[
https://issues.apache.org/jira/browse/FLINK-28915?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17579603#comment-17579603
]
hjw commented on FLINK-28915:
-----------------------------
I successful modify the Class KubernetesApplicationClusterEntryPoint.java to
achieve the purpose that support S3 schema.
Here the modify logic:
1.read the jar localtion in s3 from pipeline.jars parameters.
2.download the jar from s3 to local path of pod(jobmanager).
3.replace the local path (local schema) to pipeline.jars parameters.
I know such an implementation is not elegant and not compatible with other
remote DFS schema.(OSS.HDFS .etc). I think the more elegant implementation is
to use Flink filesystem to connect each DFS schema.
Howerver, I notice the fact that Flink filesystem is configred in Starting
flink cluster. But the job PackagedProgram is inited in
KubernetesApplicationClusterEntryPoint before the code
"ClusterEntrypoint.runclusterEntrypoint(KubernetesApplicationClusterEntryPoint)".
> Flink Native k8s mode jar localtion support s3 schema
> ------------------------------------------------------
>
> Key: FLINK-28915
> URL: https://issues.apache.org/jira/browse/FLINK-28915
> Project: Flink
> Issue Type: Improvement
> Components: Deployment / Kubernetes, flink-contrib
> Affects Versions: 1.15.0, 1.15.1
> Reporter: hjw
> Priority: Major
>
> As the Flink document show , local is the only supported scheme in Native k8s
> deployment.
> Is there have a plan to support s3 filesystem? thx.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
