[
https://issues.apache.org/jira/browse/FLINK-40782?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=18118359#comment-18118359
]
Martijn Visser commented on FLINK-40782:
----------------------------------------
[~snuyanzin] Can you keep me honest that this is indeed the right way to tackle
this one?
> Fail connector CI when ArchUnit would update a frozen violation store
> ---------------------------------------------------------------------
>
> Key: FLINK-40782
> URL: https://issues.apache.org/jira/browse/FLINK-40782
> Project: Flink
> Issue Type: Improvement
> Components: Build System / CI, Connectors / Parent
> Reporter: Martijn Visser
> Priority: Major
>
> Connector CI runs {{mvn clean deploy}} through {{ci.yml}} on the {{ci_utils}}
> branch, and every connector's {{archunit.properties}} sets
> {{freeze.store.default.allowStoreUpdate=true}}. When a violation disappears,
> ArchUnit deletes it from the store and the build stays green. The deletion is
> thrown away with the workspace, so the stale entry stays allowed. In
> FLINK-40781 this hid that the JDBC production rules analysed no classes at
> all.
> Flink itself runs CI with
> {{-Darchunit.freeze.store.default.allowStoreUpdate=false}} (see
> {{tools/ci/test_controller.sh}}). {{ci.yml}} should do the same, with an
> input to opt out. Connectors that run ArchUnit against more than one Flink
> version can use {{skip_archunit_tests}} for the others, as {{_testing.yml}}
> already shows. Connectors with stale entries will need to refreeze once.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
