voonhous commented on PR #19433:
URL: https://github.com/apache/hudi/pull/19433#issuecomment-5162862915

   Update after chasing the consistency question properly. Short version: 
**this is done as scoped, and I am withdrawing the description asks rather than 
holding it up for them.**
   
   **State.** 75/75 checks green on `ed3ccf5`, `mergeable`, no conflicts, both 
blocking items verified fixed against a clean build. All eight review threads 
are now resolved. The only thing outstanding is an approving review.
   
   **Withdrawing the two Impact asks.** Both were record quality rather than 
correctness, and the content now lives in the threads instead, which is good 
enough:
   
   - the graph-flattening consequence of promotion is on the `hudi-aws-bundle` 
thread
   - the `provided` drop on `hudi-datahub-sync-bundle` is on that thread -- 
along with a correction to my own earlier comment there, since I had asked you 
to fix reasoning that only ever existed in the thread and never in the 
description. Sorry for the noise on that one.
   
   One small thing I would still take if you are touching the description 
anyway, but not worth a round trip on its own: the **"Open question: 
consistency with the nine already-reduced bundles"** section reads as 
undecided, and it is decided -- promotion stays on these seven. Retitling it 
would stop the merged record saying the question was open at merge time. 
Entirely your call.
   
   **Release note** for the seven bundles still stands, at release time rather 
than in this PR.
   
   **Follow-ups to file, revised.** The consistency one changed shape once I 
looked at what the repo actually does -- details and evidence on the 
`hudi-aws-bundle` thread:
   
   - [ ] **Hoist the shade POM flags to the root pom**, rather than "add 
promotion to the other nine". The root already carries `maven-shade-plugin` 
with a shared `<configuration>` that all bundles inherit, so 
`createDependencyReducedPom`, `promoteTransitiveDependencies` and 
`dependencyReducedPomLocation` belong there once, per-bundle copies deleted, 
with a single documented override in `hudi-io`. Two things that reframe this: 
`hudi-io` sets `createDependencyReducedPom=false` correctly, because it is the 
only module using `shadedArtifactAttached=true`; and the "already on a 
Spark/Flink classpath" argument only covers four of the nine, so 
`hudi-cli-bundle`, `hudi-kafka-connect-bundle`, `hudi-integ-test-bundle`, 
`hudi-metaserver-server-bundle` and `hudi-timeline-server-bundle` are the five 
with real exposure and should be verified first.
   - [ ] presto coverage in `packaging/bundle-validation`, stating in the issue 
that every `-pl packaging/...` in `.github/workflows` and `scripts/` passes 
`-am`, so the green Azure run here could not have caught the regression this PR 
fixes.
   - [ ] dead relocation entry `org.apache.httpcomponents.` -> 
`org.apache.hudi.aws.org.apache.httpcomponents.` in `hudi-aws-bundle`.
   - [ ] `hudi-presto-bundle` declaring `hudi-hadoop-mr-bundle`, a fat jar 
duplicating classes it already shades.
   
   Happy for these to be filed by either of us -- say which you would rather 
pick up.
   


-- 
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.

To unsubscribe, e-mail: [email protected]

For queries about this service, please contact Infrastructure at:
[email protected]

Reply via email to