Personal Banking-Capacity

Personal Banking System - Scalable Microservices Architecture on Azure (AKS + Kafka)

🌐 Overview

• Use Case: Complex personal banking system

• Microservices: 20 Spring Boot-based microservices

• Cloud: Azure (Active-Active region setup)

• Traffic:

  • Average: 500 TPS

  • Peak: 1000+ TPS

  • Concurrent Users: 10,000 - 15,000

• Core Modules:

  • Customer Onboarding

  • Account Management

  • Fund Transfer (NEFT/RTGS/IMPS/UPI)

  • Loan Management (Personal/Home/Car)

  • Fraud Detection

  • Statements & Notifications

🏋️ Capability Map

Business Domain	Capability
Customer Management	Onboarding, KYC, Profile Update
Accounts	Creation, Closure, Linking
Transactions	NEFT, UPI, RTGS, IMPS
Loan Services	Eligibility, EMI Management
Fraud Detection	Real-time transaction analysis
Notifications	SMS, Email, In-app messages
Statements	Monthly/Annual Statements
Audit & Compliance	Logging, Traceability, Reporting

⚖️ Capability to Service Mapping

Capability	Microservice
Onboarding	customer-onboarding-service
KYC	kyc-service
Profile Management	profile-service
Account Management	account-service
Fund Transfer	fund-transfer-service
Payment Gateway	payment-gateway-service
UPI/NEFT/RTGS	payment-routing-service
Loan Management	loan-service
Loan Evaluation	loan-evaluation-service
Fraud Detection	fraud-analytics-service
Notifications	notification-service
Statement Gen	statement-service
Auth/Login	auth-service
Audit Logging	audit-service
User Session Mgmt	session-service
Reporting	report-service
Document Storage	doc-storage-service
Config Management	config-service
API Gateway	api-gateway
Orchestration	orchestration-service

📊 Resource Planning (Active-Active AKS + Kafka)

🌟 Performance Targets

• Peak TPS: 1000+

• Latency Target: <200ms for 95% of requests

• Availability: 99.99% (across regions)

🚀 AKS Cluster Planning

• Cluster Count: 2 AKS clusters (Active-Active)

• Node Pool Size (per cluster):

  • Node Type: D8s_v5 (8 vCPU, 32 GB RAM)

  • Node Count: 20 (scalable to 30)

• Total CPU/RAM (per cluster): 160 vCPU / 640 GB RAM

🎡 Kafka Planning (Azure Event Hubs / Confluent on AKS)

• Broker Nodes: 6 brokers x 2 regions

• Partitions: 200 (across topics)

• Replication Factor: 3 (high durability)

• Throughput: 2 GBps (handle 1000 TPS)

🌀 Other Resources

• Azure SQL/PostgreSQL (Active Geo-Replication)

• Blob Storage (RA-GRS)

• Redis Cache (Active-Active)

• Application Gateway + Front Door (Global Load Balancer)

🔹 Architecture Flow (Text Description)

1. User Request reaches Azure Front Door

2. Routed to region via Traffic Manager

3. Passes through Azure API Gateway

4. Authentication via auth-service

5. Based on request type:

   • Onboarding -> customer-onboarding-service

   • UPI/NEFT -> fund-transfer-service

   • Loan check -> loan-service + loan-evaluation-service

6. Transactions logged via audit-service

7. Data written to PostgreSQL + Kafka

8. Kafka triggers fraud-analytics-service

9. Notifications sent via notification-service

10. User gets async updates via websockets or push notifications

🚀 Pods & Replicas

🧶 Pod Planning

• Total Microservices: 20

• Average Resource per Pod: 500m CPU / 1 GB RAM

Service Type	Replicas (Avg)	Reason
Core Services	4-5	Consistent usage (auth/account)
High-Traffic (e.g. payments)	8-10	500+ TPS
Async Services	2-3	Notification, audit
Event Consumers	3-5	Kafka-based

Total Pods Estimate: ~100-120 per region (scalable)

⚡ Node Pool & Scaling Strategy

• HPA (Horizontal Pod Autoscaler): CPU/memory/custom metrics

• Node Autoscaler: Enabled for burst traffic

• Node Pool Design:

  • General Pool (80%) for business logic

  • Dedicated Pool (20%) for Kafka & infra services

⚠️ Enterprise Risks & Mitigation

Category	Risk	Priority	Mitigation
Business	User churn due to downtime	High	Active-Active, SLAs, SRE practices
Operations	Scaling delays during traffic burst	High	Autoscaler with buffer nodes
Technology	Kafka message loss	High	Durable storage + replication
People	Misuse of elevated roles	Medium	RBAC, Zero Trust policies
Process	Inefficient loan processing	Medium	BPM, async workflows
Security	Transaction spoofing/fraud	High	Real-time fraud analytics, OTP, MFA
Compliance	Non-adherence to RBI/SEBI	High	Audits, traceability, retention
Governance	No visibility on microservice health	Medium	Observability stack (ELK, Prometheus)