Prevent older Bodhi updates from superseding newer builds during stable pushes #13468

New issue

Open

opened 2026-08-12 05:04:31 +00:00 by gotmax23 · 1 comment

gotmax23 commented

2026-08-12 05:04:31 +00:00

Member

Describe the issue

Supersedes: #13332

(Text partially Assisted-by: GPT-5 🙈 and heavily edited, but hopefully this is better than the old ticket)

Problem

Bodhi can allow conflicting updates for the same package to coexist in certain scenarios where the update obsoleting functionality doesn't work:

If the older update is locked/in a push [e.g., stuck in testing->stable because of a feature freeze]

If the older update has multiple packages attached to it [e.g., it comes from a sidetag]

If the older update is submitted to stable? [but not yet pushed to stable]

(https://github.com/fedora-infra/bodhi/issues/1266)

If conflicting updates move from testing to stable in the same push, a race condition can occur that allows an older build of a package to supersede a newer one. Long update freezes seem to increase the possibility of conflicting updates accumulating in the testing -> stable phase and causing this race when the feature freeze ends and updates starting getting pushed to stable all at once.

Observed during the initial Fedora 44 updates push:

python3.14-3.14.4-1.fc44 superseded python3.14-3.14.4-2.fc44, which contained security fixes (#13331)
ruby-build-20260327-2.fc44 superseded ruby-build-20260412-1.fc44 (#13332)

Solution?

The affected f44 builds were manually retagged last cycle, but the underlying problem remains.

Potential solutions this include updating Bodhi to make the checks stricter to prevent conflicting updates (https://github.com/fedora-infra/bodhi/issues/1266) or creating a separate releng process/script to identify these cases and retag the correct builds.

When do you need this? (YYYY/MM/DD)

Before the F45 freezes, ideally

When is this no longer needed or useful? (YYYY/MM/DD)

n/a

If we cannot complete this, what is the impact? [Dependencies/Blocker]

We'll continue to end up with update conflicts that have to be manually reconciled each cycle.

## Describe the issue  Supersedes: https://forge.fedoraproject.org/releng/tickets/issues/13332 (Text partially Assisted-by: GPT-5 🙈 and heavily edited, but hopefully this is better than the old ticket) ### Problem Bodhi can allow conflicting updates for the same package to coexist in certain scenarios where the update obsoleting functionality doesn't work: > - If the older update is locked/in a push [e.g., stuck in `testing->stable` because of a feature freeze] > - If the older update has multiple packages attached to it [e.g., it comes from a sidetag] > - If the older update is submitted to stable? [but not yet pushed to stable] (https://github.com/fedora-infra/bodhi/issues/1266) If conflicting updates move from testing to stable in the same push, a race condition can occur that allows an older build of a package to supersede a newer one. Long update freezes seem to increase the possibility of conflicting updates accumulating in the `testing -> stable` phase and causing this race when the feature freeze ends and updates starting getting pushed to stable all at once. Observed during the initial Fedora 44 updates push: - `python3.14-3.14.4-1.fc44` superseded `python3.14-3.14.4-2.fc44`, which contained security fixes (https://forge.fedoraproject.org/releng/tickets/issues/13331) - `ruby-build-20260327-2.fc44` superseded `ruby-build-20260412-1.fc44` (https://forge.fedoraproject.org/releng/tickets/issues/13332) ### Solution? The affected f44 builds were manually retagged last cycle, but the underlying problem remains. Potential solutions this include updating Bodhi to make the checks stricter to prevent conflicting updates (https://github.com/fedora-infra/bodhi/issues/1266) or creating a separate releng process/script to identify these cases and retag the correct builds. ## When do you need this? (YYYY/MM/DD)  Before the F45 freezes, ideally ## When is this no longer needed or useful? (YYYY/MM/DD)  n/a ## If we cannot complete this, what is the impact? [Dependencies/Blocker]  We'll continue to end up with update conflicts that have to be manually reconciled each cycle.

Rows
Columns