▤ Stackers Network

July 19, 2025

Stackers Network Digest — July 19, 2025

21 threads · 53 messages · openstack-discuss

The Big Picture

Two themes dominated this week. First, an unusually rich set of operator pain points — a Galera deadlock under Keystone/Blazar load, an Epoxy cluster that wouldn't recover from a clean shutdown, Cinder retype efficiency between Ceph backends — that all touch real production deployments. Second, a maturing developer conversation about the future of pastedeploy in OpenStack services and the practical scope of the Eventlet-removal effort. The TC's R-11 summary kept the Flamingo feature freeze (August 28) firmly in view and confirmed shade and os-client-config are now deprecated in favor of openstacksdk.

Releases & Announcements

Goutham Pacha Ravi's R-11 TC summary marked shade and os-client-config as deprecated — both have long been superseded by openstacksdk, and no new releases are planned absent critical bugs. The summary reiterated the 2025.2 "Flamingo" feature freeze on August 28 and the synchronized election windows: nominations August 6–20, polling August 27 through September 17. Members must renew OIF individual membership by August 20 (00:00 UTC) to vote. The Monasca retirement proposal, Eventlet-removal milestone changes, and TC-candidate affiliation requirement remain open for review.

The Ironic team accepted HPE Superdome Redfish VirtualMedia under the /Systems/ path is already supported in sushy/ironic via bug 2039458.

Development & Technical Decisions

The biggest discussion thread of the week was Leroy Tennison's "Very basic newby questions" — what it takes to use OpenStack with existing KVM hosts vs. converting them. The responses pointed at the control-plane architecture docs and the Ops Guide on compute maintenance, and underscored that newcomers continue to find the docs landscape overwhelming.

Operator failure stories were the bigger signal. Winicius Allan's Galera deadlock thread saw Keystone logins fail with Lock wait timeout exceeded while Nova-conductor logged the same error and Blazar-manager hit QueuePool limit of size 1 overflow 50 reached. Bumping Blazar's max_pool_size from 1 to 2 cleared its symptom. The thread also surfaced a candidate root cause in Blazar's concurrent _process_event batching and references MariaDB MDEV-35124 and Nova bug 2116186 for the broader pattern.

Franck Vedel reported an Epoxy / Kolla-Ansible cluster that wouldn't return after a planned datacenter shutdown. After three days of triage nova-cell reconfigures fail at "Waiting for nova-compute services to register themselves" — a useful warning for anyone trusting that Epoxy will tolerate a full power cycle.

Yuta Kambe (Fujitsu) confirmed the Cinder retype fix — tests covering empty, attached-bootable, and detached-bootable volumes all retyped between Ceph HDD and SSD pools without growing backend usage, and instances boot cleanly afterward. On the CI front, Renu Rani and the Tacker team traced a Zuul regression to two specific commits in the Jun 29 – Jul 8 window: devstack commit 988ce885 and neutron commit 1971c4a2; Brian Haley engaged from the Neutron side.

Stephen Finucane reopened the pastedeploy question — whether to bootstrap an oslo.wsgi around minimally-maintained Paste/Routes/WebOb dependencies, or push services to drop the framework in favor of a static middleware chain. Operator feedback is being solicited.

Two specs/feature signals: Ryo Taketani shared details on the Nova Arm CCA confidential-computing feature following the 4/8 PTG, asking the Nova team for third-party CI guidance and an SEV-ES code review (libvirt/QEMU/OS prerequisites land April 2026). Thamanna Farhath asked whether /os-hypervisors will ever expose CPU/RAM allocation ratios via API — currently visible only in Horizon and nova.conf.

Rodolfo Alonso's Neutron bug report for week 28 flagged one high: networking-sfc master jobs broken by an ovs-agent crash, with patches in flight. Two mediums are tracked for remote-managed vnic-type binding (2116087) and OVN metadata breakage when allowed-address reuses the port IP (2116249).

Heads Up / Action Needed

  • Renew OpenInfra membership by August 20 (00:00 UTC) to vote in the 2026.1 PTL/TC elections; nomination window opens August 6.
  • Flamingo feature freeze: August 28. Land remaining work or seek freeze exceptions.
  • pastedeploy future: Stephen Finucane wants deployer input on whether to keep PasteDeploy stack alive or move services to static middleware.
  • shade/os-client-config deprecated: migrate to openstacksdk.
  • Epoxy cold-restart fragility: the Vedel report is worth tracking before you do scheduled-power-down maintenance.

Community & Events

Kees Meijs of Nefos stepped up to chair the Public Cloud SIG, taking over from Felix Kronlage-Dammers; the next meeting was July 16 at 07:00 UTC (meetpad). Brian Haley canceled the next two Neutron Drivers meetings (July 18 and 25) for PTO. Buddhika Godakuru of the University of Kelaniya put out a request for a Sri Lanka–area Kolla/Kayobe support provider — single-person bus-factor problem they want to insure against. Ildiko Vancsa flagged a Bitergia dashboard maintenance window on July 17, 10:00–12:00 CEST.