Published: 07-Dec-2022

Updated: 07-Dec-2022

Author: Shaun Curtis

Rethinking The Repository Pattern

An article taking a fresh look at how to implement the Repository Pattern.

Introduction

The classic repository pattern is a fairly simple way to implement database access in any application. It meets many of the normal design goals for a small application. On the other side, CQS and CQRS provide a more complex but well structured design pattern for larger more complex applications.

In this article I'll develop the basic repository pattern applying some of the fundimentally good practices used in CQS and implement a fully generic provider.

This is not a regurgitated IRepository implementation in DotNetCore with a few frills.

There's no implementation per entity class. You won't see this:

    public class WeatherForecastRepository : GenericRepository<WeatherForecast>, IWeatherForcastRepository
    {
        public WeatherForecastRepository(DbContextClass dbContext) : base(dbContext) {}
    }

    public interface IProductRepository : IGenericRepository<WeatherForecast> { }

There's no separate UnitOfWork classes: it's built in.
All standard Data I/O uses a single Data Broker.
CQS Requests, Results and Handler patterns are used in the design.

Nomenclature, Terminology and Practices

DI: Dependancy Injection
CQS: Command/Query Separation

The code is:

Net7.0
C# 10
Nullable enabled

Repo

The Repo and latest version of this article are here: Blazr.IRepository.

The Data Store

The solution needs a real data store for testing: it implements an Entity Framework In-Memory database.

I'm a Blazor developer so my test data class is WeatherForecast. The code for the data provider is in the Appendix.

This is the DbContext used by the DBContext factory.

public sealed class InMemoryWeatherDbContext : DbContext
{
    public DbSet<WeatherForecast> WeatherForecast { get; set; } = default!;
    public InMemoryWeatherDbContext(DbContextOptions<InMemoryWeatherDbContext> options) : base(options) { }

    protected override void OnModelCreating(ModelBuilder modelBuilder)
        => modelBuilder.Entity<WeatherForecast>().ToTable("WeatherForecast");
}

Testing the Factory and Context

The following XUnit test demonstrates the basic datastore setup in DI. It:

Sets up a DI container
Loads the data from the Test Provider
Tests the record count is correct
Tests one arbitary record is correct.

[Fact]
public async Task DBContextTest()
{
    // Gets the control test data
    var testProvider = WeatherTestDataProvider.Instance();

    // Build our services container
    var services = new ServiceCollection();

    // Define the DbSet and Server Type for the DbContext Factory
    services.AddDbContextFactory<InMemoryWeatherDbContext>(options
        => options.UseInMemoryDatabase($"WeatherDatabase-{Guid.NewGuid().ToString()}"));

    var rootProvider = services.BuildServiceProvider();

    //define a scoped container
    var providerScope = rootProvider.CreateScope();
    var provider = providerScope.ServiceProvider;

    // get the DbContext factory and add the test data
    var factory = provider.GetService<IDbContextFactory<InMemoryWeatherDbContext>>();
    if (factory is not null)
        WeatherTestDataProvider.Instance().LoadDbContext<InMemoryWeatherDbContext>(factory);

    // Check the data has been loaded
    var dbContext = factory!.CreateDbContext();
    Assert.NotNull(dbContext);

    var count = dbContext.Set<WeatherForecast>().Count();
    Assert.Equal(testProvider.WeatherForecasts.Count(), count);

    // Test an arbitary record
    var testRecord = testProvider.GetRandomRecord()!;
    var record = await dbContext.Set<WeatherForecast>().SingleOrDefaultAsync(item => item.Uid.Equals(testRecord.Uid));
    Assert.Equal(testRecord, record);

    // Dispose of the resources correctly
    providerScope.Dispose();
    rootProvider.Dispose();
}

The Classic Repository Pattern Implementation

Here's a nice succinct implementation that I found on the Internet.

    public abstract class Repository<T> : IRepository<T> where T : class
    {
        protected readonly DbContextClass _dbContext;

        protected GenericRepository(DbContextClass context)
            => _dbContext = context;

        public async Task<T> GetById(int id)
            => await _dbContext.Set<T>().FindAsync(id);

        public async Task<IEnumerable<T>> GetAll()
            => await _dbContext.Set<T>().ToListAsync();

        public async Task Add(T entity)
             => await _dbContext.Set<T>().AddAsync(entity);

        public void Delete(T entity)
            => _dbContext.Set<T>().Remove(entity);

        public void Update(T entity)
           =>  _dbContext.Set<T>().Update(entity);
    }
}

Picking it apart:

What happens when a null is returned, what does it mean?
Did that add/update/delete really succeed? How do I know?
How do you handle cancellation tokens? Most async methods now accept a cancellation token.
What happens when your DBSet contains a million records (maybe the DBA got something wrong last night)?
There's one of me for every data store entity in the application.

