The Big Picture

Two threads define this week, and both are about OpenStack's plumbing rather than its features. First, a developer-facing question with broad implications: what to do about the aging PasteDeploy/WSGI middleware stack as the Eventlet migration churns through oslo.service. Second, a steady run of operator pain — Galera deadlocks, Nova failing to come back after a clean shutdown, Cinder retype storage bloat — that captures the day-to-day reality of running OpenStack 2024.1/2025.1 in production. Governance was light as the summer slowdown set in, but the 2025.2 "Flamingo" feature freeze (August 28) stayed firmly on the horizon.

Development & Technical Decisions

The most strategically important thread came from Stephen Finucane: "What are we going to do about pastedeploy usage?" As part of the Eventlet migration, oslo.service is in heavy flux, and some WSGI routing code was inadvertently deprecated and then un-deprecated. Long term, that code doesn't belong in oslo.service. Finucane prototyped an oslo.wsgi package but found it would depend on Paste, PasteDeploy, Routes, and WebOb — all minimally maintained or effectively unmaintained. His question to deployers: how important are the paste.ini files in practice? Not every service uses them, and he hasn't heard complaints from those that don't. The alternative he floats is for projects to define a static middleware set (with a few config knobs) rather than carry the pastedeploy framework forward. This came up ~18 months ago without resolution, so operators with opinions on whether they rely on paste.ini customization should speak up now — the answer will shape post-Eventlet WSGI architecture.

The TC weekly summary (R-11, by Goutham Pacha Ravi) put the community eleven weeks from the Flamingo release and pointed again at the August 28 feature freeze. Two governance items of note: shade and os-client-config are now deprecated (their functionality has long lived in openstacksdk, and no new releases are planned barring critical bugs), and all CLA references are being removed from governance docs and manuals in favor of the Developer Certificate of Origin — contributors are asked to file bugs or small doc patches where stale CLA references linger. Consensus is also emerging on the relaxed Eventlet-removal timeline. On infrastructure, CentOS/Rocky 10 and Debian Trixie node images are expected from zuul-launcher soon.

Forward-looking hardware work surfaced via Fujitsu's Taketani, who asked the Nova team about third-party CI requirements for the Arm CCA (Confidential Compute Architecture) feature, including how much parallelism a third-party CI should target given Nova's ~20–30 patches/day, ongoing maintenance expectations, and a request for SEV-ES code review (a CCA dependency). Prerequisites (libvirt/QEMU/OS upstreaming) are planned for around April 2026.

Security

A membership-and-elections deadline doubles as an access-control housekeeping item: as part of the OpenInfra Foundation's transition under the Linux Foundation, individual memberships must be renewed. The summary noted a July 19 deadline to participate in the OIF Board elections, and — more relevant to most readers — a August 20 membership-renewal deadline to contest or vote in OpenStack's own PTL/TC elections, which run August 27 through September 17. Don't let a lapsed membership silently disqualify you.

Heads Up / Action Needed

• Weigh in on the pastedeploy/WSGI direction if your deployment depends on paste.ini middleware customization — silence will be read as "safe to drop."
• Flamingo feature freeze: August 28 (about eleven weeks out as of this window).
• Renew your OpenInfra Foundation membership by August 20 to vote or run in the PTL/TC elections.
• shade and os-client-config are deprecated — migrate to openstacksdk.
• Neutron drivers meetings on July 18 and 25 are canceled (no agenda, PTL on PTO).

Operator Troubleshooting

The week's operator threads read like a field guide to mid-life OpenStack clusters. The most active was a newcomer's set of "very basic" questions (Leroy Tennison) about whether OpenStack can adopt existing KVM/QEMU hypervisors and live-migrate VMs — drawing five participants and a useful orientation discussion for those evaluating OpenStack against Proxmox. Winicius Allan reported a Galera deadlock storm: Keystone and nova-conductor hitting DBDeadlock/lock-wait-timeout errors while blazar-manager exhausted its connection pool (QueuePool limit of size 1 overflow 50 reached), partially mitigated by bumping blazar's max_pool_size. Franck Vedel described a painful case where a Kolla-Ansible Epoxy cluster wouldn't bring Nova back after a clean shutdown for electrical maintenance, with nova-compute failing to register and a JSON decode error during reconfigure. Yuta Kambe (Fujitsu) closed out a Cinder retype investigation, confirming a patch that lets volumes retype without inflating backend (Ceph) usage across empty, attached-bootable, and detached-bootable cases.

Other reports: a Neutron DHCP question about pinning the metadata static-route IP in the qdhcp namespace; an Ironic/sushy confirmation that Redfish VirtualMedia under the newer /Systems/.../VirtualMedia/ path is already supported (the standard moved it from Managers); and recurring Kolla-Ansible CloudKitty GUI 404/rating failures (DNS resolution of the cloudkitty host and an invalid hashmap "compute" service). Rodolfo Alonso's Neutron bug report for week 28 flagged a High networking-sfc breakage from the OVS agent crashing, plus medium OVN metadata and binding-profile validation bugs, all with patches assigned.

Community & Events

Kees Meijs volunteered to take over chairing the Public Cloud SIG from Felix Kronlage-Dammers, and will send meeting invitations going forward. Ildikó Vancsa gave notice of a short Bitergia metrics dashboard outage on July 17 for an OpenSearch cluster update. As elsewhere this month, several meetings thinned out for the northern-hemisphere summer break.