Connectors¶
Event sources, message sinks, and bridges to the outside world. Each connector ships both halves where it makes sense — drop in source-only, sink-only, or pair them for a round-trip.
-
Sub-microsecond IPC + UDP. Live subscribe or archive replay. Single-host or LAN.
-
File tail + append-only sink with size/time-based rotation.
-
Chronicle Queue + Map: persistent, memory-mapped, microsecond-latency.
-
Producer + consumer for cross-process / cross-host messaging.
-
UDP multicast for LAN-scoped pubsub without a broker.
Choosing a connector¶
The fast rule:
| Need | Connector |
|---|---|
| Tail a JSON or text file, replay later | connector-file |
| High-throughput persistent log on disk | connector-chronicle |
| Cross-process messaging via Kafka broker | connector-kafka |
| LAN broadcast / discovery / heartbeats | connector-multicast |
| Sub-microsecond IPC, cold-start replay | connector-aeron |
If you outgrow one tier, swap to the next without changing your processor — connector-file → chronicle → kafka is a common progression as throughput rises.
Operational defaults shared across connectors¶
- All file paths auto-create their parent directories.
- All
tearDown()calls are idempotent. - Source-side
doWork()is configurable via an idle strategy on the agent runner. - Sink-side
sendToSink()is single-writer (the dispatcher serialises calls).
See Operational guide for the full production checklist.