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. Event lookup + ClickHouse medallion DSL.

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 → silver → gold; viewer DSL + promote API
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}'

Medallion (ClickHouse only): scroll to Medallion on the viewer, edit the DSL, then Promote all. That creates silver/gold tables + MVs and backfills from bronze. Gold charts export as SVG or an iframe at /widgets/{table}. Details: docs/medallion.md.

curl -s -X POST http://localhost:3000/api/medallion/promote \
  -H 'Content-Type: application/json' \
  -d '{"tier":"all","backfill":true,"dsl":"silver silver_orders from orders.events\n  customer:String = customerName\n  amount:Float64 = totalAmount\n\ngold gold_orders_daily from silver_orders\n  group event_date\n  orders:count\n  revenue:sum(amount)\n"}'

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);