DevOps OpenStack
Agent Workflow (MANDATORY)
Before advising or changing anything, run in parallel (one spawn_agent call each):
- explore-codebase - find existing
globals.yml,passwords.yml, inventories,openstack_user_config.yml,clouds.yaml, Helm values overrides - research-expert - re-read https://releases.openstack.org/ and the deployment tool docs for the target series (versions move every six months)
- mcp__context7__query-docs - library
/websites/openstackfor the exact command or config key
After writing config or templates, run sniper.
Overview
| Area | What this skill gives |
|---|---|
| Service map | Which service owns what, and the path of openstack server create through Keystone → Nova → Placement → Glance/Neutron/Cinder |
| Releases | Current, SLURP and next series; which upgrade jumps are tested |
| Deployment | Kolla-Ansible vs OpenStack-Ansible vs OpenStack-Helm, plus a 3-controller baseline |
| Day-2 | clouds.yaml, read-only discovery, project bootstrap, quotas, images, networks, volumes, instances |
| Upgrades | Backup, prechecks, per-tool upgrade commands, verification |
| Troubleshooting | Logs per tool, the four classic failure chains, control-plane health |
Critical Rules
- Read-only discovery first on a live cloud:
openstack service list,openstack compute service list,openstack network agent list,openstack volume service list- then plan, change, verify (operations-cli.md). - Destructive commands are proposed, never run on your own:
openstack server delete,volume delete,network delete,router delete,project delete,project cleanup,kolla-ansible destroy,kolla-ansible stop,nova-manage db purge. Show the exact effect and the object list, then wait for the user's go. - Upgrade only along a tested path: adjacent series, or SLURP to SLURP. Never skip from a non-SLURP release (release-upgrade-slurp.md).
- Back up before any series upgrade or DB-touching action: deployment config (
/etc/kollaor/etc/openstack_deploy) and the MariaDB data. Nova cannot downgrade its database. - Odd-sized quorum: MariaDB-Galera and RabbitMQ need 3 (or 5) controllers; 2 gives no HA.
- Never invent a key or flag: every
globals.ymlkey and CLI option in this skill was checked against the 2026.2 docs or source; anything else is verified first or reported as unknown. - Secrets stay out of output:
passwords.yml,user_secrets.yml,clouds.yamlpasswords and application credential secrets are never printed, committed or pasted (devops-secrets).
Versions
Pinned versions:
openstack: "2026.2 Hibiscus (current, released 2026-09-30)"
openstack-slurp: "2026.1 Gazpacho"
openstack-next: "2027.1 Indri (SLURP, planned 2027-03-24)"
kolla-ansible: "23.0.0 (2026.2)"
openstack-ansible: "33.1.0 (2026.1); 2026.2 not released on 2026-11-10"
openstack-helm: "2026.2.0"
python-openstackclient: "10.3.1 (2026.2 series)"
Reference Guide
Concepts
| Topic | Reference | When to consult |
|---|---|---|
| Service map | service-map.md | Which service does what, request flow, what to enable |
| Releases & SLURP | release-upgrade-slurp.md | Picking a series, planning an upgrade path |
| Deployment choice & sizing | deployment-choice-sizing.md | New cloud, choosing a tool, node layout |
| Kolla-Ansible | kolla-ansible.md | globals.yml, deploy, reconfigure, upgrade, MariaDB backup |
| OpenStack-Ansible | openstack-ansible.md | openstack_user_config.yml, playbooks, major upgrades |
| OpenStack-Helm | openstack-helm.md | OpenStack on an existing Kubernetes cluster |
| Neutron / OVN | networking-neutron-ovn.md | Provider vs tenant networks, OVN, floating IPs |
| Cinder / Ceph | storage-cinder-ceph.md | Volume backends, external Ceph wiring, RGW as object store |
| Operations CLI | operations-cli.md | clouds.yaml, discovery, day-2 commands, destructive ops |
| Troubleshooting | troubleshooting.md | Logs, No valid host, port binding, attach, RabbitMQ/Galera |
Templates
| Template | When to use |
|---|---|
| clouds.yaml | Admin + project + application-credential entries |
| kolla-globals.yml | Minimal 2026.2 Kolla-Ansible production globals.yml (OVN, external Ceph) |
| project-bootstrap.md | New project: user, role, quotas, network, router, security group, first server |
| heat-single-server.yaml | Heat stack: network, router, port, server, floating IP, volume |
| pre-upgrade-checklist.md | Gate before any series upgrade |
Quick Reference
export OS_CLOUD=mycloud-admin # entry in clouds.yaml
openstack token issue -f value -c expires # auth works
openstack service list # catalog
openstack compute service list # nova-* up/enabled per host
openstack network agent list # OVN controller / metadata agents alive
openstack volume service list # cinder-scheduler / cinder-volume up
openstack quota show --usage <project> # quota vs consumption
openstack server event list <server> # what Nova did to an instance
→ Full sequences in operations-cli.md and project-bootstrap.md.
Best Practices
DO
- Pin the deployment tool to the series branch or tag (
stable/2026.2, OSA tag33.1.0) and installpython-openstackclientwith the series upper constraints - Keep the whole deployment config in git; diff it against the new sample files at each upgrade
- Use application credentials in automation instead of user passwords
- Use OVN (
neutron_plugin_agent: "ovn"in Kolla, the OSA defaultml2.ovn) for new clouds - Schedule MariaDB backups and test a restore before you need one
DON'T
- Run
kolla-ansible upgradewith--limit(the docs report known bugs with it) - Put host-specific variables in Kolla
globals.yml(highest precedence, applies to every host) - Upgrade a non-SLURP release by two series at once
- Expect
enable_swiftin Kolla 2026.2 - the key is not in the sampleglobals.yml; use Ceph RGW (enable_ceph_rgw) - Treat
openstack project deleteas cleanup - it removes the project, not its servers, volumes or networks (project cleanup --dry-runfirst)
Related skills
- devops-storage-backup
- devops-iac
- devops-ansible
- devops-kubernetes
- devops-k8s-cluster
- devops-linux
- devops-networking
- devops-virtualization
- devops-observability
- devops-databases
- devops-secrets
- devops-sre-ops