▤ Stackers Network

June 21, 2025

Stackers Network Digest — June 21, 2025

21 threads · 60 messages · openstack-discuss

The Big Picture

Two big things this week, both with hard deadlines: the DCO is now ten days away (effective 2025-07-01 on review.opendev.org — no Signed-off-by means rejection at upload), and Cyborg is officially in the danger zone. Ghanshyam Maan's call for new Cyborg maintainers follows Sean Mooney's earlier alarm: the gate has been broken for weeks, the unblocking patch 949772 has no response from current cores, and the TC has proposed in governance change 952798 to mark Cyborg inactive. If marked inactive before Milestone-2, there will be no Cyborg release in the 2025.2 cycle. Operators and downstream consumers who care about FPGA / accelerator support should speak up now via the TC's R-15 weekly summary or directly on the gerrit review.

The eventlet-removal saga continued — gibi's thread on dropping eventlet grew to 15 messages and seven participants and looped in governance change 952903. The free-threaded CPython discussion (3.14 whatsnew, 3.15 threading) is informing what "after eventlet" looks like long-term.

Releases & Announcements

The VMware Migration Guide is live. Kendall Nelson and the VMware Working Group published a detailed migration-from-VMware playbook. This is positioned as the deeper follow-up to the earlier whitepaper — meaningful at a moment when the OpenStack-as-VMware-alternative narrative is having a real industry moment.

Development & Technical Decisions

Heat plans to remove standalone API services. Takashi Kajinami's operator-facing notice gives the timeline: deprecate heat-api and heat-cfn-api (the eventlet-based standalone scripts) in 2025.1 via backport, remove during 2025.2. Operators must move to running Heat under uwsgi or apache+mod_wsgi. Object if you'll be affected — Takashi wrote: "If we hear no objections/concerns for a week then, we aim to merge the removal during this cycle."

Custom RBAC trap, episode three. Thamanna Farhath's follow-up post now lays out three rule variants for os_compute_api:servers:index:get_all_tenants — all of which list servers correctly but return 404 on detail/action. A separate Cinder thread from Serhii Kryhin (GoDaddy) on SystemReader + all_tenants in Cinder is the same theme: the filter is stripped at cinder/volume/api.py:667-674. The pattern across services is consistent enough that a cross-project SystemReader audit seems overdue.

Tags are not consistently case-sensitive. Rodolfo Alonso Hernandez raised an excellent cross-project bug: Neutron, Nova, Octavia define the (resource, tag) primary key without sanitization, but a case-insensitive collation (e.g. utf8_general_ci) silently breaks the API-WG tag guidelines that say tags are case-sensitive. Real symptoms vary by service: Neutron 409s, Nova silently drops the duplicate tag. See Launchpad 2114819. Decision needed: enforce case-sensitive collation, or normalize tags to lower case everywhere.

Nova rebuild + virtual-size pre-check missing. Joao Vicentini Fracarolli (StarlingX) filed Nova bug 2114855: the API accepts a server rebuild where the image's virtual size exceeds flavor disk, fails in nova-compute, and leaves the VM in ERROR. The API check only compares on-disk size. A reasonable fix is API-level validation.

Cinder/Ceph deferred deletion leaks .deleted rbd images. Vinicius Rocha Lobo (StarlingX) detailed in bug 2114853 that volumes created from Glance images leave behind protected snapshots and .deleted rbd images even with enable_deferred_deletion, rbd_flatten_volume_from_snapshot, and deferred_deletion_purge_interval all set. This is a meaningful operational issue on Caracal-class deployments.

Arnaud Morin asked Nova about combining multiple PCI alias types in a single flavor — e.g. pci_passthrough:alias='h200a:4|h200b:4' for two GPU pools. Not currently possible; see PCI passthrough docs.

Janaka Wickramasinghe (FC + Windows Failover Cluster): Cinder volumes attach as device='disk', but Windows VPD Page 0x83 needs device='lun'. No supported way today. Anyone running Windows Failover Clusters on OpenStack should join this thread.

Templateless provisioning thread closed graciously. Prem Mishra accepted that OpenStack's design is image-first and is shifting his approach to pre-baked Glance images with user_data. A small but pleasant outcome.

Heads Up / Action Needed

  • Cyborg risk: respond by Milestone-2 (releases.openstack.org/flamingo/schedule.html) or accept no 2025.2 Cyborg release.
  • DCO: T-minus 10 days. Configure git commit -s.
  • Heat operators: speak up on standalone-API removal this week.

Community & Events

The Ops Radio Hour on June 27 at 13:00 UTC will cover switches & OVN transitions, RabbitMQ upgrades, and Summit Forum session planning. Etherpad: openstack_ops_radiohour; Matrix room: #openstack-ops:opendev.org.

Ildiko Vancsa shared the metrics analysis the OpenInfra community managers are running — Review Efficiency Index, Time-to-Review/Merge, patchsets/review, opened-vs-closed, maintainer counts across 13 project teams over five release cycles (Bitergia dashboard, extraction script). The goal is a baseline against which contributor-experience interventions can be evaluated.

The Manila weekly meeting is cancelled for June 18 (US/Brazilian holidays).