v0.3.0

What's Changed

• Enhance CI by @ADWasHere in #26
• Add major and minor versioning by @ADWasHere in #27
• Added auto support branch by @ADWasHere in #28
• Add structured logging by @ADWasHere in #34
• Improve logging functionality with MDC by @ADWasHere in #35
• Implement logging convention and add missing logging by @ADWasHere in #36
• Enhance logging functionality by @ADWasHere in #38
• Add lag logging and adjust documentation by @ADWasHere in #39

Full Changelog: v0.2.0...v0.3.0

v0.2.0

What's Changed

• P4.2 schema typ for jsonb filtering by @ADWasHere in #9
• Added a validator to incoming events by @ADWasHere in #10
• P5 multi tenant/p5.1 by @ADWasHere in #18
• Add air quality schemas and views; update Docker configuration by @ADWasHere in #19
• P5.2.1/migration tool by @ADWasHere in #20
• P5.2 - Added tenant-per-schema by @ADWasHere in #21
• P5.3 - Make readmodel tenant aware by @ADWasHere in #22
• Make Views tenant-aware by @ADWasHere in #23
• P5.6 Prepare release v0.2.0 by @ADWasHere in #25

Full Changelog: v0.1.0...v0.2.0

v0.1.0

v0.1.0 — First usable realese

Refrax turns one event-sourced log into multiple standard-conformant projections
(native, NGSI-LD) that stay consistent by construction. No field without a declared
domain meaning and a vocabulary binding can appear in an exposed representation —
the gate enforces that structurally, not by discipline.

This rekease is usable end to end: you declare a schema, ingest events, and read
them back through views and projectors, served from real read models.

What you can do today

• Ingest events into an append-only log: POST /v1/events (idempotent per eventId).
• Declare an event type as JSON — its exposable fields, their roles and vocabulary
  URIs — loaded and validated at startup. Anything undeclared is internal by default.
• Structural leak guarantee: internal payload fields (deviceDbId, userId, …)
  can never appear in a projection, in any format.
• Views narrow what a consumer sees (data minimisation) and declare queryable axes;
  a view can only shrink the schema, never widen it.
• Projectors: native and NGSI-LD (with a @context derived from the schema).
  Views and projectors are orthogonal — any view renders in any format.
• Read side (CQRS): a persisted, incremental consumer maintains reading_latest
  (current state) and reading_series (a TimescaleDB hypertable). Reads are served
  from the read models: /latest, /stream, /series.
• Disposable & replayable: wipe the read models and rebuild from the log — full or
  per-entity. Ingest is decoupled: the log keeps accepting while the read side lags,
  crashes or rebuilds.

Quick start

docker compose up -d          # TimescaleDB + PostGIS
./mvnw quarkus:dev            # app on :8787

# ingest
curl -X POST http://localhost:8787/v1/events -H 'Content-Type: application/json' -d '{
  "type":"AirQualityReading","eventId":"11111111-1111-1111-1111-111111111111",
  "observedAt":"2026-07-02T10:00:00Z",
  "payload":{"sensorId":"sensor-42","metric":"PM2.5","value":12.3,"unit":"ug/m3","deviceDbId":999}
}'

# read (deviceDbId never appears)
curl 'http://localhost:8787/v1/views/air-quality-full/latest?sensor=sensor-42'
curl 'http://localhost:8787/v1/views/air-quality-full/latest?sensor=sensor-42&format=ngsi-ld'

Requirements

• JDK 25
• Docker (for the database, and for the test suite via Dev Services)
• TimescaleDB + PostGIS (brought up by docker-compose.yml)

Not yet (pre full-version scope)

• No authentication/authorisation — endpoints are open. Do not expose publicly.
• Single tenant (tenant_id is fixed); multi-tenancy is not implemented.
• GDPR enforcement (crypto-shredding) is not built — the seam is kept open (read models never materialise a personalData field in cleartext).
• PostGIS is enabled but the geo read model / location axis are not built yet.
• Reads are eventually consistent (projection lag ~seconds).
• The API and storage layout are unstable and may change without notice.

Feedback and design discussion welcome — the most valuable contribution is to the idea.