[
https://issues.apache.org/jira/browse/FLINK-29084?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17584207#comment-17584207
]
Weike Dong commented on FLINK-29084:
------------------------------------
Oh yes, but that issue is still left unresolved. So probably we need to address
this issue to prevent further potential "bugs" like this :D
> Program argument containing # (pound sign) mistakenly truncated in Kubernetes
> mode
> ----------------------------------------------------------------------------------
>
> Key: FLINK-29084
> URL: https://issues.apache.org/jira/browse/FLINK-29084
> Project: Flink
> Issue Type: Bug
> Components: Deployment / Kubernetes
> Affects Versions: 1.13.6, 1.14.5, 1.15.1
> Environment: Flink 1.13.6
> Native Kubernetes (Application Mode)
> Reporter: Weike Dong
> Priority: Minor
>
> We have found that when submitting jobs in native-Kubernetes mode, the main
> arguments of the Flink program would be truncated if it contains a #
> character.
> For example, if we pass 'ab#cd' as the argument for Flink programs, Flink
> actually gets only 'ab' from the variable
> `$internal.application.program-args` at runtime.
> After searching into the code, we found the reason might be that when
> `org.apache.flink.kubernetes.kubeclient.decorators.FlinkConfMountDecorator#buildAccompanyingKubernetesResources`
> transform Flink config data `Map` into `ConfigMap`, fabric8 Kubernetes
> client converts it to YAML internally, without any escaping procedures.
> Afterwards, when there is a # character in the YAML line, the decoder treats
> it as the start of a comment, thus the substring after the # character is
> ignored erroneously.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