The Implementation

Requests and Results

Request objects encapulate what we request and result objects the data and status information we expect back. They are records: defined once and then consumed.

Commands

A Command is a request to make a change to the data store: Create/Update/Delete operations. We can define one like this:

public record CommandRequest<TRecord>
{
    public required TRecord Item { get; init; }
    public CancellationToken Cancellation { get; set; } = new ();
}

Commands only return status information: no data. We can define a result like this:

public record CommandResult
{
    public bool Successful { get; init; }
    public string Message { get; init; } = string.Empty;

    private CommandResult() { }

    public static CommandResult Success(string? message = null)
        => new CommandResult { Successful = true, Message= message ?? string.Empty };

    public static CommandResult Failure(string message)
        => new CommandResult { Message = message};
}

At this point it's worth noting one small exception to the return rule: the Id for an inserted record. If you aren't using Guids to give your records unique identifiers, then the database generated Id is status information.

Item Requests

A Query is a request to get data from the data store: no mutation. We can define an item query like this:

public sealed record ItemQueryRequest
{
    public required Guid Uid { get; init; }
    public CancellationToken Cancellation { get; set; } = new();
}

And the return result: the requested data and status.

public sealed record ItemQueryResult<TRecord>
{
    public TRecord? Item { get; init;} 
    public bool Successful { get; init; }
    public string Message { get; init; } = string.Empty;

    private ItemQueryResult() { }

    public static ItemQueryResult<TRecord> Success(TRecord Item, string? message = null)
        => new ItemQueryResult<TRecord> { Successful=true, Item= Item, Message= message ?? string.Empty };

    public static ItemQueryResult<TRecord> Failure(string message)
        => new ItemQueryResult<TRecord> { Message = message};
}

List Queries

List queries present a few extra challenges.

They should never request everything. In edge conditions there may be 1,000,000+ rows in a table. Every request should be constrained. The request defines StartIndex and PageSize to both constrain the data and provide paging. If you set the page size to 1,000,000, will your data pipeline and front end handle it gracefully?
They need to handle sorting and filtering. The request defines these as Linq Expressions.

public sealed record ListQueryRequest<TRecord>
{
    public int StartIndex { get; init; } = 0;
    public int PageSize { get; init; } = 1000;
    public CancellationToken Cancellation { get; set; } = new ();
    public bool SortDescending { get; } = false;
    public Expression<Func<TRecord, bool>>? FilterExpression { get; init; }
    public Expression<Func<TRecord, object>>? SortExpression { get; init; }
}

The result returns the items, the total item count (for paging) and status information. Items are always returned as IEnumerable.

public sealed record ListQueryResult<TRecord>
{
    public IEnumerable<TRecord> Items { get; init;} = Enumerable.Empty<TRecord>();  
    public bool Successful { get; init; }
    public string Message { get; init; } = string.Empty;
    public long TotalCount { get; init; }

    private ListQueryResult() { }

    public static ListQueryResult<TRecord> Success(IEnumerable<TRecord> Items, long totalCount, string? message = null)
        => new ListQueryResult<TRecord> {Successful=true,  Items= Items, TotalCount = totalCount, Message= message ?? string.Empty };

    public static ListQueryResult<TRecord> Failure(string message)
        => new ListQueryResult<TRecord> { Message = message};
}

Handlers

Handlers are small single purpose classes that handle requests and return results. They abstract the nitty-gritty execution from the higher level Data Broker.

Command Handlers

The interface provides the abstraction.

public interface ICreateRequestHandler
{
    public ValueTask<CommandResult> ExecuteAsync<TRecord>(CommandRequest<TRecord> request)
        where TRecord : class, new();
}

And the implementation does the real work.

