Wrong CosmosDB port when running API tests on top of Aspire

[dilyandimitrov]

I have an Aspire setup with CosmosDB and on top of that I run API tests using the DistributedApplicationTestingBuilder.

I have a POST/PUT endpoint that puts some data into CosmosDB, I want to make a test verifying that the data ended up in the DB as it is expected.

Inside the endpoint logic I am using the CosmosClient that the Aspire is putting available into the Services (DI container):

builder.AddAzureCosmosClient(connectionName: "cosmosdb", configureClientOptions: (CosmosClientOptions options) =>
{
    options.SerializerOptions = new CosmosSerializationOptions()
    {
        PropertyNamingPolicy = CosmosPropertyNamingPolicy.CamelCase
    };
}, configureSettings: c => c.Credential = new DefaultAzureCredential());

To verify if my data is in the DB I am creating a new CosmosClient manually inside the test which I am going to use to query the DB.

From the DistributedApplication I can use the await Fixture.App.GetConnectionStringAsync("cosmosdb") to fetch the connection string with the correct PORT where the CosmosDB is running on (AccountKey=C2y6yDjf5/R+ob0N8A7Cgv30VRDJIWEHLM+4QDU5DE2nQ9nDuVTqobD4b8mGGyPMbIZnqyMsEcaGQy67XIw/Jw==;AccountEndpoint=http://localhost:60980/)

But when I try to query the DB I get the following error:

Endpoint not reachable. Failed Location: http://127.0.0.1:8081/;

When I create the CosmosClient inside the test I also get some useful debug logs like:

DocDBTrace Information: 0 : DocumentClient with id 2 initialized at endpoint: [http://localhost:60980/ ](http://localhost:60980/)with ConnectionMode: Gateway, connection Protocol: Https, and consistency level: null
DocDBTrace Information: 0 : Current WriteEndpoints = ([http://127.0.0.1:8081/ ](http://127.0.0.1:8081/)) ReadEndpoints = ([http://127.0.0.1:8081/ ](http://127.0.0.1:8081/))

from which you can see that the Read and Write endpoints are not pointing to the right port (60980 in this case)

The code that I use to create the CosmosClient inside the tests:

var cosmosdb = (await Fixture.App.GetConnectionStringAsync("cosmosdb"))?.Replace(";Database=cosmosdb", "");
var cosmosClient = new CosmosClient(cosmosdb , clientOptions:
    new CosmosClientOptions
    {
        ConnectionMode = ConnectionMode.Gateway,
        SerializerOptions = new CosmosSerializationOptions { PropertyNamingPolicy = CosmosPropertyNamingPolicy.CamelCase }
    });

What am I missing? Why the port for the Read and Write cosmosdb endpoints on my manually created CosmosClient is not pointing to the port I am giving it via the connection string? (The endpoint property inside the cosmos client actually has the right port)

[davidfowl]

I think you're going to need to share a sample so we can see all of the code.

[dilyandimitrov]

So here it is the AppHost program.cs file (the snippet with the cosmosdb):

#pragma warning disable ASPIRECOSMOSDB001
var cosmos = builder
        .AddAzureCosmosDB("cosmos")
        .RunAsPreviewEmulator(e => 
        {
            e.WithDataExplorer();
        });
#pragma warning restore ASPIRECOSMOSDB001

var cosmosdb = cosmos.AddCosmosDatabase("cosmosdb");

builder.AddProject<Projects.XXXXXX>("api")
    .WithReference(cosmosdb)
    .WithEnvironment("CosmosDb:DatabaseName", "db_local");

builder.Build().Run();

For the testing I have this ApiFixture initializing the Aspire AppHost:

public class ApiFixture : IDisposable
{
    public required HttpClient HttpClient { get; set; }
    public required DistributedApplication App { get; set; }
    public required JsonSerializerOptions JsonSerializerOptions { get; set; } 

    public ApiFixture()
    {
        Initialize().GetAwaiter().GetResult();

