The Big Picture

A genuinely quiet week — 17 threads, no security advisories, and a lot of single-message project notices — as the community heads into the December slowdown. The signal worth pulling out: a deliberate round of legacy cleanup (old Unmaintained branches headed to EOL, puppet-vitrage retirement, an Ironic driver deprecation) and a recurring operator-strategy question that resurfaces every so often — should you run PostgreSQL in production? We're at R-16/R-17 in the countdown to 2026.1 "Gazpacho," with 2025.2 "Flamingo" trailing deliverables being wrapped up this week.

Database Strategy

"Using PostgreSQL in production." Jiatong Shen opened the week's most-discussed thread (10 messages, 7 participants) asking about the pros and cons of PostgreSQL versus MariaDB for OpenStack, motivated by PG's fast-growing community and Chinese-vendor PG-compatibility offerings, plus a question about online migration tooling for legacy environments. This is a perennial and somewhat sensitive topic: OpenStack formally dropped PostgreSQL support years ago because maintaining a second database backend was deemed too much testing overhead (a point Clark Boylan cited in his recent process essay). Operators weighing PG should go in clear-eyed that it is not the upstream-tested path — the active discussion is worth reading, but the corpus doesn't show a changed upstream stance.

Development & Technical Decisions

Old Unmaintained branches are going EOL. Előd Illés (on behalf of the stable/unmaintained effort, and noting Ericsson's sponsorship as maintainer) announced he's initiating End of Life for the oldest unmaintained branches — starting with Victoria and up through Yoga — with transition patches coming the following week. The driver is decay in CI: the ubuntu-bionic node type is gone, py36 jobs have disappeared, and tempest/grenade jobs are failing, making meaningful testing infeasible. He explicitly pushed back on the idea of merging to these branches without tests, to avoid silently shipping regressions.

More retirements and deprecations. Takashi Kajinami proposed retiring puppet-vitrage, since vitrage was marked inactive this cycle and likely won't see a 2026.1 release or distro packaging. Jacob Anders announced the iRMC hardware type (Fujitsu PRIMERGY) is deprecated in 2026.1 and slated for removal from master in 2026.2 — its third-party CI has been broken since 2019 and the vendor is unreachable; affected operators should plan migration to Redfish (or, with limits, the legacy IPMI driver). Brian Haley again asked for help making neutron-dynamic-routing work without eventlet — the migration is still stuck and needs someone familiar with the project.

Skyline OIDC bug. Taavi Ansper reported a likely bug in Skyline under kolla-ansible 2025.2: OIDC logins fail with a request-ID header pattern-mismatch (X-Openstack-Request-Id empty, failing the ^req-... regex). A fix was proposed; oddly, 2025.1 doesn't set the header either yet still works, which remains unexplained.

Heads Up / Action Needed

• Old Unmaintained branches (Victoria→Yoga) heading to EOL — if you depend on backports there, this is your notice; transition patches are imminent.
• iRMC hardware type deprecated in 2026.1, removal from master in 2026.2 — start migration planning now if you run Fujitsu PRIMERGY via iRMC.
• EC2/S3 token middleware safe to remove. Arnaud Morin confirmed his team isn't using the keystonemiddleware ec2token/s3token code, clearing one objection to deprecating it.
• Freezer IRC meeting time poll (Dmitriy Rabotyagov) is open until December 15 for anyone interested in reviving regular Freezer development syncs.

Community & Events

The TC summary (Goutham Pacha Ravi) confirmed the just-merged SIG-governance and os-net-config resolutions and the completed Monasca deliverable retirements, and renewed the call for TC/PTL election officials (#openstack-election on OFTC). The Public Cloud SIG met December 3 (bi-weekly). On the operator help front, the usual mix appeared: a Neutron L3 HA / OVS thread on packet loss and VRRP route-flapping after restarting openvswitch-vswitchd on the active controller, a Keystone-with-389ds LDAP question about detecting expired passwords during bind, and the Octavia "No Route to Host" amphora health-check issue (resolved by setting octavia_network_type: "tenant" in kolla-ansible). With several teams already on holiday cadence, expect the lists to stay light into the new year.