Injects the DBContext Factory.
Implements Unit of Work Db contexts through the DbContext factory.
Uses the Add method on the context to add the record to EF.
Calls SaveChangesAsync, passing in the Cancellation token, and expects a single change to be reported.
Provides status information if things go wrong.

public sealed class CreateRequestHandler<TDbContext>
    : ICreateRequestHandler
    where TDbContext : DbContext
{
    private readonly IDbContextFactory<TDbContext> _factory;

    public CreateRequestHandler(IDbContextFactory<TDbContext> factory)
        => _factory = factory;

    public async ValueTask<CommandResult> ExecuteAsync<TRecord>(CommandRequest<TRecord> request)
        where TRecord : class, new()
    {
        if (request == null)
            throw new DataPipelineException($"No CommandRequest defined in {this.GetType().FullName}");

        using var dbContext = _factory.CreateDbContext();

        dbContext.Add<TRecord>(request.Item);
        return await dbContext.SaveChangesAsync(request.Cancellation) == 1
            ? CommandResult.Success("Record Updated")
            : CommandResult.Failure("Error updating Record");
    }
}

The Update and Delete handlers are the same but use different dbContext methods: Update and Remove.

Item Request Handler

The interface.

public interface IItemRequestHandler
{
    public ValueTask<ItemQueryResult<TRecord>> ExecuteAsync<TRecord>(ItemQueryRequest request)
        where TRecord : class, new();
}

And the server implementation. Note:

Injects the DBContext Factory.
Implements Unit of Work Db contexts through the DbContext factory.
Turns off tracking. There's no mutation involved in this transaction.
Checks to see if it can use an Id to get the item - the record implements IGuidIdentity.
If not, tries FindAsync which uses the inbuilt Key methodology to get the record.
Provides status information if things go wrong.

public sealed class ItemRequestHandler<TDbContext>
    : IItemRequestHandler
    where TDbContext : DbContext
{
    private readonly IDbContextFactory<TDbContext> _factory;

    public ItemRequestHandler(IDbContextFactory<TDbContext> factory)
        => _factory = factory;

    public async ValueTask<ItemQueryResult<TRecord>> ExecuteAsync<TRecord>(ItemQueryRequest request)
        where TRecord : class, new()
    {
        if (request == null)
            throw new DataPipelineException($"No ListQueryRequest defined in {this.GetType().FullName}");

        using var dbContext = _factory.CreateDbContext();
        dbContext.ChangeTracker.QueryTrackingBehavior = QueryTrackingBehavior.NoTracking;

        TRecord? record = null;

        // first check if the record implements IGuidIdentity.  If so we can do a cast and then do the query via the Uid property directly 
        if ((new TRecord()) is IGuidIdentity)
            record = await dbContext.Set<TRecord>().SingleOrDefaultAsync(item => ((IGuidIdentity)item).Uid == request.Uid, request.Cancellation);

        // Try and use the EF FindAsync implementation
        if (record is null)
            record = await dbContext.FindAsync<TRecord>(request.Uid);

        if (record is null)
            return ItemQueryResult<TRecord>.Failure("No record retrieved");

        return ItemQueryResult<TRecord>.Success(record);
    }
}

List Request Handler

The interface.

public interface IListRequestHandler
{
    public ValueTask<ListQueryResult<TRecord>> ExecuteAsync<TRecord>(ListQueryRequest<TRecord> request)
        where TRecord : class, new();
}

And implementation.

Note there are two internal methods:

_getItemsAsync Gets the items. This builds an IQueryable object and returns a materialized IEnumerable. You must execute the query before the factory disposes the DbContext.
_getCountAsync Gets the count of all the records based on the filter.

    private async ValueTask<IEnumerable<TRecord>> _getItemsAsync<TRecord>(ListQueryRequest<TRecord> request)
        where TRecord : class, new()
    {
        if (request == null)
            throw new DataPipelineException($"No ListQueryRequest defined in {this.GetType().FullName}");

        using var dbContext = _factory.CreateDbContext();
        dbContext.ChangeTracker.QueryTrackingBehavior = QueryTrackingBehavior.NoTracking;

        IQueryable<TRecord> query = dbContext.Set<TRecord>();
        if (request.FilterExpression is not null)
            query = query
                .Where(request.FilterExpression)
                .AsQueryable();

        if (request.SortExpression is not null)

            query = request.SortDescending
                ? query.OrderByDescending(request.SortExpression)
                : query.OrderBy(request.SortExpression);

        if (request.PageSize > 0)
            query = query
                .Skip(request.StartIndex)
                .Take(request.PageSize);

        return query is IAsyncEnumerable<TRecord>
            ? await query.ToListAsync()
            : query.ToList();
    }

The Repository Class Replacement

First the interface.

