Cluster Autoscaling

Phase Г.4 of the GA roadmap: from CP, click Cluster → Add node → CP provisions a VM on Hetzner / DigitalOcean and the node auto-registers within ~3 minutes. No SSH, no copy-token-then-curl.

Walkthrough

1. CP → Cluster → Add node.
2. Provider — Hetzner / DigitalOcean (Fly.io / AWS later).
3. Token — paste API token. Verifies on submit; stored encrypted in cc_settings (sensitive prefix autoscale.); reusable for later provisions.
4. Region — dropdown of ListRegions.
5. Plan — table of ListPlans with vCPU / RAM / Disk / monthly cost.
6. Hostname + SSH key — optional pre-existing SSH key in the provider for break-glass console access.
7. Confirm — preview “We will create a cax21 VM in nbg1 for ~€8.21/mo. Provisioning will take ~3 min.”
8. Run — SSE-backed progress: provisioning → bootstrapping → registering → ready.
9. Done — green check + “Open node detail” CTA.

How it works

CP → Provider.Provision(spec)             — creates VM with cloud-init user-data
                ↓
            (~30 s)                       — VM boots, cloud-init writes /etc/quazzar/registration.env
                ↓
       install.sh --register=…            — runs the standard installer + POSTs registration heartbeat
                ↓
         Node.Register(token)             — CP validates the one-time token + binds the node to the tenant
                ↓
              ready                        — node visible in the Cluster page

Cloud-init bundle:

#cloud-config
write_files:
  - path: /etc/quazzar/registration.env
    permissions: '0600'
    owner: root:root
    content: |
      QUAZZAR_REGISTER_URL=https://cp.quazzar.cloud/cp/api/v1/cluster/register
      QUAZZAR_REGISTER_TOKEN={one_time_token}
      QUAZZAR_TENANT_ID={tenant_id}
runcmd:
  - curl -fsSL https://get.quazzar.cloud/install | bash -s -- --unattended --register=/etc/quazzar/registration.env

The one-time token has a 30-min TTL; failed bootstraps after that auto-decommission the partial VM.

Idempotency

• Hetzner: native Idempotency-Key header.
• DigitalOcean: quazzar-idempotency:<hash> tag on the droplet.

A retry of the same job_id never creates a second VM.

Pricing

The wizard shows monthly cost per plan during selection. Hetzner exposes prices via API; DO does not directly — we ship a hand-curated price table updated quarterly. Banner: “Prices indicative. Confirm with provider.”

Plans

License gates: cluster_autoscale (Team+), cluster_autoscale_node_cap (Team=5 / Enterprise=∞). Manually-registered nodes (install.sh run by hand) don’t count toward the cap.

Cost & safety

• Monthly budget per tenant via autoscale.<tenant>.budget_cents setting; provisioning blocked above it.
• Token scoping — Hetzner: Read & Write; DO: droplet:create + droplet:read + droplet:delete minimum. Docs walk operators through provider IAM.
• Auto-decommission — failed bootstraps after 30 min auto-delete the partial VM.

API

GET    /cp/api/v1/cluster/providers                         → list
POST   /cp/api/v1/cluster/providers/{name}/verify           { token }
GET    /cp/api/v1/cluster/providers/{name}/regions          ?token=…
GET    /cp/api/v1/cluster/providers/{name}/plans            ?token=…&region=…
POST   /cp/api/v1/cluster/nodes                             → 202 { job_id }
GET    /cp/api/v1/cluster/jobs                              → list
GET    /cp/api/v1/cluster/jobs/{id}                          → status + progress
POST   /cp/api/v1/cluster/jobs/{id}/cancel                  → 204
DELETE /cp/api/v1/cluster/nodes/{id}                        → tear down + revoke
POST   /cluster/register                                    — public; one-time token

Admin-only. Provider tokens never returned to the frontend after creation.

Roadmap follow-ups

• С.4.b: Fly.io + AWS EC2 + GCP Compute adapters.
• Auto-scaling on metric thresholds.
• Multi-region cluster fabric.
• Bring-your-own-cluster (Vultr, Linode, etc) via a generic cloud-init mode.