TOADSource ↗

Four tools, clear responsibilities

An understandable platform.

TOAD composes mature open tools instead of hiding the cloud behind another permanent control plane. Each layer has one job and one recovery story.

Small but complete

One useful node

The same routing, certificates, monitoring and application contract on the smallest practical footprint.

  • 1 manager
  • Direct ingress
  • Best for evaluation and small services
Internet
Public IP
Manager 1
Traefik
Apps
Observe

The layers

  1. Heat

    Own the resources

    Networks, security groups, servers, floating IPs, load balancing, and delegated DNS remain declared in OpenStack.

  2. Ansible

    Configure the hosts

    Docker installation, hardening, Swarm membership, ingress, and monitoring use narrow idempotent playbooks.

  3. Swarm

    Run the services

    A small orchestration surface provides rolling updates, secrets, configs, placement, health, and rollback.

  4. Traefik

    Route the applications

    Labels become HTTPS routes through one shared overlay, without direct application ports.

Update availability is declared

Best effort

The one-replica, lowest-cost path starts the replacement first and rolls back failures, without promising uninterrupted traffic.

Rolling

TOAD requires two replicas, a health check, one-at-a-time start-first replacement, and a monitored automatic rollback.

Maintenance

Single-owner and local-state services declare a stop-first interruption instead of hiding it behind a zero-downtime claim.

For an existing healthy public service, deployment continuously probes every declared endpoint while Swarm updates it and saves timestamped JSON evidence. Rolling acceptance fails if any request fails. Traefik keeps the HTTPS route while Swarm's virtual IP owns changing backend membership. Protecting against a server failure additionally requires replicas on independent nodes.

Availability boundaries

Ingress

Traefik OSS uses one certificate-state writer. Managers and Octavia are resilient; that task is not active/active.

Local data

Named volumes are single-node state. Availability comes from an explicit storage profile and tested encrypted recovery.

Provider outage

In-cluster monitoring cannot see a total cloud failure. A credential-free external probe runs elsewhere.