The very important bit is the generic TRecord definition on each method, not on the interface. This removes the need for entity specific implementations.

public interface IDataBroker
{
    public ValueTask<ListQueryResult<TRecord>> GetItemsAsync<TRecord>(ListQueryRequest<TRecord> request) where TRecord : class, new();
    public ValueTask<ItemQueryResult<TRecord>> GetItemAsync<TRecord>(ItemQueryRequest request) where TRecord : class, new();
    public ValueTask<CommandResult> UpdateItemAsync<TRecord>(CommandRequest<TRecord> request) where TRecord : class, new();
    public ValueTask<CommandResult> CreateItemAsync<TRecord>(CommandRequest<TRecord> request) where TRecord : class, new();
    public ValueTask<CommandResult> DeleteItemAsync<TRecord>(CommandRequest<TRecord> request) where TRecord : class, new();
}

And the implementation. Each handler is registered in DI and injected into the broker.

public sealed class RepositoryDataBroker : IDataBroker
{
    private readonly IListRequestHandler _listRequestHandler;
    private readonly IItemRequestHandler _itemRequestHandler;
    private readonly IUpdateRequestHandler _updateRequestHandler;
    private readonly ICreateRequestHandler _createRequestHandler;
    private readonly IDeleteRequestHandler _deleteRequestHandler;

    public RepositoryDataBroker(
        IListRequestHandler listRequestHandler,
        IItemRequestHandler itemRequestHandler,
        ICreateRequestHandler createRequestHandler,
        IUpdateRequestHandler updateRequestHandler,
        IDeleteRequestHandler deleteRequestHandler)
    {
        _listRequestHandler = listRequestHandler;
        _itemRequestHandler = itemRequestHandler;
        _createRequestHandler = createRequestHandler;
        _updateRequestHandler = updateRequestHandler;
        _deleteRequestHandler = deleteRequestHandler;
    }

    public ValueTask<ItemQueryResult<TRecord>> GetItemAsync<TRecord>(ItemQueryRequest request) where TRecord : class, new()
        => _itemRequestHandler.ExecuteAsync<TRecord>(request);

    public ValueTask<ListQueryResult<TRecord>> GetItemsAsync<TRecord>(ListQueryRequest<TRecord> request) where TRecord : class, new()
        => _listRequestHandler.ExecuteAsync<TRecord>(request);

    public ValueTask<CommandResult> CreateItemAsync<TRecord>(CommandRequest<TRecord> request) where TRecord : class, new()
        => _createRequestHandler.ExecuteAsync<TRecord>(request);

    public ValueTask<CommandResult> UpdateItemAsync<TRecord>(CommandRequest<TRecord> request) where TRecord : class, new()
        => _updateRequestHandler.ExecuteAsync<TRecord>(request);

    public ValueTask<CommandResult> DeleteItemAsync<TRecord>(CommandRequest<TRecord> request) where TRecord : class, new()
        => _deleteRequestHandler.ExecuteAsync<TRecord>(request);
}

Testing the Data Broker

We can now define a set of tests for the data broker. I've included two here. The rest are in the Repo.

First two methods to create our root DI container and populate the database.

private ServiceProvider BuildRootContainer()
{
    var services = new ServiceCollection();

    // Define the DbSet and Server Type for the DbContext Factory
    services.AddDbContextFactory<InMemoryWeatherDbContext>(options
        => options.UseInMemoryDatabase($"WeatherDatabase-{Guid.NewGuid().ToString()}"));
    // Define the Broker and Handlers
    services.AddScoped<IDataBroker, RepositoryDataBroker>();
    services.AddScoped<IListRequestHandler, ListRequestHandler<InMemoryWeatherDbContext>>();
    services.AddScoped<IItemRequestHandler, ItemRequestHandler<InMemoryWeatherDbContext>>();
    services.AddScoped<IUpdateRequestHandler, UpdateRequestHandler<InMemoryWeatherDbContext>>();
    services.AddScoped<ICreateRequestHandler, CreateRequestHandler<InMemoryWeatherDbContext>>();
    services.AddScoped<IDeleteRequestHandler, DeleteRequestHandler<InMemoryWeatherDbContext>>();