        JsonSerializerOptions = new(JsonSerializerDefaults.Web);
        JsonSerializerOptions.Converters.Add(new JsonStringEnumConverter());
    }

    public void Dispose()
    {
        HttpClient.Dispose();
        App.Dispose();
    }

    private async Task Initialize()
    {
        var appHost = await DistributedApplicationTestingBuilder.CreateAsync<Projects.XXXXX_AppHost>();
        App = await appHost.BuildAsync();
        await App.StartAsync();
        HttpClient = App.CreateHttpClient("api");
    }
}

And in the test which tests an API Endpoint that creates/updates data into the cosmosdb after the endpoint call I try to initialize a new CosmosClient which should verify if the data is in the CosmosDB:

var cosmosdb = (await Fixture.App.GetConnectionStringAsync("cosmosdb"))?.Replace(";Database=cosmosdb", ""); // probably this replace is not needed
var cosmosClient = new CosmosClient(cosmosdb, clientOptions:
    new CosmosClientOptions
    {
        ConnectionMode = ConnectionMode.Gateway,
        SerializerOptions = new CosmosSerializationOptions { PropertyNamingPolicy = CosmosPropertyNamingPolicy.CamelCase }
    });
var container = cosmosClient.GetContainer("db_local", Entry.ContainerName);

var query = new QueryDefinition("....");
var iterator = container.GetItemQueryIterator<Entry>(query);

using var queryIterator = container.GetItemQueryIterator<Entry>(query);
while (queryIterator.HasMoreResults)
{
        var item = await queryIterator.ReadNextAsync();
}

And on the last line where the query iterator reads the data I get the Endpoint not reachable. Failed Location: http://127.0.0.1:8081/;.

Note: In the code for the API endpoint where I create/update the data in CosmosDB, I use the injected CosmosClient via:

builder.AddAzureCosmosClient(connectionName: "cosmosdb", configureClientOptions: (CosmosClientOptions options) =>
{
    options.SerializerOptions = new CosmosSerializationOptions()
    {
        PropertyNamingPolicy = CosmosPropertyNamingPolicy.CamelCase
    };
}, configureSettings: c => c.Credential = new DefaultAzureCredential());

When this API is deployed to Azure we set this connecting string cosmosdb to the Cosmos Account in Azure (AccountEndpoint=https://xxx.documents.azure.com:443/) and when we run the API locally we rely on the Aspire to set the connecting string with the right port which it does.

The overall issue is that if I try to use the Aspire's connecting string (the port number) to create another CosmosClient something is off and cannot reach the right Read or Write endpoint in CosmosDB spun by the Aspire (in particular, by the DistributedApplicationTestingBuilder)

[davidfowl]

And cosmos is running successfully? Did you wire up the apphost logs to xunit's outputs so you can get better visibility into what is being spun up?

[dilyandimitrov]

Yeah, when I spin up the tests, I can browse the Explorer, see that the data is there (created from the endpoint).

From the logs I see the following when the first CosmosClient is created (the one injected from the DI container in the endpoint code):

DocumentClient with id 1 initialized at endpoint: http://localhost:61252/ with ConnectionMode: Gateway, connection Protocol: Https, and consistency level: null
...
Current WriteEndpoints = (http://localhost:61252/) ReadEndpoints = (http://localhost:61252/)

and when I manually create it I see something like:

DocumentClient with id 2 initialized at endpoint: http://localhost:61252/ with ConnectionMode: Gateway, connection Protocol: Https, and consistency level: null
...
Current WriteEndpoints = (http://127.0.0.1:8081/) ReadEndpoints = (http://127.0.0.1:8081/)

[davidfowl]

Well, not the explorer, I mean the logs from the orchestrator itself that show all of the verbose data about everything being spun up.

[dilyandimitrov]

aspire_cosmos_logs.txt

Not sure if this is what you want, but I am sending you all the logs regarding the spinning up the cosmos in aspire in the tests