Inject dependencies into Cratis Arc query controllers

MVC query controllers use ASP.NET Core dependency injection to receive storage and application services. Put those dependencies in the constructor and keep caller-supplied arguments on the action so the request contract stays separate from the services that perform the read.

Constructor injection

The controller below receives a MongoDB collection and an options object that caps the number of results. It uses the shared domain concepts and an ASP.NET Core Arc host with the MongoDB provider configured. Use this DebitAccount declaration instead of the model-bound alternative:

using System;
using System.Collections.Generic;
using System.Threading.Tasks;
using Microsoft.AspNetCore.Authorization;
using Microsoft.AspNetCore.Mvc;
using Microsoft.Extensions.Options;
using MongoDB.Driver;

namespace Banking.Accounts;

public class AccountQueryOptions
{
    public int MaxResults { get; set; } = 100;
}

public record DebitAccount(AccountId Id, AccountName Name, decimal Balance);

[Authorize(Roles = "AccountReader")]
[Route("api/accounts")]
public class AccountsController(
    IMongoCollection<DebitAccount> collection,
    IOptions<AccountQueryOptions> options) : ControllerBase
{
    [HttpGet("positive")]
    public async Task<IEnumerable<DebitAccount>> PositiveAccounts() =>
        await collection.Find(account => account.Balance > 0)
            .SortBy(account => account.Name)
            .Limit(Math.Clamp(options.Value.MaxResults, 1, 1000))
            .ToListAsync();
}

PositiveAccounts() reads the configured cap, clamps it to the allowed range, and applies Limit() before ToListAsync() executes the query. Keep fluent operations such as Limit() and Skip() before execution: FindAsync() returns a cursor that no longer exposes them. This fixed cap does not provide pages or total-count metadata. Return an IQueryable<T> for renderer-based paging when clients need to navigate the full result set.

Other providers and lifetimes

You can inject a registered EF Core DbContext or application service in the same way. Database reads do not require Chronicle. Resolve scoped services from the request scope instead of capturing them in a singleton. If you cache caller- or tenant-specific results, the cache key and invalidation must preserve those boundaries.

Let storage failures surface as failures so clients can distinguish an unavailable database from a successful query with no matches. A streaming action also needs subscription-owned teardown: injecting the producer’s dependencies does not dispose its upstream reactive subscriptions.

Once the controller has its dependencies, bind caller arguments on the action. Then choose a return type that gives the client the response contract it needs.