Architecture Overview
FluidGrids is a managed SaaS product. There is no self-hosted option. This page explains how a workflow actually runs, so you can reason about latency, concurrency, and failure behaviour.
How a run executes
┌─────────────────────────────────────────────┐
│ Workspace public gateway │
│ (authentication, RBAC, routing) │
└─────────────────────────────────────────────┘
│
▼
┌─────────────────────────────────────────────┐
│ Workflow service │
│ ┌─────────────────┐ ┌─────────────────┐ │
│ │ GraphQL API │ │ Job publisher │ │
│ └─────────────────┘ └─────────────────┘ │
└─────────────────────────────────────────────┘
│
▼ job / event bus
┌─────────────────────────────────────────────┐
│ FluidGrids workers │
│ consume jobs → run each node in a │
│ sandbox (per-node microservice) │
└─────────────────────────────────────────────┘
All requests go through the workspace public gateway, which authenticates the caller and enforces RBAC; services are never called directly. The workflow service exposes the GraphQL API and publishes execution jobs onto a durable job/event bus. The workers tier consumes those jobs and runs each node in a sandbox — every node is its own microservice — reporting status and events back over the same bus.
Two consequences worth knowing:
- Runs are asynchronous and durable. Triggering a workflow enqueues a job. If a worker dies mid-run, the job is redelivered rather than lost.
- Nodes are isolated from each other. A node that hangs or crashes cannot take down the run engine or another workspace's execution.
What this means for you
| Concern | Behaviour |
|---|---|
| Per-node timeout | Enforced by the sandbox; long nodes are cut off, not left running |
| Concurrency | Bounded per workspace by your plan's limits |
| Retries | Failed jobs are retried according to the workflow's retry policy |
| Ordering | Per-run ordering is preserved; independent runs execute in parallel |
Availability and rollout
FluidGrids is continuously delivered — changes ship to production after passing automated checks, with no maintenance window and no action required from you. Live status and incident history: status.burdenoff.com.