Graph, pipeline, and Compose

Three related files with different jobs. The graph describes connection intent; the pipeline resolves it; Compose runs the supported services.

From intent to containers

Start from graph.sample.yaml. Maintain your installation graph alongside the other YAMLs. Generation expands it from enabled cameras, models, and collection sessions.

graph.yaml · you own

Templates for nodes, inputs, outputs, edges, conditions, and queue settings. Edit it when changing supported pipeline paths.

pipeline.yaml · generated

The concrete nodes and connections for this installation. Disabled cameras and inactive conditional branches are omitted.

pipeline.compose.yaml · generated

The Docker entry point. Includes service Compose files with images, commands, mounts, endpoints, restart policies, and health checks.

The processing path

Each enabled camera gets its capture service. Models may be shared; tracking and enrichment remain scoped to cameras. With tracking disabled, inference can feed enrichment directly.

01Camera
02Inference
03Tracking
04Enrichment
05Rules

Delivery and optional branches

Rules send accepted events and alerts to row builder, and alerts to notifier. Monitor sends health metadata to row builder. Collector joins original camera frames with inference detections.

01Row builder
02Persistent buffer
03HTTP ingest
04Branch database

One entry point, separate service files

Keep the generated files together. Use the umbrella for the whole installation, or select a service name from its service list.

Generated files

config/.generated/
  pipeline.yaml
  pipeline.compose.yaml   # umbrella: project, volumes, includes
  camera.compose.yaml
  inference.compose.yaml
  tracking.compose.yaml
  enrichment.compose.yaml
  rules.compose.yaml
  row_builder.compose.yaml
  buffer.compose.yaml
  ingest.compose.yaml
  notifier.compose.yaml
  monitor.compose.yaml
  collector.compose.yaml