[
https://issues.apache.org/jira/browse/FLINK-23737?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17398571#comment-17398571
]
Chesnay Schepler edited comment on FLINK-23737 at 8/13/21, 10:14 AM:
---------------------------------------------------------------------
> that makes sure the native k8s deployment works independent from the
> DockerHub review processes.
That's again more of an issue of our release process, with us announcing
releases before all artifacts are published. Which furthermore only happened
_once_.
Just last week we had an issue where we deployed incorrectly built images
partially breaking production environments _for all versions_ (which this
ticket would limit to one release).
As is we're not saving anything. We create confusion among users because there
are 2 sources of images, that potentially differ if something comes up in the
DockerHub review process. We both have to do the DockerHub release process, and
release stuff on our own + maintaining scripts and setting up infrastructure to
ensure we aren't messing things up.
What's worse, we never even tried to find a solution with DockerHub.
We could for example publish our RCs there as well; which would both find
issues in the images earlier and allows us to _actually_ test the image we
would be releasing.
was (Author: zentol):
> that makes sure the native k8s deployment works independent from the
> DockerHub review processes.
That's again more of an issue of our release process, with us announcing
releases before all artifacts are published. Which furthermore only happened
_once_.
Just last week we had an issue where we deployed incorrectly built images
breaking production environments _for all versions_ (which this ticket would
limit to one release).
As is we're not saving anything. We create confusion among users because there
are 2 sources of images, that potentially differ if something comes up in the
DockerHub review process. We both have to do the DockerHub release process, and
release stuff on our own + maintaining scripts and setting up infrastructure to
ensure we aren't messing things up.
What's worse, we never even tried to find a solution with DockerHub.
We could for example publish our RCs there as well; which would both find
issues in the images earlier and allows us to _actually_ test the image we
would be releasing.
> Ensure publish-to-dockerhub.sh only publish specific flink versions
> -------------------------------------------------------------------
>
> Key: FLINK-23737
> URL: https://issues.apache.org/jira/browse/FLINK-23737
> Project: Flink
> Issue Type: New Feature
> Components: Build System
> Reporter: Yun Tang
> Priority: Major
> Fix For: 1.14.0
>
>
> Current publish-to-dockerhub.sh in
> https://github.com/apache/flink-docker/blob/master/publish-to-dockerhub.sh
> actually publish all images in current maintaining flink versions. However,
> this is not necessary and release manager should only publish the specific
> version he needs.
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
