A realistic Australian core banking platform demonstrating enterprise microservices patterns — four Java/Spring Boot services wired together with local ACID transactions, Saga/Outbox/CDC messaging, full observability (Jaeger + Prometheus + Loki + Grafana), and a foundation for an AI-driven root cause analysis system.
Australian core banking sandbox: Spring Boot microservices with Saga/Outbox/CDC, Istio service mesh, full observability stack( Jaeger + Prometheus + Loki + Grafana), and a Claude MCP server for AI-driven root cause analysis.
Goal: A running, end-to-end system where every enterprise pattern (ACID, Saga, Outbox, CDC, distributed tracing, service mesh) is implemented and queryable — proving the patterns work together, not just in theory.
Per-service databases, WAL logical replication for CDC
Kafka
3.9.2 (KRaft)
Event streaming, no ZooKeeper
Redis
7.2
Idempotency keys (24 h TTL)
Debezium Connect
3.1
CDC: polls outbox tables → Kafka topics
OTel Collector
0.149.0
Receives OTLP, fans out to Jaeger/Prometheus/Loki
Jaeger
2.17.0
Distributed trace UI
Prometheus
3.10.0
Metrics
Loki
3.7.0
Log aggregation
Grafana
11.6.14
Unified dashboards
Key Patterns Implemented
ACID Transfer (account-service)
Debit, credit, and outbox row written in a single PostgreSQL transaction. Balance reads use SELECT FOR UPDATE (pessimistic lock) to prevent concurrent overdraft. All monetary values stored as DECIMAL(15,4).
Outbox + CDC (Debezium → Kafka)
Transactional outbox table captures domain events inside the same ACID transaction as the balance update. Debezium reads PostgreSQL WAL and routes events to Kafka topics via the EventRouter SMT — no dual-write, no lost events.
Saga via Kafka
payment-service orchestrates the transfer saga. ledger-service consumes banking.transfer.initiated, posts the double-entry journal, and publishes banking.transfer.confirmed (or routed to DLT on failure). State transitions are event-driven.
payment-service checks a Redis key before forwarding any transfer request. Duplicate requests within 24 hours return the original response without re-executing.
Circuit Breaker
payment-service → account-service calls are wrapped in a Resilience4j circuit breaker. Trips after 5 failures in a 10-call sliding window; half-open probe after 30 s.
Observability
Spring Boot 4 auto-configures OTel. Every service exports traces, metrics, and logs via OTLP to the collector. Grafana datasources (Jaeger, Prometheus, Loki) are pre-provisioned for correlation.
Quick Start
Prerequisites
Podman (rootful, WSL2 backend on Windows) — not Docker
Java 21 + Maven (or use IntelliJ bundled Maven)
kind 0.24+ for Kubernetes phases
1. Start the full stack
podman compose up -d
This starts all four service databases, Kafka, Redis, Debezium Connect, and the full observability stack.
2. Build and run services
# From repo root
"C:/Program Files/JetBrains/IntelliJ IDEA Community Edition 2025.2.4/plugins/maven/lib/maven3/bin/mvn" \
-Dmaven.test.skip=true clean package
podman compose up -d account-service payment-service ledger-service notification-service
3. Register the Debezium connector
bash scripts/register-connector.sh
# or manually:
curl -s -X POST http://localhost:8085/connectors \
-H "Content-Type: application/json" \
-d @infra/debezium/outbox-connector.json
Podman (rootful, WSL2) — no Docker Desktop dependency
B
BankForge
Harry-Zhao-AU·15 ดาว·
Australian core banking sandbox: Spring Boot microservices with Saga/Outbox/CDC, Istio service mesh, full observability stack( Jaeger + Prometheus + Loki + Grafana), and a Claude MCP server for AI-dri
Australian core banking sandbox: Spring Boot microservices with Saga/Outbox/CDC, Istio service mesh, full observability stack( Jaeger + Prometheus + Loki + Grafana), and a Claude MCP server for AI-driven root cause analysis.
ตรวจสอบล่าสุดเมื่อ 2026-10-28 Australian core banking sandbox: Spring Boot microservices with Saga/Outbox/CDC, Istio service mesh, full observability stack( Jaeger + Prometheus + Loki + Grafana), and a Claude MCP server for AI-driven root cause analysis.