env to the {{lib}} dir under YARN's workding dir.
--
This message was sent by Atlassian Jira
(v8.20.1#820001)