    // Create the container
    return services.BuildServiceProvider();
}

private IDbContextFactory<InMemoryWeatherDbContext> GetPopulatedFactory(IServiceProvider provider)
{
    // get the DbContext factory and add the test data
    var factory = provider.GetService<IDbContextFactory<InMemoryWeatherDbContext>>();
    if (factory is not null)
        WeatherTestDataProvider.Instance().LoadDbContext<InMemoryWeatherDbContext>(factory);

    return factory!;
}

The GetItems test:

[Fact]
public async Task GetItemsTest()
{
    // Get our test provider to use as our control
    var testProvider = WeatherTestDataProvider.Instance();

    // Build the root DI Container
    var rootProvider = this.BuildRootContainer();

    //define a scoped container
    var providerScope = rootProvider.CreateScope();
    var provider = providerScope.ServiceProvider;

    // get the DbContext factory and add the test data
    var factory = this.GetPopulatedFactory(provider);

    // Check we can retrieve thw first 1000 records
    var dbContext = factory!.CreateDbContext();
    Assert.NotNull(dbContext);

    var databroker = provider.GetRequiredService<IDataBroker>();

    var request = new ListQueryRequest<WeatherForecast>();
    var result = await databroker.GetItemsAsync<WeatherForecast>(request);

    Assert.NotNull(result);
    Assert.Equal(testProvider.WeatherForecasts.Count(), result.TotalCount);

    providerScope.Dispose();
    rootProvider.Dispose();
}

The Add Item test:

[Fact]
public async Task AddItemTest()
{
    // Get our test provider to use as our control
    var testProvider = WeatherTestDataProvider.Instance();

    // Build the root DI Container
    var rootProvider = this.BuildRootContainer();

    //define a scoped container
    var providerScope = rootProvider.CreateScope();
    var provider = providerScope.ServiceProvider;

    // get the DbContext factory and add the test data
    var factory = this.GetPopulatedFactory(provider);

    // Check we can retrieve thw first 1000 records
    var dbContext = factory!.CreateDbContext();
    Assert.NotNull(dbContext);

    var databroker = provider.GetRequiredService<IDataBroker>();

    // Create a Test record
    var newRecord = new WeatherForecast { Uid = Guid.NewGuid(), Date = DateOnly.FromDateTime(DateTime.Now), TemperatureC = 50, Summary = "Add Testing" };

    // Add the Record
    {
        var request = new CommandRequest<WeatherForecast>() { Item = newRecord };
        var result = await databroker.CreateItemAsync<WeatherForecast>(request);

        Assert.NotNull(result);
        Assert.True(result.Successful);
    }

    // Get the new record
    {
        var request = new ItemQueryRequest() { Uid = newRecord.Uid };
        var result = await databroker.GetItemAsync<WeatherForecast>(request);

        Assert.Equal(newRecord, result.Item);
    }

    // Check the record count has incremented
    {
        var request = new ListQueryRequest<WeatherForecast>();
        var result = await databroker.GetItemsAsync<WeatherForecast>(request);

        Assert.NotNull(result);
        Assert.Equal(testProvider.WeatherForecasts.Count() + 1, result.TotalCount);
    }

    providerScope.Dispose();
    rootProvider.Dispose();
}

Wrapping Up

What I've presented here is a hybrid Repository Pattern. It maintains the Repository Pattern's simplicity, and adds some of the best CQS Pattern features.

Abstracting the nitty-gritty EF and Linq code to individual handlers keeps the classes small, succinct and single purpose.

The single Data Broker simplifies data pipeline configuration for the Core and Presentation domains.

To those who believe that implementing any database pipeline over EF is an anti-pattern, my answer is: I use EF as just another Object Request Broker [ORB]. You can plug this pipeline into Dapper, LinqToDb, ... . I never build core business logic code (data relationships) into my Data/Infrastructure Domain: [personal view] crazy idea.

