The Big Picture

With Gazpacho past feature freeze and racing toward the coordinated 2026.1 release, this week was about end-game release mechanics and the new Hibiscus (2026.2) Technical Committee taking shape. The R-4 countdown began, deliverables that hadn't yet released were chased down, and the freshly elected TC was confirmed. Underneath the release machinery, the same theme that has run through the whole month persisted: a parade of CI breakages from upstream Python tooling churn — this week tox, packaging, and oslo.service all took their turn — keeping maintainers in firefighting mode at the worst possible time.

Releases & Announcements

2026.2 "Hibiscus" TC results, confirmed. Ian Y. Choi posted (to both openstack-discuss and openstack-announce) the five elected Technical Committee members: Amy Marrich (spotz), Dan Smith (dansmith), Goutham Pacha Ravi (gouthamr), Michal Nasiadka (mnasiadka), and Dmitriy Rabotyagov (noonedeadpunk).

PTL voting is underway. Polls for the contested PTL races — Barbican and Horizon — are open until March 18, 23:45 UTC via CIVS, with the usual eligibility (a commit in the team's deliverables over 2025.2–2026.1) and the CIVS email opt-in requirement. Mauricio Harley (Barbican) made his case for re-election; the Skyline situation was messier, with Wu Wenxiang noting the team missed the formal PTL window over the Chinese New Year holiday but intending to keep driving the project, prompting a TC nudge about onboarding more core contributors.

Release countdown reached R-4 (Mar 02–06, per Előd Illés). The pressure to finalize is on: the prior week's reminder still stands that horizon and several Ironic deliverables (bifrost, ironic-ui, networking-baremetal, networking-generic-switch) must cut releases before March 12 to make the final Gazpacho cut.

Development & Technical Decisions

CI churn, again — three fresh breakages. Radomir Dopieralski reported that an urllib upgrade in global requirements broke Selenium, knocking out Horizon's integration tests at a sensitive pre-release moment; the team's stopgap is a locally downgraded constraints file. Takashi Kajinami flagged that tox 4.46.0 (which needs virtualenv 20.39.0) conflicts with the upper-constraints virtualenv pin (20.38.0), breaking cinderclient/heatclient functional jobs with 'PythonSpec' object has no attribute 'machine', and advised against rechecks while the bump is debated. Brian Haley chased a No module named 'packaging.pylock' failure in neutron-functional where packaging 26.0 was somehow pulled in despite a 25.0 pin. And Bernard Cafarelli's Neutron bug-deputy report led with a critical issue: oslo.service 4.5.0 broke the Neutron periodic and maintenance workers (bug 2142122), with several related medium/low findings from the same release.

Neutron debates pinning OVN in the functional gate. Jakub Libosvar revived the question of whether the functional job should keep tracking the moving OVN LTS branch (24.03, nearly two years old) or pin a specific OVN hash. The status quo blocks features that need newer OVN — his BGP work required skipping many tests — and risks breakage from bad LTS backports. A previous pin request was denied to preserve the LTS policy; he's pushing to change it. Open debate.

The Ceph-down I/O-error question recurs. Thamanna Farhath (2023.1 + Ceph, 3-node) asked how to pause or block VM writes when Ceph loses quorum to avoid guest filesystem corruption — guests keep writing and throw I/O errors while storage is unavailable. This is the same operator pain point that surfaces periodically; the upstream stance is that guest I/O errors are the intended back-pressure and Nova won't auto-pause VMs on storage loss.

A few more technical threads. Dominik Danelski opened a Nova design question about whether a failed batch-scheduled instance should fail and clean up all siblings or just the offending one — the current _cleanup_build_artifacts behavior is inconsistent with both interpretations, blocking his Placement-allocation cleanup fix (review 968446). Mathieu Gagné moved to remove TungstenFabric support from openstack-helm given the project's closure and 5–8-year-old images. Vincent Godin reported a Keystone app-credential reactivation delay after TTL expiry, and Bence Romsics raised a multicast/allowed-address-pairs inconsistency in Neutron.

Heads Up / Action Needed

• R-4 release crunch: finalize RC work; horizon and the listed Ironic deliverables must release before March 12.
• PTL voting (Barbican, Horizon) open until March 18, 23:45 UTC — confirm CIVS opt-in.
• Manila NetApp NAE feature-freeze exception requested by Gireesh Awasthi; Manila Gazpacho bug squash runs March 2–5 (kickoff March 2, 1500 UTC).
• Masakari leadership unresolved: with the nomination window closed without a PTL, the team is in conversation with the TC about whether it can continue active for 2026.2.
• Avoid rechecks while the tox/virtualenv and packaging conflicts are sorted.

Community & Events

The Ops Radio Hour held its Feb 27 session (HA architectures with Tyler Stachecki, backups, automation) and published survey results showing strong support for alternating time slots and a bi-weekly/monthly cadence. The Public Cloud SIG met, and the Digital Sovereignty WG continued building toward its landing page. The TC's R-5 summary flagged an ongoing concern about Requirements team health and the lack of a confirmed Deployment Project Liaison team — a gap that risks forcing a TC appointment the committee would rather avoid through community engagement — plus continued work to stabilize CI on older stable branches.