Compute environments

A VegaFlow compute environment supplies isolated capacity and network placement for connections, QuickFlows and pipelines. Definitions remain independent from compute, so an environment can stop, start or resize without changing what a flow means.

Create an environment

POST .../vegaflow/clusters
Content-Type: application/json

{
  "cluster_name": "production-ingestion",
  "size": "small",
  "minimum_scale": 0,
  "maximum_scale": 4,
  "auto_stop_minutes": 15
}


Open **VegaFlow → Clusters** and choose a clear name, an available node type and lifecycle settings. Use the fields in the current cluster form or API schema.

| Setting | Requirement |
| --- | --- |
| Node type | Required; select from the available choices. |
| Minimum nodes | At least `1`. |
| Maximum nodes | At least `1` and greater than or equal to minimum nodes. |
| Auto-scale | Allows capacity to vary within the configured node range. |
| Auto-resume | Allows a stopped cluster to become available when work needs it. |
| Auto-stop | Allows idle shutdown; the auto-stop time must be positive when enabled. |

Choose the region and private network path that can reach every system referenced by the flow. Keep production, development and regulated workloads in separate environments when they require independent capacity or access policy.

## Start and stop

```http
PUT .../vegaflow/clusters/{cluster_id}/start
PUT .../vegaflow/clusters/{cluster_id}/stop

Start and stop requests are idempotent. A successful start request means the desired state was accepted; wait for running before discovery or execution. A stopped environment can remain selected in a definition and will resume or fail according to the run’s readiness policy.

Displayed status can lag a lifecycle transition. Wait for it to settle before submitting a conflicting start or stop action.

Choose a size

Workload	Starting point	Watch
Periodic database sync	Small, auto-stop	source throughput and destination commit time
Many concurrent streams	Medium	parallel stream limits and source connection caps
Large backfill	Compute or memory optimized	bytes read, transform memory and destination throughput
Python/SQL pipeline	Match task profile	peak task memory and concurrent tasks
ML training	Dedicated accelerated profile where available	dataset staging and checkpoint volume

Auto-stop

Auto-stop considers active flow work, not browser sessions. Set a positive window longer than the normal gap between related scheduled jobs. With auto-resume enabled, a new run can resume the selected environment; include startup time in short schedule intervals and latency objectives.

Permissions and readiness

Collection permissions control listing and creation. Instance permissions control visibility and actions on an individual cluster. Give flow operators the access needed to use the cluster and manage its lifecycle. See access control.

If configuration is rejected, check the node range and auto-stop time. If a saved cluster does not become ready, check node type availability, capacity, status updates and permissions before changing its configuration.

Deletion

An environment referenced by an active QuickFlow, pipeline or schedule cannot be deleted. Repoint or archive those definitions first. Run history keeps the environment identity used by completed work.