[
https://issues.apache.org/jira/browse/BEAM-10986?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17205714#comment-17205714
]
Kenneth Knowles commented on BEAM-10986:
----------------------------------------
I will emphasize "discovering and acquiring" to the list of "naming and
starting up" expansion service. The piece I was a bit curious about was stuff
around here:
[https://github.com/apache/beam/blob/ac35099e438d5fb8d471263e0485edc8c2eb6886/sdks/python/apache_beam/transforms/external.py#L522]
It eventually bottoms out at
[https://github.com/apache/beam/blob/6612b24ada9382706373db547b5606d6e0496b02/sdks/python/apache_beam/utils/subprocess_server.py#L209
|https://github.com/apache/beam/blob/6612b24ada9382706373db547b5606d6e0496b02/sdks/python/apache_beam/utils/subprocess_server.py#L209]which
replaces colons with dashes to guess a maven artifact id. This is a heuristic
that we super don't guarantee.
I agree wholeheartedly with docker maybe being overkill for this. I'm happy
enough with a bundled jar as the form of executable distribution. I think the
*big* missing piece here is how you robustly acquire whatever the artifact is.
Docker has the benefit of having a URL by default, while maven coords for the
released bundled jar work pretty well too.
> :build task doesn't build shaded jar with shadow >5.0.0
> -------------------------------------------------------
>
> Key: BEAM-10986
> URL: https://issues.apache.org/jira/browse/BEAM-10986
> Project: Beam
> Issue Type: Bug
> Components: build-system
> Reporter: Brian Hulette
> Assignee: Brian Hulette
> Priority: P0
> Fix For: 2.25.0
>
> Time Spent: 5h 10m
> Remaining Estimate: 0h
>
> Not 100% sure but I believe this will prevent us from releasing shaded jars.
> See
> [https://lists.apache.org/thread.html/r12c785c2b01017329204ad816740e80b70c19c5f2bb1ea01535a987d%40%3Cdev.beam.apache.org%3E]
> for details.
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
