Stratum

Operating it

A cloud is a thing somebody runs. This is that side.

The console has an administration side in three layers: metal, substrate, platform. Click a picture to enlarge.

The Stratum console: Administration home
Administration home What is healthy, what needs a hand, and the three layers.

Metal

Nodes, disks, pools, the network fabric.

  • The Stratum console: Nodes
    Nodes Every node with its role, readiness and load; drain and evacuate.
  • The Stratum console: Capacity
    Capacity How many more of each shape the cluster can place.

Substrate

Cluster HA, libvirt, OVN, Postgres, Ceph, reconcilers, health.

  • The Stratum console: Reconcilers
    Reconcilers Every converging loop: what it owns and when it last ran.
  • The Stratum console: Ceph
    Ceph Ceph health, OSDs, pools, capacity and recovery.
  • The Stratum console: Platform health
    Platform health Checks over every substrate component, with their alarms.

Platform

Organizations, quotas, the catalog, identity, audit.

  • The Stratum console: Organizations
    Organizations The tenants: who they are and what they hold.
  • The Stratum console: Quotas
    Quotas Limits per project per resource, against what is used.
  • The Stratum console: Database catalog
    Database catalog The engine versions and classes tenants pick from.
  • The Stratum console: Policy simulator
    Policy simulator Would this principal be allowed this action, and which policy decided?
  • The Stratum console: Audit
    Audit Every call, hash-chained, filterable, exportable.
  • The Stratum console: Alerts
    Alerts What is firing, with silences and notification channels.

Day to day

Each is also an API call and a CLI verb.

Onboard a tenant Create the organization and project, apply a quota tier.
Watch the fleet Nodes, capacity, storage health and the platform-health pack on one home page.
Add a node A join code, the same ISO on the new machine, an approval. A cluster is 3 to 5 nodes.
Take a node out Drain it and instances move off; bring it back and it reports ready.
Publish what tenants use Instance types, global images, database engine versions, floating-IP pools.
Answer “why was this refused?” The policy simulator names the policy that decided.
Get paged when it matters Alarm rules deliver to webhooks and push channels; silences for known noise.
Prove what happened A verifiable audit trail, exportable as CSV.
Back up the cloud itself Database, secrets store and credentials go to a target outside the cluster. Off until you set one.
Upgrade A signed release: plan, start, rolled node by node, one command to roll back. 0.7.0 starts the upgrade chain; 0.7.0 to 0.8.1 is the first upgrade.

Running it for others

The demo is this: one install and a site that gives each sign-up a tenant.

Where the operator docs live

Install, cluster management, upgrades and runbooks are in the documentation.