Multi-Database Architecture — Complete Guide

Introduction

Multi-Database Architecture — Complete Guide is essential for .NET developers building ShopNest.DataAccess — Enterprise High-Performance Data Platform — Toolliyo's 100-article ADO.NET Core master path covering SqlConnection, stored procedures, transactions, connection pooling, ASP.NET Core integration, Azure SQL, and ten enterprise portfolio projects. Every article includes minimum two enterprise real-world examples (ICICI banking, TCS ERP reporting, insurance batch, legacy modernization).

In Indian delivery projects (TCS, Infosys, Wipro), interviewers expect multi-database architecture with real banking transfers, ERP GL reports, or legacy stored procedure modernization — not toy animal demos. This article delivers production depth on Dashboard Reporting.

After this article you will

• Explain Multi-Database Architecture in plain English and in SQL Server / ADO.NET terms
• Implement multi-database architecture in ShopNest.DataAccess — Enterprise High-Performance Data Platform (Dashboard Reporting)
• Compare SQL-concat / sync anti-patterns vs production-ready parameterized async ADO.NET
• Answer fresher, mid-level, and senior ADO.NET and SQL Server interview questions confidently
• Connect this lesson to Article 65 and the 100-article ADO.NET Core roadmap

Prerequisites

• Software: .NET 8 SDK, SQL Server Express / LocalDB or Docker, SSMS or Azure Data Studio
• Knowledge: C# basics and SQL Server
• Previous: Article 63 — Data Warehousing Concepts — Complete Guide
• Time: 28 min reading + 30–45 min hands-on in SSMS

Concept deep-dive

Level 1 — Analogy

ADO.NET is a direct phone line to SQL Server — your C# speaks SQL through SqlConnection without an ORM translator in the middle.

Level 2 — Technical

Multi-Database Architecture establishes ShopNest.DataAccess foundations — SqlClient stack, connection strings, and when raw ADO.NET beats EF Core for Dashboard Reporting.

Level 3 — Data platform view

[ASP.NET Core API / MVC Controller]
       ▼
[Application Service — IOrderRepository interface]
       ▼
[ADO.NET Repository — SqlConnection + SqlCommand + SqlParameter]
       ▼
[SQL Server — Tables · Indexes · Stored Procedures · Transactions]
       ▼
[Connection Pool · Read Replica · Azure SQL · Serilog + SQL Profiler]

Project structure

ShopNest.DataAccess/
├── src/
│   ├── ShopNest.DataAccess.Api/       ← ASP.NET Core Web API
│   ├── ShopNest.DataAccess.Core/      ← Repository interfaces + DTOs
│   ├── ShopNest.DataAccess.AdoNet/    ← SqlConnection, SPs, transactions
│   ├── ShopNest.DataAccess.Reports/   ← Streaming readers, GL reports
│   └── ShopNest.DataAccess.Tests/     ← Testcontainers SQL integration
├── sql/
│   ├── migrations/
│   └── stored-procedures/             ← usp_Orders_*, usp_Payments_*
└── docker-compose.yml                 ← SQL Server 2022

Hands-on implementation — Dashboard Reporting

Write Multi-Database Architecture in ShopNest.DataAccess for Dashboard Reporting: SqlConnection/SqlCommand with parameters, async calls, and verify in SSMS with execution plan.

1. Open ShopNest.DataAccess repository for this lesson module.
2. Use SqlConnection with await using and connection string from IConfiguration.
3. Add SqlParameter for every user input — never string concatenation.
4. Use ExecuteReaderAsync for reads; transactions for multi-statement writes.
5. Verify in SSMS — check execution plan, row counts, and connection pool metrics.

