▤ Stackers Network

September 06, 2025

Stackers Network Digest — September 06, 2025

20 threads · 40 messages · openstack-discuss

The Big Picture

Flamingo (2025.2) is in its final weeks — Thierry Carrez marked R-3 in the release countdown, feature freeze has passed, and project teams are pivoting to release candidates. The dominant cross-project story is a newly surfaced MariaDB 11.8 charset/collation compatibility problem that touches the whole stack. Alongside it: a milestone moment for the eventlet-removal effort with Ironic, Barbican and Heat fully migrated, and a TC actively working through the consequences of low PTL nomination turnout for the 2026.1 cycle.

Releases & Announcements

Elod Illes reminded PTLs that 2025.2 cycle highlights need to be finalized this week so they can feed marketing prep. Thierry Carrez also flagged three cycle-with-intermediary deliverables that need a refresh before final release — Horizon (last released 25.4.0 on 2025-05-27), ironic-python-agent-builder (6.1.0, 2025-06-10), and networking-baremetal (6.6.0, 2025-06-10) — to avoid shipping outdated versions.

Goutham Pacha Ravi's TC R-4 summary confirms voting is open for the four TC seats and the Horizon PTL race, closing September 17 at 23:45 UTC. Election officials flagged that a chunk of the electorate hadn't opted in to CIVS emails — affected voters can opt in here. Four project teams have no PTL nominees: Monasca, OpenStack Charms, Venus, and Vitrage. The TC will work on appointments and/or retirement decisions.

The October 2025 PTG signup window is closing fast (sign up here) and registration is open at ptg.openinfra.dev.

Security

Herve Beraud's eventlet update noted that CVE-2025-58068 (CVSS 6.3) in Eventlet's WSGI parser — enabling HTTP request smuggling — has been fixed in eventlet 0.40 (specific release noted in the thread). Operators running affected versions should upgrade.

Development & Technical Decisions

The MariaDB 11.8 problem. Seunghun Lee, working on bumping Kolla's MariaDB, surfaced a sweeping compatibility issue: MariaDB 11.5+ changed the default collation for utf8mb3 (MDEV-25829), and OpenStack services are inconsistent about how they declare charset/collation on table creation. Services that explicitly set charset on tables (Nova, Keystone, Cinder, Designate) behave differently from those that don't (Glance, Neutron, Octavia, Placement), and Magnum is internally mixed — producing observable bugs like 2121797. The thread surfaced both upgrade-path risk and fresh-deploy inconsistency. Existing migration tooling such as Percona's pt-online-schema-change may help in places, and a galera_server patch is up. Expect this to become a Gazpacho-cycle goal.

Eventlet removal milestone. Herve Beraud reported that Ironic, Barbican, and Heat have completed their eventlet migrations, and Eventlet 0.40.2 introduced Python 3.14 support. Roughly 17% of impacted runtime code has been migrated, 13 projects are fully done, and over 332 patches have been submitted. Ironic's team published a detailed write-up of their migration. Futurist gained a DynamicThreadPoolExecutor to replace dynamic GreenPool, and oslo.service's backend system is now documented. The migration timeline at the governance level has been adjusted to align with Nova's dual-mode support. Ongoing work continues for Neutron, Nova, Manila, Magnum, Mistral, Watcher, Designate, and Trove.

Routed provider networking edge case. Nathan Harper hit a Nova/Neutron interaction limit that may surprise operators: in a multi-rack routed provider networking setup using SR-IOV per rack, scheduling correctly filters hypervisors by aggregate, but the PCI device request sent to nova-compute always carries the physnet of the same segment — pointing at a TODO in nova/network/neutron.py calling out "Additional work will be required to handle the case of multiple vlan segments associated with different physical networks." So despite the documentation describing this pattern, it may never have actually worked end-to-end. Worth confirming before designing on top of it.

Cinder driver process improvement. Nimesh Desai's thread on resurrecting a WIP cinder patch whose original author has left the IBM organization drew helpful guidance — abandon-and-reupload is the standard path when ownership can't be transferred. The IBM driver patch has been picked up under new ownership.

Adrian Jarvis is writing a Keystone patch to rehash deprecated password hashes on successful auth — a smoother migration path for old clouds upgrading through Rocky→Train→Epoxy than forcing user-side resets. Patch up; contributions welcome from anyone who would use it.

The Magnum team is proposing Scott Davidson as core reviewer for magnum-capi-helm — Jake Yip's +1 cites consistent useful review comments. Chang Xue also opened Nova bug 2121607 tracking the post-Caracal API latency growth.

Heads Up / Action Needed

  • Vote in the 2026.1 elections — polls close Sep 17 at 23:45 UTC. Check your spam folder, or opt in to CIVS emails.
  • Finalize Flamingo cycle highlights for marketing.
  • Refresh stale intermediary releases: Horizon, ironic-python-agent-builder, networking-baremetal.
  • Sign up for the October PTG (registration via the OpenInfra form).
  • The QA team is collecting PTG topics — add yours before Sep 12.

Community & Events

Clark Boylan flagged an Open Source Pavilion booth opportunity at the OpenInfra Summit 2025 — projects can book time to demo work and engage attendees. James Denton continued his ML2/OVN logical port statistics discussion, suggesting an RFE to extend OVN network logging beyond security groups to monitor single ports. Doug Goldstein, in a TC campaign post, laid out priorities: easier paths to contribution, better tooling and docs, and clearer routes from new contributor to core. Ghanshyam Maan announced he'll be on PTO until after the 2025.2 release — Tempest reviews may be slower.