fix: Run release verification with --profile=ci#20987
Conversation
viirya
left a comment
There was a problem hiding this comment.
It is good to restore the verification script. However I wonder if we should still stick with debug mode when running verification script before release? If a debug check is actually effective and correct, is it possibly we skip it both in ci and verification script?
I guess we can go back to debug once we upgrade to the new version? I think it is good to have the debug assertions running as extra verification / test for the assumptions. |
Yea, sounds good to me. |
I think we should be verifying with the same setup as is used in CI Maybe we should contemplate removing the ci profile and just always use the debug profile? I can't remember why we put in the CI profile in the first place 🤔 |
## Which issue does this PR close? - Related to apache#20689 - Closes apache#20992 - part of apache#19692 ## Rationale for this change While verifying DataFusion 53.0.0, we found that the `verify_release_candidate` script was not working correctly, as it was hitting an assertion Specifically: ```shell $ ./dev/release/verify-release-candidate.sh 53.0.0 2 ... joins::sort_merge_join::tests::test_should_emit_early_when_have_enough_data_to_emit test result: FAILED. 1190 passed; 78 failed; 0 ignored; 0 measured; 0 filtered out; finished in 8.89s error: test failed, to rerun pass `-p datafusion-physical-plan --lib` + cleanup + '[' no = yes ']' + echo 'Failed to verify release candidate. See /var/folders/1l/tg68jc6550gg8xqf1hr4mlwr0000gn/T/datafusion-53.0.0.XXXXX.d5aUZcsu7v for details.' Failed to verify release candidate. See /var/folders/1l/tg68jc6550gg8xqf1hr4mlwr0000gn/T/datafusion-53.0.0.XXXXX.d5aUZcsu7v for details. ``` The failure is due to an overzealous assert in arrow-rs (see apache/arrow-rs#9506) that will be fixed in the next release The reason this isn' triggered on CI is that sqllogictests are run with `--profile=ci` -- see https://github.com/apache/datafusion/blob/11b9693952cd419b73dd03cc39f22c8b343bc05c/.github/workflows/rust.yml#L299-L298 ## What changes are included in this PR? 1. Update verify_release.sh to use `--profile=ci` when running sqllogictests to mirror CI ## Are these changes tested? I tested it manually (STILLRUNNING) ```shell $ ./dev/release/verify-release-candidate.sh 53.0.0 2 ... /var/folders/1l/tg68jc6550gg8xqf1hr4mlwr0000gn/T/datafusion-53.0.0.XXXXX.j0FgM9yH5J + TEST_SUCCESS=yes + echo 'Release candidate looks good!' Release candidate looks good! + exit 0 + cleanup + '[' yes = yes ']' + rm -fr /var/folders/1l/tg68jc6550gg8xqf1hr4mlwr0000gn/T/datafusion-53.0.0.XXXXX.j0FgM9yH5J ``` ## Are there any user-facing changes? No
Which issue does this PR close?
53.0.0(Feb 2026 / Mar 2026) #19692Rationale for this change
While verifying DataFusion 53.0.0, we found that the
verify_release_candidatescript was not working correctly, as it was hitting an assertionSpecifically:
$ ./dev/release/verify-release-candidate.sh 53.0.0 2 ... joins::sort_merge_join::tests::test_should_emit_early_when_have_enough_data_to_emit test result: FAILED. 1190 passed; 78 failed; 0 ignored; 0 measured; 0 filtered out; finished in 8.89s error: test failed, to rerun pass `-p datafusion-physical-plan --lib` + cleanup + '[' no = yes ']' + echo 'Failed to verify release candidate. See /var/folders/1l/tg68jc6550gg8xqf1hr4mlwr0000gn/T/datafusion-53.0.0.XXXXX.d5aUZcsu7v for details.' Failed to verify release candidate. See /var/folders/1l/tg68jc6550gg8xqf1hr4mlwr0000gn/T/datafusion-53.0.0.XXXXX.d5aUZcsu7v for details.The failure is due to an overzealous assert in arrow-rs (see apache/arrow-rs#9506) that will be fixed in the next release
The reason this isn' triggered on CI is that sqllogictests are run with
--profile=ci-- seehttps://github.com/apache/datafusion/blob/11b9693952cd419b73dd03cc39f22c8b343bc05c/.github/workflows/rust.yml#L299-L298
What changes are included in this PR?
--profile=ciwhen running sqllogictests to mirror CIAre these changes tested?
I tested it manually (STILLRUNNING)
Are there any user-facing changes?
No