Telemetry Pipeline

Telemetry in DataHub follows a publish-subscribe pipeline designed for durability and real-time evaluation.

Pipeline overview

1. Adapter publishes JSON event to NATS
2. JetStream stream DATAHUB stores the message (7-day retention)
3. telemetry-worker persists history + updates Redis latest values
4. rule-worker evaluates matching rules → creates alarms → enqueues actions
5. action-worker delivers notifications
6. Console reads latest values from Redis and history from PostgreSQL

Event contract

Every adapter publishes to subject datahub.telemetry.<plantCode>.<deviceCode>:

{
  "eventId": "uuid",
  "tagPath": "gebze/press-line-01/s7-1500-01/motor.current",
  "plantCode": "gebze",
  "machineCode": "press-line-01",
  "deviceCode": "s7-1500-01",
  "tagCode": "motor.current",
  "value": 12.7,
  "valueType": "DOUBLE",
  "unit": "A",
  "quality": "GOOD",
  "source": "SIMULATOR",
  "sourceTimestamp": "2026-07-21T12:00:00.000Z",
  "metadata": {}
}

See Adapter contract for full field definitions.

Storage layers

Layer	Technology	Purpose
Latest values	Redis hash `datahub:latest`	Real-time console display
Device liveness	Redis `datahub:lastseen:<devicePath>`	CONNECTED if seen within 15 s
History	PostgreSQL `TelemetryEvent`	Time-series queries and exports
Rate metrics	Redis counters	Operations Center throughput KPIs

Active tags only in live views

The platform distinguishes live data (Redis latest cache) from history (PostgreSQL):

Inactive tags do not update Redis or receive new history rows — adapters may still publish, but the worker skips them
Operations Center, UNS Explorer, and device Latest values show active tags only by default
Deactivating a tag clears its entry from the live Redis mirror

History already stored before deactivation remains queryable. See Plants, machines, and devices — active tags.

The telemetry worker batches PostgreSQL writes for higher throughput. Inactive tags are filtered before persistence. For production scale options (batch size, optional plant-scoped consumers), see Production deployment.

Idempotency

eventId is unique in PostgreSQL. Duplicate NATS deliveries are skipped — adapters can safely retry publishes.

Control subjects

Configuration changes trigger reloads without restarting workers:

datahub.rule.changed — rule-worker refreshes its in-memory rule cache
datahub.device.changed — orchestrator reconciles adapter processes

External adapters

Adapters can be written in any language. Publish JSON to NATS using the event contract above. In-process drivers (SIMULATOR, MQTT) are managed by the adapter orchestrator.