The Big Picture

This was the home stretch before OpenStack 2026.1 "Gazpacho," which ships April 1. The release reached week R-2, release candidates are being tagged for cycle-with-rc deliverables, and the Technical Committee's focus has shifted to release-critical testing and to filling leadership gaps. The dominant developer-facing theme was CI and requirements turbulence: a Django security-support deadline forcing a freeze exception for Horizon, gate breakages, and a fresh look at how OpenStack consumes PyPI. Meanwhile, the PTL elections for Horizon and Barbican entered their final days.

Releases & Announcements

The release countdown hit R-2 (March 16–20) per Elõd Illés's post, and the TC's R-3 summary confirmed all cycle-with-rc deliverables would have published a release candidate by week's end to support testing for the April 1 coordinated release. Looking ahead to 2026.2 "Hibiscus," the TC has updated tested runtimes and is strongly recommending teams test Python deliverables against Python 3.14 — the unit-test job will run as non-voting for now, pending an OS that ships 3.14 natively (such as the upcoming Ubuntu 26.04 LTS).

Jeremy Stanley again pressed PTLs to finalize Gazpacho cycle highlights in openstack/releases; timing is especially tight this round because KubeCon EU falls the week before the release, so marketing needs the material early. Allison Price issued the bi-annual call for volunteers for the OpenInfra Live "Gazpacho" episode on Thursday, April 2 at 1400 UTC — historically the project's most popular webinar — and wanted the lineup locked by early in the following week.

Development & Technical Decisions

Django's security clock is forcing Horizon's hand. Radomir Dopieralski filed a requirements freeze exception to bump Django from 4.2.28 to 5.2.12, because security support for the 4.2 series ends next month and 5.2 is an LTS release with two more years of support. The Horizon team has been making the project Django 5.2-compatible this cycle, but gate disruptions kept the enabling patches from merging before the freeze. This is a genuine "ship-with-a-supported-dependency-or-not" decision the requirements team had to weigh against late-cycle risk.

CI cost and resilience drew real discussion. Radomir also opened a thread proposing a local PyPI cache/mirror inside the CI infrastructure, observing that package installation can eat more than half of a test run. Benefits floated: faster installs, resilience against PyPI/network outages, reduced load on PyPI, and finer control over available package versions (pinning setuptools, banning bad versions, staging pre-releases). It's an exploratory proposal looking for infra-knowledgeable owners, not a committed plan.

Neutron's OVN gate testing strategy is being reconsidered. Jakub Libosvar's thread on the OVN version used in the functional gate questioned the long-standing choice to always test against the LTS OVN. The current approach decorates BGP tests that require newer OVN schema to skip them — which Jakub now argues is wrong, since production code genuinely can't work with the older OVN and the test should fail to flag it. The two paths under debate: move the functional job to a newer OVN, or add an experimental job; either way, drop the skip decorator and blacklist those tests from the voting job.

Two PTG-bound technical proposals stand out. Jay Faulkner opened a cross-project Ironic/Nova discussion on nova-compute startup, which can take 15–20 minutes (or worse) for Ironic-managed nodes and forces multi-minute outages on every code deploy — a serious problem for in-place rebuild use cases. His ideas include incremental placement updates from Ironic, bulk placement API updates instead of N calls for N nodes, and possibly a staged-service handoff to avoid the outage; he wants to converge on a direction at the PTG and spec it for the "I" release, contingent on Nova core review commitment. Separately, Lajos Katona pushed a WIP Keystone patch (review 976618, bug 2138725) to make LDAP password-expiration handling work with Keystone's bind request, and is asking whether to take it to the PTG.

On the packaging fallout from RDO's RPM sunset, Neal Gompa weighed in that AlmaLinux contributors interested in OpenStack RPMs should engage upstream RDO rather than build a parallel effort.

Heads Up / Action Needed

• PTL elections for Horizon and Barbican were in their final days; the TC urged eligible Active Project Contributors (and OpenInfra Individual Members) to vote and to ping #openstack-election on OFTC if they couldn't find their ballot.
• Neutron bug deputy reports (Slawek Kaplonski for the week of March 2; a fairly quiet week) flagged a critical networking-sfc py312 breakage already fixed by Rodolfo, plus an unstable functional test being temporarily skipped to protect the gate.
• Core team changes: Douglas Viroel proposed adding Alfredo Moralejo to watcher-core and watcher-dashboard-core and removing four long-inactive cores (effective after the March 19 IRC meeting); Michał Nasiadka proposed Bertrand Lanson for the new kolla-reviewers group; Daniel Bengtsson opened a one-week vote to add Ghanshyam Maan to oslo-core.
• Operator gotchas worth knowing: a Keystone Epoxy upgrade failure with legacy $6$ password hashes (KeyError: '$6$'); a kolla-ansible Flamingo VPNaaS misconfiguration (needs an explicit service_provider in neutron.conf); and a persistent kolla all-in-one cannot fork child process libvirt failure that resisted all the obvious resource-limit checks (likely a cgroup v2 angle).

Community & Events

Ildiko Vancsa shared next steps from the Flamingo-cycle contributor/maintainer survey, surfacing recurring pain points — unclear or unmaintained review priorities, and inconsistent places where contributors look for them — and proposed teams experiment with making review priorities explicit and proactively communicating reviewer availability. The TC also celebrated Outreachy intern Rose Kimondo completing her work on the Manila UI, and asked the community to help retain her. The Digital Sovereignty WG and Public Cloud SIG continued their regular cadence.