[
https://issues.apache.org/jira/browse/FLINK-9612?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16517525#comment-16517525
]
Eron Wright commented on FLINK-9612:
-------------------------------------
Yes it makes sense that the overlays would be selective and configurable, and
especially true that the Flink binaries aren't needed in most scenarios
involving a docker image. Specifically on that, I wonder if the Flink conf
directory should be treated differently from the bin/libs (perhaps as a
different overlay), since the image might be 'stock'.
> Add option for minimal artifacts being pulled in Mesos
> ------------------------------------------------------
>
> Key: FLINK-9612
> URL: https://issues.apache.org/jira/browse/FLINK-9612
> Project: Flink
> Issue Type: Improvement
> Components: Configuration, Docker, Mesos
> Reporter: Addison Higham
> Priority: Major
>
> NOTE: this assumes mesos, but this improvement could also be useful for
> future container deployments.
> Currently, in mesos, the FlinkDistributionOverlay copies the entire `conf`,
> `bin`, and `lib` folders from the running JobManager/ResourceManager. When
> using docker with a pre-installed flink distribution, this is relatively
> inefficient as it pulls jars that are already baked into the container image.
> A new option that disables pulling most (if not all?) of the
> FlinkDistributionOverlay could allow for much faster and more scalable
> provisions of TaskManagers. As it currently stands, trying to run a few
> hundred TaskManagers is likely to result in poor performance in pulling all
> the artifacts from the MesosArtifactServer
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
