The Big Picture

With the coordinated 2025.2 "Flamingo" release now just weeks out (the release team marked week R-2 in Thierry Carrez's countdown, and the TC's R-3 summary put it three weeks away), this week's traffic was dominated by end-game release mechanics: last-minute dependency regressions, freeze exceptions, and a scramble to land or restore features before the door closes. Underneath the release churn, operators surfaced a cluster of real-world pain points — stuck live migrations, MariaDB compatibility, and snapshot-deletion permission errors — while governance continued to wrestle with leaderless project teams heading into the 2026.1 "Gazpacho" cycle.

Releases & Announcements

The headline release blocker was a pyroute2 regression threatening Flamingo. Balazs Gibizer (gibi) flagged bug 2109396 in os-vif: pyroute2 0.9.1 caused intermittent failures during interface attach/detach. The team had bumped to 0.9.4 believing it fixed the issue, but the problem persists there too. Gibi asked the requirements team whether they can roll the dependency back to <0.9 this close to release, or whether the only viable path is rolling forward on stable/2025.2 and master once an upstream fix lands. This is an unresolved dependency-management decision with direct release-timing implications — worth watching if you run anything touching os-vif.

Relatedly, Stephen Finucane requested a Requirements Freeze Exception (RFE) for openstacksdk (at elodilles's request) to release 4.7.0. That release resolves a minor bug currently blocking a bump to openstacksdk's minimum version before Flamingo ships; 4.7.0 is otherwise mostly new features, bug fixes, and test cleanup.

The TC weekly summary also recorded a milestone: the cross-project goal to migrate from WSGI scripts to module paths is now complete across all OpenStack services — relevant context for the documentation gap discussed below.

Development & Technical Decisions

Two threads converged on the WSGI migration's loose ends. Thomas Goirand (zigo) raised that while distros are moving Debian packages from wsgi-script= to module= entries, the correct module path (e.g. magnum.wsgi.api:application) is documented nowhere consistent — sometimes only buried in a release note, and for Magnum the note merely says the old form is deprecated without naming the replacement. He's pushing for an OpenStack-wide standard so distros and operators don't have to reverse-engineer each project from devstack.

On Kolla, Amir Hossein Ahmadi made a focused call for reviews to reinstate Swift support for 2025.2. Swift was dropped from Kolla months ago over role-modernization needs, lack of maintainership, and CI failures; Amir volunteered to fix all three, reports the Zuul jobs are now consistently green, and has stepped up as maintainer. With the release deadline close, the remaining step is getting the dev-swift-modernization patch series reviewed and merged in time.

Mohammed Naser asked the Neutron/TaaS team to split the tap-as-a-service client into a separate package, since the combined package currently drags in all of Neutron just to install the CLI — a packaging cleanup with no apparent objection.

Operators raised several concrete issues. Karl Kloppenborg reported live migrations never completing memory copy (memory stuck at 100% remaining across many minutes), and Thamanna Farhath asked separately about migration delays, "no valid host found" under high overcommit, and using a dedicated migration network. Other operator threads covered Cinder snapshot deletes stuck in "error deleting" with rbd.PermissionError on a Kolla-Ansible + Ceph setup, a Nova CPU flag (ibpb) rejected as invalid despite the host supporting it, retrieving overcommit-adjusted hypervisor capacity via the API, and Masakari HACluster containers crash-looping. An "OpenStack wide MariaDB 11.8 problem" thread (Sean Mooney) also continued — a recurring compatibility theme this period.

Community & Events

Governance was preoccupied with leaderless teams for 2026.1 "Gazpacho." Goutham Pacha Ravi (gouthamr) posted separate PTL-volunteer calls for Charms, Vitrage, and Venus, all of which still lack a PTL nominee. The TC's R-3 summary noted that the proposed retirement of the Monasca project team remains under active discussion — including whether to fork it into a new opendev namespace under a volunteer maintainer — and that the TC is adding metadata to flag inactive projects as a health signal for operators.

The PTL and TC election polls remain open until September 17 at 23:45 UTC, covering four TC seats and the Horizon PTL race. Election officials reached out to voters who hadn't opted into CIVS emails; if you can't find your ballot, contact them promptly.

Sean Mooney proposed Watcher core-team changes: immediate removal of six long-inactive cores (last contributions ranging from 2018 to 2022), with advance notice that additional low-activity cores face removal at the start of 2026.2 unless they re-engage during 2026.1. Jimmy McArthur asked whether anyone is still working on or interested in Freezer, probing the project's activity level.

On events, Kendall Nelson issued the final call for the October 2025 virtual PTG (October 27–31): teams must complete the signup survey by September 14 at 07:00 UTC, and registration is free and open. The Public Cloud SIG held its bi-weekly meeting on September 10.

Heads Up / Action Needed

• Flamingo dependency risk: Track the pyroute2 0.9.x regression (os-vif bug 2109396) if you depend on interface attach/detach reliability; the rollback-vs-roll-forward decision is still open.
• Reviews wanted: The Kolla Swift reinstatement series (topic dev-swift-modernization) needs reviews before the 2025.2 deadline.
• Vote by Sept 17, 23:45 UTC: TC (four seats) and Horizon PTL polls close; check for your CIVS ballot.
• PTG signup by Sept 14, 07:00 UTC: Teams must complete the survey to get a slot at the Oct 27–31 virtual PTG.
• PTL volunteers needed: Charms, Vitrage, and Venus are leaderless for Gazpacho.