Anti-pattern (SQL concat, sync calls, DataSet for huge reports)

// ❌ BAD — SQL concat, sync call, no disposal
public List<Order> GetOrders(string status) {
    var conn = new SqlConnection(_connStr);
    conn.Open();
    var cmd = new SqlCommand("SELECT * FROM Orders WHERE Status = '" + status + "'", conn);
    var reader = cmd.ExecuteReader(); // sync, blocks thread pool
    // connection never disposed — pool exhaustion under load
    return Parse(reader);
}

Production-style ADO.NET data access

// ✅ CORRECT — Multi-Database Architecture on ShopNest (Dashboard Reporting)
public async Task<IReadOnlyList<OrderDto>> GetByStatusAsync(string status, CancellationToken ct) {
    await using var conn = new SqlConnection(_config.GetConnectionString("ShopNestDb"));
    await conn.OpenAsync(ct);
    await using var cmd = new SqlCommand("usp_Orders_GetByStatus", conn) {
        CommandType = CommandType.StoredProcedure
    };
    cmd.Parameters.Add("@Status", SqlDbType.NVarChar, 20).Value = status;
    var list = new List<OrderDto>();
    await using var reader = await cmd.ExecuteReaderAsync(ct);
    while (await reader.ReadAsync(ct))
        list.Add(new OrderDto(reader.GetInt32(0), reader.GetDecimal(1), reader.GetString(2)));
    return list;
}

Complete example

// Multi-Database Architecture — ShopNest.DataAccess (Dashboard Reporting)
await using var conn = new SqlConnection(_connectionString);
await conn.OpenAsync(ct);

ADO.NET vs EF Core vs Dapper — when to use which

Rule of thumb: EF Core for productivity; ADO.NET for performance-critical SQL, stored procedures, bulk ops, and legacy SP bridges.

SQL performance and connection management — Multi-Database Architecture

• Connection pooling — default enabled; never disable without load testing; watch pool exhaustion (error 10053/10054)
• Parameterized queries — always use SqlParameter; prevents SQL injection and enables plan cache reuse
• Async — ExecuteReaderAsync/ExecuteNonQueryAsync free thread pool under load
• CommandBehavior.SequentialAccess — stream large BLOB/text columns without loading full row into memory
• Indexes — align with WHERE/JOIN columns; use SQL Server DMVs to find missing indexes

Real-World Example 1 — Insurance Claims Batch Processing

MANDATORY enterprise scenario (Insurance (LIC-style)): Multi-Database Architecture in ShopNest.DataAccess Dashboard Reporting.

Business problem

Nightly batch validates 500K claims against policy rules. SqlBulkCopy inserts staging rows; stored procedure usp_ValidateClaims runs set-based SQL; invalid claims roll back in one transaction per batch of 5,000.

Architecture

CSV ingest → SqlBulkCopy → Staging.Claims
  → BEGIN TRANSACTION per batch
  → EXEC usp_ValidateClaims @BatchId
  → COMMIT or ROLLBACK + ErrorLog

Production ADO.NET code

using var bulk = new SqlBulkCopy(conn, SqlBulkCopyOptions.TableLock, tx)
{
    DestinationTableName = "Staging.Claims",
    BatchSize = 5000,
    BulkCopyTimeout = 600
};
await bulk.WriteToServerAsync(dataTable, ct);
await using var validate = new SqlCommand("usp_ValidateClaims", conn, tx) { CommandType = CommandType.StoredProcedure };
validate.Parameters.Add("@BatchId", SqlDbType.Int).Value = batchId;
await validate.ExecuteNonQueryAsync(ct);

Outcome

Batch window reduced from 6 hours to 90 minutes; zero partial batches after explicit transaction boundaries.

Real-World Example 2 — Hospital Patient Records (Apollo-Style)

MANDATORY enterprise scenario (Healthcare): Multi-Database Architecture in ShopNest.DataAccess Dashboard Reporting.

Business problem

Patient vitals, lab results, and billing hit different schemas. ADO.NET multi-result sets from usp_GetPatientDashboard return 3 grids in one round trip — faster than 3 EF Core queries with Include chains.