Appendix

The Data Store

The test system implements an Entity Framework In-Memory database.

I'm a Blazor developer so naturally my demo data class is WeatherForecast. Here's my data class. Note it is a record for immutability and I've set some arbitary default values for testing purposes.

public sealed record WeatherForecast : IGuidIdentity
{
    [Key] public Guid Uid { get; init; } = Guid.Empty;
    public DateOnly Date { get; init; } = DateOnly.FromDateTime(DateTime.Now);
    public int TemperatureC { get; init; } = 60;
    public string? Summary { get; init; } = "Testing";
}

First a class to generate a data set. This is a Singleton pattern class (not a DI singleton). Methods such as GetRandomRecord are for testing.

public sealed class WeatherTestDataProvider
{
    private int RecordsToGenerate;

    public IEnumerable<WeatherForecast> WeatherForecasts { get; private set; } = Enumerable.Empty<WeatherForecast>();

    private WeatherTestDataProvider()
        => this.Load();

    public void LoadDbContext<TDbContext>(IDbContextFactory<TDbContext> factory) where TDbContext : DbContext
    {
        using var dbContext = factory.CreateDbContext();

        var weatherForcasts = dbContext.Set<WeatherForecast>();

        // Check if we already have a full data set
        // If not clear down any existing data and start again
        if (weatherForcasts.Count() == 0)
        {
            dbContext.AddRange(this.WeatherForecasts);
            dbContext.SaveChanges();
        }
    }

    public void Load(int records = 100)
    {
        RecordsToGenerate = records;

        if (WeatherForecasts.Count() == 0)
            this.LoadForecasts();
    }

    private void LoadForecasts()
    {
        var forecasts = new List<WeatherForecast>();

        for (var index = 0; index < RecordsToGenerate; index++)
        {
            var rec = new WeatherForecast
            {
                Uid = Guid.NewGuid(),
                Summary = Summaries[Random.Shared.Next(Summaries.Length)],
                Date = DateOnly.FromDateTime(DateTime.Now.AddDays(index)),
                TemperatureC = Random.Shared.Next(-20, 55),
            };
            forecasts.Add(rec);
        }

        this.WeatherForecasts = forecasts;
    }

    public WeatherForecast GetForecast()
    {
        return new WeatherForecast
        {
            Uid = Guid.NewGuid(),
            Summary = Summaries[Random.Shared.Next(Summaries.Length)],
            Date = DateOnly.FromDateTime(DateTime.Now.AddDays(-1)),
            TemperatureC = Random.Shared.Next(-20, 55),
        };
    }

    public WeatherForecast? GetRandomRecord()
    {
        var record = new WeatherForecast();
        if (this.WeatherForecasts.Count() > 0)
        {
            var ran = new Random().Next(0, WeatherForecasts.Count());
            return this.WeatherForecasts.Skip(ran).FirstOrDefault();
        }
        return null;
    }

    private static WeatherTestDataProvider? _weatherTestData;

    public static WeatherTestDataProvider Instance()
    {
        if (_weatherTestData is null)
            _weatherTestData = new WeatherTestDataProvider();

        return _weatherTestData;
    }

    public static readonly string[] Summaries = new[]
    {
        "Freezing", "Bracing", "Chilly", "Cool", "Mild", "Warm", "Balmy", "Hot", "Sweltering", "Scorching"
    };

}

The DbContext.

public sealed class InMemoryWeatherDbContext
    : DbContext
{
    public DbSet<WeatherForecast> WeatherForecast { get; set; } = default!;
    public InMemoryWeatherDbContext(DbContextOptions<InMemoryWeatherDbContext> options) : base(options) { }

    protected override void OnModelCreating(ModelBuilder modelBuilder)
        => modelBuilder.Entity<WeatherForecast>().ToTable("WeatherForecast");
}

Table of Contents

Rethinking The Repository Pattern

Introduction

Nomenclature, Terminology and Practices

Repo

The Data Store

Testing the Factory and Context

The Classic Repository Pattern Implementation

The Implementation

Requests and Results

Commands

Item Requests

List Queries

Handlers

Command Handlers

Item Request Handler

List Request Handler

The Repository Class Replacement

Testing the Data Broker

Wrapping Up

Appendix

The Data Store