▤ Stackers Network

August 02, 2025

Stackers Network Digest — August 02, 2025

30 threads · 69 messages · openstack-discuss

The Big Picture

Two formal retirements moved into motion this week: the TC's proposal to retire Monasca ahead of the Flamingo release, with current core/PTL support, and Jeremy Stanley's announcement that the RefStack service has been decommissioned and the interop/refstack/refstack-client repos will be retired following the 2023 board resolution that relaxed RefStack as a trademark prerequisite. Both clear the deck before the August 28 Flamingo feature freeze, which the TC R-9 summary kept squarely in focus. Alongside the cleanups, two operationally-charged debates ran in parallel: secrets management for Kolla-Ansible deployments and the boundary between vendor "callhome" telemetry and operator control.

Releases & Announcements

Goutham Pacha Ravi's R-9 TC summary reconfirmed: feature freeze 2025-08-28; PTL/TC nominations 2025-08-06 to 2025-08-20; voters must be active OIF members and have opted in via CIVS before August 20. The TC approved revised Eventlet-removal milestones, still aiming to drop Eventlet entirely in the 2027.2 ("J") release, and the aetos project now has a canonical metric-storage service type. Goutham also opened the formal Monasca retirement proposal, listing 17 repositories slated for retirement before Flamingo; reviving Monasca later remains possible if a maintainer group steps up.

The RefStack decommissioning is effective immediately per the 2023 board minutes; trademark licensing has not used RefStack since March 2024.

Ghanshyam Maan proposed adding sean-k-mooney and tosky to Devstack core, with the plan to add them by Friday, August 8 absent objections. Carlos Silva announced the Manila Flamingo mid-cycle on August 6–7 (14–16 UTC; etherpad).

Development & Technical Decisions

The week's most-discussed thread was Satish Patel's Kolla-Ansible secrets management discussion (10 messages, 6 participants), which started with "Why not Ansible Vault?" and grew into a substantive defense-in-depth debate about Barbican/Castellan integration. The conclusion-in-progress: Ansible Vault, application credentials and Castellan-backed secrets each play different roles, and operators want clearer guidance — and a recent kolla-ansible commit is a concrete data point. Two earlier threads from elnaz cl (1, 2) underscored the demand for an oslo.config {{ secret:UUID }} pattern via Barbican.

Brian Rosmaita posted a thorough critique of the IBM SVf "callhome" Cinder patch and the NetApp precedent — concluding that there's no existing precedent for vendor "phone home" in Cinder drivers, that the NetApp send_ems_log_message traffic is operator-controlled (so likely fine), and that the IBM patch needs clearer commit messaging if data isn't actually sent off-host. The discussion has direct relevance to the GDPR/operator-trust questions raised earlier this cycle.

Jeff Yang opened a thread documenting severe VM-to-VM throughput degradation on OVN — host-to-host iperf3 reports a clean 12 Gbits/sec with zero retransmits, but VM-to-VM through the overlay drops to ~4 Gbits/sec with hundreds of retransmits per second. The thread connects to Karl Kloppenborg's broader tenancy network MTU discussion, which is collecting operator experience with jumbo frames on tenant networks vs. 1500 byte northbound.

Vasyl Saienko asked whether Nvidia ASAP² OVS hardware offload on ConnectX-6 DX is "a myth" — switchdev set up per the OpenStack offload docs and Nvidia docs, but no offload visible for VLAN or VXLAN. The thread is collecting real-world configs from people who got it working.

Chang Xue (Bloomberg) asked the Neutron team whether operators and third-party ML2 mechanism drivers can safely extend binding:profile with custom keys (blueprint) and have those values persist across upgrades and migrations — a concrete request for guidance on namespacing/schema and supported extension patterns.

Vincent Godin continued chasing Cinder NetApp multipath on NVMe-over-TCP, and separately pinpointed the cause to the NetApp initialize_connection returning only target_portals[0] instead of all four discovered paths — operators on C800 arrays are getting single-path mounts where they expect four.

Renu Rani's Tacker investigation was confirmed resolved by removing logger from Q_ML2_PLUGIN_MECHANISM_DRIVERS — verified by Zuul build 48d74246.

TRB PREM KUMAR reported a Horizon/CloudKitty interaction that crashes /admin/hashmap/ with HTTP 404 after creating an invalid hashmap service named compute — useful for Kolla operators turning on CloudKitty, and a candidate for input validation. He also flagged Ceilometer central/compute containers stuck unhealthy (linked Kolla bug 2081190).

Heads Up / Action Needed

  • August 6: PTL/TC nominations open; August 20: nominations close and OIF membership/CIVS opt-in deadline.
  • August 28: Flamingo feature freeze.
  • Object now to Monasca retirement if you have a maintainer plan (proposal thread).
  • RefStack is gone — purge dependencies on refstack.openstack.org.
  • IBM SVf "callhome": the Cinder community has not endorsed off-host telemetry by default — watch patch 951829 for resolution.
  • Kendall Nelson is recruiting operators to contribute to the OpenStack for AI whitepaper; join the AI working group list.

Community & Events

Ian Y. Choi highlighted CMU student contributions to the Zanata-to-Weblate migration — three students (Furqan Saeed, Simon Yu, Jana Hussien) shipped 34 unit tests for i18n tooling and Zuul script migration work over a six-week mentorship (demo 1, demo 2; patches 953745, 955982, 955439). The next Public Cloud SIG meeting is August 13 at 07:00 UTC. Jakub Libosvar's Neutron bug-deputy report covering July 14–28 noted one Critical (now being investigated) and several Incomplete bugs needing reporter follow-up.