The Big Picture

A quieter, operations-heavy week on the list, with the northern-hemisphere summer slowing governance activity but not the steady stream of real-world troubleshooting. The headline development for planners was the proposed 2026.1 "Gazpacho" release schedule landing for community comment, while the 2025.2 "Flamingo" cycle continued its countdown toward the August 28 feature freeze. Notably, a Neutron thread raised a potential VM traffic-isolation problem that operators should read closely.

Releases & Announcements

Előd Illés circulated the proposed cycle and release schedule for 2026.1 "Gazpacho", inviting comments on the milestone timing via the Gerrit review. If you care about how next cycle's deadlines fall, now is the time to weigh in on the patch rather than after it merges.

The TC weekly summary (R-10, by Goutham Pacha Ravi) noted the community is ten weeks from the Flamingo coordinated release and five weeks from feature freeze (2025-08-28). No new governance changes merged this week, though several remain under review (see below).

Development & Technical Decisions

The TC's open governance queue is the clearest signal of where things are heading. Proposals currently under review include retiring networking-midonet, removing Monasca from active projects, making the Eventlet-removal deadlines more acceptable for operators, and requiring TC candidates to declare their affiliation. Discussion also covered who should maintain the service-types-authority repo — some favored bringing in the openstacksdk-core team, but the prevailing view was that it should stay under TC oversight given its governance weight, with a renewed commitment to review changes more actively.

On the libraries front, Stephen Finucane's pkg_resources/setuptools PSA matured into a detailed writeup: the last Python 3.8-compatible setuptools doesn't normalize package names passed to pkg_resources.require, which now collides with OpenStack's (correct) normalization of wheel filenames. Symptoms are confusing "distribution was not found" errors. It bites release tooling that calls pkg_resources directly, and projects like Designate and Octavia that used the legacy "scripts" keyword; fixes are up for review.

A meaty operator-investigation thread stands out for its diagnostic value. The Tacker/Neutron regression (Renu Rani) was traced with real precision: the Kubernetes-based job failed because Ubuntu Jammy's default setuptools 59.6.0 lacks get_requires_for_build_editable (fixed by setting Global_Venv=True so Devstack uses setuptools 80), and the OpenStack-based job's "Internal Server Error" and "Neutron did not start" failures were narrowed to a specific Devstack commit (May 20) and a Neutron master commit (July 1). That's a model bug report for anyone chasing CI regressions.

Rodolfo Alonso's Neutron bug report for week 28 flagged a High issue — networking-sfc jobs broken by the OVS agent crashing — plus medium-severity OVN metadata and remote-managed vnic-type binding-profile validation bugs, all with patches in flight.

Security & Isolation

One thread deserves operator attention as a possible isolation issue rather than a routine question. Nguyễn Hữu Khôi reported that instances on the same VLAN and compute node can see each other's traffic, as if port mirroring were enabled, and that anti-affinity rules are the only workaround found so far. On VMware, this would be handled by toggling promiscuous mode on the port group; the OpenStack equivalent and root cause were still being discussed. If you run flat/VLAN tenant networks, it's worth verifying your own environment doesn't exhibit the same behavior. The recurring Cinder "call home" debate also continued, with IBM following up on its SVf driver callhome patch and the comparison to NetApp's ActiveIQ/EMS approach (enabled by default, capturing details from logs) still unresolved.

Heads Up / Action Needed

• Comment on the Gazpacho (2026.1) release schedule on the Gerrit patch if milestone timing matters to your team.
• Flamingo feature freeze: August 28 — five weeks out as of this window.
• Review the TC governance proposals (networking-midonet retirement, Monasca removal, Eventlet timeline relaxation, TC affiliation declaration) while they're open.
• Neutron drivers meetings on July 18 and 25 were canceled (no agenda, PTL on PTO); open neutron-specs still need review.

Community & Events

Ildikó Vancsa reminded the community of the Ops Radio Hour on July 25 (1300 UTC), with a packed agenda: OVN migration (and a documentation proposal), Swift compliance processes, GPUs, VMware migration, and vTPM behavior during live migrations and reboots. The group is also trialing Matrix for more synchronous chat between calls. Several teams scaled back for summer: the Tacker IRC meeting was skipped for a Japanese holiday, and Horizon canceled its July 23 and 30 meetings with about half the team on PTO. On the operator help side, recurring topics included Neutron metadata-agents working but reporting as not alive after a Caracal→Epoxy upgrade (Eugen Block), NVMe/TCP multipath only yielding a single path on NetApp backends, instance freeze during Ceph-backed snapshots on 2024.1, disabling a Cinder Availability Zone for legacy workloads, the Ironic networking-baremetal vs networking-generic-switch distinction, and a Kolla-Ansible proxysql failure on aarch64 (Raspberry Pi 5) due to an unsupported system page size. A University in Sri Lanka also put out a call for an OpenStack support provider to reduce single-person dependency on their Kayobe deployment.