▤ Stackers Network

December 27, 2025

Stackers Network Digest — December 27, 2025

9 threads · 12 messages · openstack-discuss

The Big Picture

This was the deep holiday week — between Christmas and New Year — and traffic on openstack-discuss was light: nine threads, mostly opened-and-not-answered operator questions, plus an update from the Masakari restart-and-revitalize effort. No security advisories, no releases, and no urgent action items. Operators looking ahead to January should still note the small but useful items: the Masakari team's restart plan now has a concrete cadence and merge rules, and a couple of Nova design questions (PCI alias + unified limits, rabbit-ping in heartbeats) are sitting on the list waiting for input when people are back at their desks.

Releases & Announcements

Nothing released this week. The most concrete community update is the Masakari restart kick-off summary from Sei Sano (NTT). The December 22 kick-off — held even though the PTL was absent — settled on bi-weekly IRC meetings on Mondays, 07:00–08:00 UTC (16:00–17:00 JST), with willingness to switch to weekly if attendance warrants. The team also adopted an explicit merge rule: patch authors and reviewers/mergers should be from different affiliations whenever possible, and Sei (as a core reviewer) will wait for an external +1 before merging. The short-term goal is project maintenance leading up to the Gazpacho release. The first meeting of 2026 is planned for Monday, January 5 at 07:00 UTC. (Etherpad, IRC log, poll thread.) Operators relying on Masakari-driven HA should plug in here.

Development & Technical Decisions

Nova: unified limits + PCI alias is double-counting. Masahito Muroi described a real configuration problem worth picking up in January. With report_in_placement=true, pci_in_placement=true, a [pci] device_spec mapping NVIDIA T4s to CUSTOM_NVIDIA_T4, and a [pci] alias referencing the same resource_class, the UnifiedLimitsDriver calculates the CUSTOM_NVIDIA_T4 quota correctly — but nova-scheduler issues an /allocation_candidates request with duplicated CUSTOM_NVIDIA_T4 resources (once globally and once as a granular resourcesbb775ad3-...-0=CUSTOM_NVIDIA_T4:1), causing scheduling to misbehave. He outlined two paths forward: have nova-scheduler drop duplicated requests, or introduce a placement:<RC> flavor extra-spec prefix to indicate the resource class comes from Placement. Real design question; needs feedback from Nova cores.

Cross-project heartbeat ping (still open). Thomas Goirand reposted the main-thread RabbitMQ ping discussion to keep it visible across the holidays. Karol Klimaszewski's Nova ephemeral-storage thread (CloudFerro's SPDK proposal) is also still on the list. Both deserve a closer look in January.

Heads Up / Action Needed

  • Masakari host-failure detection in a 3-node controller+compute cluster. Thamanna Farhath reports that on OpenStack 2023.1 (Antelope) deployed with Kolla-Ansible — Masakari (hostmonitor, instancemonitor, engine, api) + Pacemaker + Corosync on three combined controller+compute nodes, no pacemaker-remote, recovery_method=auto — process-level failures are detected correctly, but powering off a node or stopping corosync produces no HOST_DOWN notification and no evacuation, even though Pacemaker marks the node OFFLINE. Two existing reviews (masakari-monitors/+/922336 and +/937638) plus the consul-usage docs are the immediate reference material. With the Masakari team restarting (see above), this is a great test case for early-2026 attention.
  • Kolla / RabbitMQ stream-queue errors after upgrade. Taavi Ansper reports operational deployments littered with PRECONDITION_FAILED - stream queue 'q-reports-plugin_fanout' in vhost '/' does not have a running replica on the local node and "connection already closed" errors from oslo.messaging._drivers.impl_rabbit. Deployment works despite the noise. If you've seen this pattern post-stream-queue migration, share.
  • Heat templates inside Mistral workflows. elnazcloud@gmail.com asks for the correct format to pass a Heat template to the built-in heat.stacks_create action when CLI openstack stack create works fine but the workflow returns "Template not in valid format: The template is not a JSON object or YAML mapping". Anyone with Mistral-Heat integration experience, this one's open.
  • Ceilometer + Gnocchi: list-field iteration in event payloads. Nitin Gupta asks whether anyone has a workaround for Ceilometer not iterating over list fields like reservations in event payloads, which prevents producing per-reservation vcpus/memory_mb metrics into Gnocchi. Possible known issue.

Community & Events

A quiet week by design. The Masakari team's restart kickoff is the one piece of forward momentum: a 30-something-thread project is back on the calendar with explicit rules, a goal (Gazpacho-cycle maintenance), and an open invitation for observers — bi-weekly Mondays 07:00 UTC, with Japan OpenStack user community members already lined up to join. Wishing the OpenStack community a happy new year.