▤ Stackers Network

November 29, 2025

Stackers Network Digest — November 29, 2025

20 threads · 42 messages · openstack-discuss

The Big Picture

A short, US-Thanksgiving-thinned week with no advisories and modest traffic, but a few decisions worth tracking. The through-line was maintenance discipline: the TC pushing toward formalizing SIG governance and wrapping up the Monasca retirement, plus a CloudFerro proposal to upstream a serious ephemeral-storage feature into Nova. Operators kept hammering at perennial problems — slow migration status updates, GPU passthrough on Q35 UEFI guests, and the long-running machine-identity gap. We're at R-18 toward 2026.1 "Gazpacho".

Releases & Announcements

The TC's R-18 weekly summary from Goutham Pacha Ravi reports the November 27 deadline for "trailing" deliverables shipping 2025.2 "Flamingo" releases — covering packaging, deployment and lifecycle-tooling projects. Most other teams are now in the thick of their Gazpacho roadmaps; Nova and Manila have populated the release calendar with their cycle-specific deadlines. The TC continues consolidating SIGs under openstack/governance and is wrapping up the Monasca project's retirement after concluding there's not enough community interest to meet OpenStack's maintenance and release standards. (Note: with the TC's new APAC-friendly monthly meeting slot still bedding in, attendance has been light — give it another cycle.)

Security

No new advisories this week, but the post-quantum cleanup begun the prior week continues — Artem Goncharov's proposal to drop unused crypto dependencies from openstack/requirements is the active concrete action (passlib, scrypt, python-gnupg). Tracking on the post-quantum wiki.

Development & Technical Decisions

CloudFerro proposes upstreaming SPDK-based ephemeral storage for Nova. Karol Klimaszewski outlined two features his team has developed and would like to contribute upstream:

  1. SPDK-based ephemeral storage backend — a new [libvirt]/images_type=spdk_lvol value that backs instance local disks with SPDK logical volumes on local NVMe, exposed to guests as vhost-blk source_type='vhostuser' disks via SPDK's JSON-RPC. They report working create/delete/cold-migrate/shelve/snapshot/unshelve flows; live migration is on the roadmap.
  2. Multiple ephemeral storage backend handling — a new [libvirt]/supported_image_types option plus a libvirt:image_type flavor extra-spec, so a single compute can offer (and the scheduler can target) different backends. Depends on #1.

He's asking whether a blueprint + spec per feature is the right next step. This is a substantive contribution; community feedback will shape how (and whether) it lands.

Arbitrary QEMU args — should Nova allow them? Nathan Harper revived the question after a Sean Mooney comment that Nova has a policy of disallowing raw QEMU args. Their operator-team needs specific flags on certain VMs, and they're currently maintaining a local Nova patch since libvirt-hooks proved unworkable. The thread is asking for "unsupported YMMV" escape-hatch options rather than a true API surface.

Slow migration status updates in Nova 2025.1. Nguyễn Hữu Khôi reported that openstack server migration list shows running for ~12 minutes after a migration completes — sparked a longer thread including a related issue Nell Jerram is debugging where a migrated instance refuses traffic post-live-migration despite Nova reporting success.

Machine identities, still no upstream answer. Sean Mooney reiterated that there's no built-in equivalent to AWS IAM roles (docs) or GCP instance identity; experiments with JWTs exist but were never upstreamed. Nathan Harper pointed to a BBC Nova fork commit as the kind of work happening privately. A topic worth a real spec.

Blazar — lease-data notifications proceed. Matt Crees confirmed Nitin Gupta's Blazar patch for extending the notification payload with the full lease data will be reviewed; the conversation is now about backporting to stable/2025.1 and unmaintained/2024.1.

Barbican — please review the SoftHSMv2 + CKM_AES_KEY_WRAP_PAD rewrap fix. First-time contributor kaoru watanabe is seeking review on barbican/+/965443 (Launchpad #2130214), which fixes barbican-manage hsm rewrap_pkek in their production configuration. The patch has been sitting for nearly a month.

Heads Up / Action Needed

  • Q35 UEFI + GPU passthrough on Windows guests is broken on host-model/Icelake-Server. Joel McLean shared detailed reproduction notes for NVIDIA A100 passthrough on Emerald Rapids hosts (OpenStack 2024.1, libvirt 8.0.0, QEMU 6.2.0): Ubuntu guests need a kernel command-line workaround, but Windows VMs go into a boot loop with Icelake-Server or Icelake-Server-noTSX while Broadwell-noTSX-IBRS and Cascadelake-Server-noTSX work. Linked QEMU's preferred CPU models and Ubuntu libvirt bug 2106812. Useful prior art for anyone in the same boat.
  • OpenStack-Ansible RabbitMQ workaround (continued): George (lmihaiescu@gmail.com) reports that the suggested group_vars override for signed_by doesn't work because roles/rabbitmq_server/vars/debian.yml takes precedence; he edited that file directly. If you applied the workaround from the prior week, double-check the actual variable that's winning.
  • Tacker, Kolla and Ironic have all pre-announced meeting cancellations for the end-of-year holiday window — Ironic will skip Dec 22 and 29, resuming January 5.

Community & Events

The November 21 Ops Radio Hour recap (Ildikó Váncsa) is notable: the group agreed OpenStack is "overextending the capabilities of RabbitMQ" but couldn't identify a clearly better replacement, with oslo.messaging driver plug-ins floated as a longer-term path. A related OpenInfra Summit Europe session was recommended; the meeting recording is here. The next Ops Radio Hour is December 19 at 1300 UTC. The Neutron VPNaaS multi-tunnel question from the prior week got a response from Lajos Katona pointing at existing bugs 2120316 and 2078845.