Skip to main content
  • Kind: Observation
  • Version: 1
  • Fleet Manager owner: Fleet Manager squad, #fleet-manager-chat (individual claimant pending)
  • Counterpart owner: unassigned, Infrastructure Engineering (IaaS substrate); whether one owner covers all deployments or one per deployment set is to be confirmed with them
  • Last reviewed: 2026-09-09

​
What flows

The per-satellite openstack-agent (src/fleet_manager/agents/openstack/, ADR-0036) polls Ironic nodes every 60 s and mirrors provision and power state into device conditions (ProvisioningHealthy, MaintenanceClear, Enrolled, InValidationPool, InstanceAllocated, CustomerAllocated, CustomerReady) from which the hub derives operational_status (ADR-0024). It sweeps assigned nodes every 5 min to rerun Ironic reconciliation and detect orphaned Nova instances, and resyncs impairment state every 15 min, writing and removing the Placement CUSTOM_CUSTOMER_READY trait for quarantine and release. Workflows call the same services through src/fleet_manager/core/clients/openstack/: enrolment, cleaning and inspection against Ironic; instance create for image-node and SLURM adoption against Nova; the Ironic-direct deploy path (ADR-0042) for standalone-Ironic sites. Application-credential access rules are the contract in OpenStack Application Credentials.

​
What breaks if this changes

An Ironic API change or a missing access rule fails enrolment, cleaning and deploy workflows at runtime. A Placement outage leaves quarantine and release half-applied.

​
Enforcement

  • Principal: ironic-agent-<site> and temporal-worker-<site>.
  • Mechanism: towards Fleet Manager, FM OAuth clients (private-key JWT) in the hub chart satellite roster; towards OpenStack, the per-site application credentials in clouds.yaml held in the site 1Password vault (see OpenStack Application Credentials). A Keystone 401 or a credential rotation is the OpenStack side, not the Fleet Manager client.
  • Satellite roster client ironic-agent-<site> (scopes device:update, device-condition:create, operations:read|write) and temporal-worker-<site> in deployment/charts/hub/values.yaml.
  • Access rules: the application-credentials page above; the PR checklist in AGENTS.md requires it to move with any new Ironic, Nova or Placement call.
  • Settings: src/fleet_manager/settings/agents/openstack.py.

​
Gaps

  • No written notice period from Infrastructure Engineering for OpenStack upgrades.
  • Standalone-Ironic sites have no Nova, so the Placement trait path does not apply there; the file does not yet describe that variant per site.

​
Change log

  • 2026-09-09 v1: created; describes present behaviour.