Architecture

usp_GetPatientDashboard @PatientId → Result set 1: Demographics
  → Result set 2: Latest vitals → Result set 3: Pending bills

Production ADO.NET code

await using var reader = await cmd.ExecuteReaderAsync(ct);
var demo = ReadDemographics(reader);
await reader.NextResultAsync(ct);
var vitals = ReadVitals(reader);
await reader.NextResultAsync(ct);
var bills = ReadBills(reader);

Outcome

Dashboard load 320ms vs 890ms EF Core; HIPAA audit trail via ADO.NET audit interceptor.

ADO.NET with ASP.NET Core — Multi-Database Architecture

Register IDashboardReportingRepository as Scoped in DI. Never hold SqlConnection across requests. Use IConfiguration for connection strings; User Secrets locally, Azure Key Vault in production.

builder.Services.AddScoped<IOrderRepository, OrderRepository>();
builder.Services.AddHealthChecks().AddSqlServer(connectionString);

Stored procedures and SQL safety

Enterprise ShopNest modules use versioned stored procedures (usp_ prefix). Never concatenate user input — always SqlParameter. Log slow queries (>500ms) with Serilog and review execution plans in SSMS.

Common errors & fixes

• SQL built with string concatenation from user input — Use SqlParameter with typed values for every dynamic value.
• Not disposing SqlConnection / SqlDataReader — Use await using for connection, command, and reader — return connections to pool.
• Loading million-row reports into DataTable — Stream with SqlDataReader and yield batches; avoid DataSet for large data.
• Hard-coding connection strings in repository classes — IConfiguration + User Secrets locally; Azure Key Vault in production.

Best practices

• 🟢 SqlParameter for every dynamic value — zero string concatenation
• 🟢 await using for SqlConnection, SqlCommand, SqlDataReader — return to pool
• 🟡 Async ADO.NET end-to-end on ASP.NET Core request paths
• 🟡 Stream large reports with SqlDataReader; avoid DataSet for millions of rows
• 🔴 SqlTransaction for multi-statement financial writes with explicit rollback
• 🔴 Connection strings in Key Vault — never committed to Git

Interview questions

Fresher level

Q1: What is Multi-Database Architecture in ADO.NET Core?
A: Multi-Database Architecture on ShopNest Dashboard Reporting: SqlConnection lifecycle, SqlCommand with parameters, async execution, and disposal for connection pool health.

Q2: ADO.NET vs EF Core — when to use which?
A: ADO.NET for stored procedures, bulk load, streaming reports, and legacy SQL; EF Core for rapid CRUD and migrations. ShopNest uses both.

Q3: How do you prevent SQL injection in ADO.NET?
A: Always SqlParameter with typed SqlDbType — never string concatenation, even for "trusted" internal tools.

Mid / senior level

Q4: Explain connection pooling and why disposal matters.
A: SqlConnection.Close/Dispose returns the physical connection to the pool. Leaked connections exhaust Max Pool Size and cause timeouts.

Q5: How do you handle transactions in ADO.NET?
A: SqlTransaction with try/commit/catch/rollback; choose isolation level (ReadCommitted default); retry deadlocks with Polly.

Q6: How would you optimize a slow stored procedure report?
A: Check execution plan in SSMS, add covering indexes, avoid SELECT *, stream with SqlDataReader, consider read replica for analytics.

Coding round

Implement a parameterized ADO.NET repository method for ShopNest Dashboard Reporting — show SqlConnection, SqlCommand, SqlParameter, async disposal, and one xUnit integration test.

Summary & next steps

• Article 64: Multi-Database Architecture — Complete Guide
• Module: Module 7: Advanced Enterprise Topics · Level: ADVANCED
• Applied to ShopNest.DataAccess — Dashboard Reporting

Previous: Data Warehousing Concepts — Complete Guide
Next: Read Replica Strategies — Complete Guide

Practice: Run today's SQL in SSMS with execution plan — commit with feat(adonet): article-064.

FAQ