ANASy

Analytics Sync. Keep OLTP thin: publish fat events once to Kafka. Any number of sinks consume them.

OLTP  --EventPublisher-->  Kafka  --group A-->  ClickHouse
                           └────  --group B-->  DuckDB
                                    viewer  (look up event_id)

Kafka is the product boundary. ClickHouse and DuckDB are reference sinks, not the architecture.

Modules

Module	Layer
event-connector-starter	Core. Inject EventPublisher.
examples/starter-example	Sample OLTP app. Copy this pom.
sinks/clickhouse-batch-sink	Optional. Batch JDBC into ClickHouse.
sinks/duckdb-batch-sink	Optional. Upsert into a DuckDB file.
viewer/	Demo dashboard. Where is this event.

Java 21, Spring Boot 3.4, Maven. Group io.anasy. Viewer is Next.js.

Docs

Doc	Read when
docs/PLAN.md	Locked decisions
docs/hld.md	Architecture, fan-out
docs/lld.md	Starter, publisher, how to add a sink
docs/scaling.md	Partitions, groups, per-sink tuning
docs/medallion.md	Bronze now; silver/gold when a query is hot
docs/sinks/clickhouse.md	ClickHouse config, schema, queries
docs/sinks/duckdb.md	DuckDB file store, PK upsert
AGENTS.md	Agent rules

Quick path

Full stack (Kafka, ClickHouse, both sinks, sample OLTP, viewer):

docker compose -f examples/docker-compose.yml up --build

Open http://localhost:3000. Publish from the viewer, or:

curl -s -X POST http://localhost:8080/orders \
  -H 'Content-Type: application/json' \
  -d '{"customerName":"Ada","totalAmount":42.5}'

Host-only Kafka (no warehouses): docker compose up -d at the repo root.

OLTP usage — persist, then publish. publishAndWait when the caller must see the broker ack:

events.publish("orders.events", order.id().toString(), fatEvent);