June 21, 2025

Stackers Network Digest — June 21, 2025

21 threads · 60 messages · openstack-discuss

The Big Picture

The most consequential conversation of the week was the eventlet removal effort reaching the "when do we drop support?" stage — Balazs Gibizer's thread drew the widest engagement on the list, signaling that the community-wide migration off eventlet is moving from groundwork to hard decisions. Running alongside it: the DCO transition's July 1 deadline kept getting reinforced (R-15, Milestone-2 approaching), and the Cyborg project's viability came to a head with a formal call for maintainers and a proposal to mark it inactive. Operators, meanwhile, surfaced a fresh batch of policy/RBAC and storage edge cases.

Releases & Announcements

Goutham Pacha Ravi's TC weekly summary placed the 2025.2 "Flamingo" release 15 weeks out, with Milestone-2 a couple of weeks away. That milestone is the governance checkpoint where deliverables are vetted: teams deemed "inactive" won't have releases tagged, so the TC stressed that any community member can propose marking a team inactive before M2 to avoid shipping unmaintained code. The summary again pushed the DCO transition (sign-offs required from July 1) and asked PTLs to spread the word in team meetings.

Kendall Nelson announced the VMware Working Group's VMware Migration Guide is live — a deeper follow-up to the original whitepaper, meant for sharing with industry and prospective adopters, with Superuser case studies to follow.

Development & Technical Decisions

Eventlet removal is the story to watch. Balazs Gibizer's "[eventlet-removal] When to drop eventlet support" thread was by far the week's busiest (15 messages, 7 participants), tagged across release and TC — a sign the project is now negotiating concrete timelines and cross-project coordination rather than just per-team experiments. Relatedly, the heat team (Takashi Kajinami) detailed its plan to remove the standalone heat-api and heat-cfn-api services, which are blocked on eventlet because there's no good non-eventlet replacement for the standalone scripts; the strategy is to backport deprecation to 2025.1 and remove in 2025.2, forcing operators onto uwsgi/apache WSGI deployments.

Cyborg's maintenance crisis went public. Ghanshyam Maan issued a call for Cyborg maintainers after the gate broke, a fix sat unreviewed, and current cores stayed silent through two TC meetings. He proposed marking the project inactive — which would mean no Cyborg deliverables this cycle — and invited anyone using or willing to maintain it to step up.

Policy/RBAC limitations were a strong undercurrent. A long custom-roles thread (Thamanna Farhath) confirmed that across-project instance listing can be granted via policy, but detail and action calls still 404 without the admin role — a recurring enforcement gap. The Cinder team got a pointed question (Serhii K.) about why the SystemReader role can't use the all_tenants filter when listing volumes (the filter is stripped at the API layer), which blocks auditing and CMDB integration; Nova, Keystone, and Neutron already support analogous patterns. And Rodolfo Alonso Hernandez raised a cross-project tag API case-sensitivity problem: tags are spec'd as case-sensitive, but Neutron's lack of input sanitization plus table collation choices mean behavior diverges across projects (Neutron 409s, Nova silently drops the duplicate) — should the API enforce case-insensitivity or pin a case-sensitive collation?

Nova hardware and storage details drew solid discussion: an inquiry on attaching a Cinder volume as device='lun' (vs device='disk') for Windows Failover Cluster SCSI VPD page 0x83; a question on multiple PCI types with the same address and whether one flavor can request h200a:4|h200b:4; and continued back-and-forth on Arm CCA feature CI requirements (Ryo Taketani) — clarifying whether third-party CI on emulated vs bare-metal environments is acceptable and how long it must be maintained. A StarlingX-reported Nova rebuild bug (#2114855) noted the API accepts a rebuild when the image's virtual size exceeds flavor disk, then fails in compute, leaving the VM in ERROR; the question is whether the API should reject it up front. StarlingX also filed a Cinder/Ceph RBD bug where deferred-deletion leaves protected snapshots and .deleted volumes behind indefinitely.

Heads Up / Action Needed

  • DCO sign-offs required July 1 — from then, Gerrit rejects changes without Signed-off-by. Start using git commit -s now.
  • Milestone-2 is the inactivity checkpoint — Cyborg is at real risk of being marked inactive (no maintainers responding). If you depend on it, engage now.
  • Eventlet removal timelines are being set — projects with eventlet-bound services (notably standalone heat APIs) should track the migration and plan operator-facing changes.
  • Neutron bug-deputy report (Bence Romsics) flagged a high-priority randomly-failing functional test (bug 2113861, backport in progress) and an RFE for ports to reference their bound segment_id.

Community & Events

Ildikó Vancsa kept promoting the Ops Radio Hour (June 27, 1300 UTC) with OVN transition, RabbitMQ upgrades, and Summit Forum planning on the agenda, and reminded everyone the group is trialing Matrix (#openstack-ops:opendev.org). She also opened a thoughtful "bridging the gap between community and contributing orgs" thread, sharing OpenInfra's contributor metrics (Review Efficiency Index, time-to-review/merge, maintainer counts across 13 teams over five cycles) — explicitly framed as a baseline, not a leaderboard. Carlos Silva canceled the June 18 Manila meeting for the holiday.