← Interactive architectureSparkPilot

SparkPilot deployment

Open SVGDownload

Control plane, customer account, persistent state and external services.

SparkPilot runtime deployment diagram

Runtime flow

  1. The browser enters through HTTPS at the Application Load Balancer. Auth0 sign-in establishes the application session.
  2. Next.js proxies authenticated requests to the private FastAPI address. The API checks identity and resource permissions.
  3. For environment setup and run submission, the API commits the work and outbox record, then publishes its identifier to the appropriate SQS queue.
  4. The worker receives work, loads saved state and advances provisioning or workload submission. Each queue has its own dead-letter queue.
  5. API and worker processes share durable PostgreSQL records. The worker's reconciler saves the AWS job's observed state.
  6. SparkPilot assumes the customer's delegation or provisioning role. EMR workloads use their separate execution identity for scripts, data and logs.
  7. The agent thread calls the external model API over HTTPS. Python controls tool execution and action approval.
Deployment assumptions and verification gaps
Configuration, source evidence and AWS references

Source snapshot 0851cee, inspected October 16, 2026. The maps show the public HTTPS profile and agent-enabled worker configuration. They are source-reviewed diagrams; active DNS/CDN routing, replica counts, availability-zone placement and egress settings require a current deployment inventory.

Reviewed against the AWS Well-Architected Framework and official reference diagrams, accessed October 16, 2026. Official AWS icons from the July 31, 2026 package. Visual references: AWS Architecture Icons, ECS inbound networking, cross-account SaaS access and numbered reference-diagram example.