Deploy Aspire applications with Radius

이 콘텐츠는 아직 번역되지 않았습니다.

Radius is a compute environment for aspire publish and aspire deploy. Adding it doesn’t change local execution: aspire run still runs your resources locally.

Prerequisites

A Kubernetes cluster with Radius v0.60.0 or later; v0.60.2 is recommended
The rad CLI on PATH and a workspace configured with rad init for your target cluster
Container images that the cluster can pull

Upgrade older control planes with rad upgrade kubernetes. Before deployment, Aspire checks the control plane targeted by the active Radius workspace, honoring ASPIRE_RADIUS_KUBE_CONTEXT. A detected version below v0.60 fails with ASPIRERADIUS091; an unreadable version isn’t proof of compatibility. Older control planes can silently drop fields from the generated recipe pack.

Add the compute environment

aspire add radius

Aspire CLI는 대화형입니다. 프롬프트 시 알맞은 검색 결과 선택:

Select an integration to add:

> radius (Aspire.Hosting.Radius)
> Other results listed as selectable options...

#:package Aspire.Hosting.Radius@*

<PackageReference Include="Aspire.Hosting.Radius" Version="*" />

TypeScript
C#

import { 
function createBuilder(): IDistributedApplicationBuilder
Creates a new distributed application builder
createBuilder } from './.aspire/modules/aspire.mjs';

const 
const builder: IDistributedApplicationBuilder
builder = await 
function createBuilder(): IDistributedApplicationBuilder
Creates a new distributed application builder
createBuilder();
const 
const radius: RadiusEnvironmentResource
radius = await 
const builder: IDistributedApplicationBuilder
builder.
IDistributedApplicationBuilder.addRadiusEnvironment(name: string): RadiusEnvironmentResource
Adds a Radius compute environment to the application model.
addRadiusEnvironment('radius');
const 
const web: ContainerResource
web = await 
const builder: IDistributedApplicationBuilder
builder.
IDistributedApplicationBuilder.addContainer(name: string, image: string | AddContainerOptions): ContainerResource
Adds a container resource to the application.
addContainer('web', 'nginx');
await 
const web: ContainerResource
web.
ContainerResource.withHttpEndpoint(options?: {
    port?: number;
    targetPort?: number;
    name?: string;
    env?: string;
    isProxied?: boolean;
} | undefined): ContainerResource (+1 overload)
Adds an HTTP endpoint
withHttpEndpoint({ 
targetPort?: number | undefined
targetPort: 80 });
await 
const web: ContainerResource
web.
ContainerResource.withComputeEnvironment(computeEnvironmentResource: IComputeEnvironmentResource): ContainerResource
Configures the compute environment for the compute resource.
withComputeEnvironment(
const radius: RadiusEnvironmentResource
radius);

await 
const builder: IDistributedApplicationBuilder
builder.
IDistributedApplicationBuilder.build(): DistributedApplication
Builds the distributed application
build().
DistributedApplication.run(cancellationToken?: cancellationToken): void
Runs the distributed application
run();

var builder = DistributedApplication.CreateBuilder(args);
var radius = builder.AddRadiusEnvironment("radius");

builder.AddContainer("web", "nginx")
    .WithHttpEndpoint(targetPort: 80)
    .WithComputeEnvironment(radius);

builder.Build().Run();

Generate artifacts for review, then deploy to the configured environment:

aspire publish -o radius-artifacts
aspire deploy

Container workloads are emitted as Radius.Compute/containers. Project resources require a prebuilt, pushed image attached with WithContainerImage / withContainerImage; the Radius publisher doesn’t build that project image for you. Include any files requested by container-file publishing in the externally built image.

Understand recipe-backed connections

Redis, PostgreSQL, MongoDB, SQL Server, and RabbitMQ are backing resources provisioned by Radius recipes, not ordinary container workloads. Recipes determine their deployed addresses. Aspire projects consumer connection strings, individual connection properties, URIs, and service-discovery values from the backing Radius resource, not from the local Aspire endpoint.

Resource	Deployed credential behavior
PostgreSQL	Required username and password properties receive the same parameters used by the consumer connection string
RabbitMQ	An explicit username is required; the password is supplied through a `Radius.Security/secrets` resource ID
MongoDB and SQL Server	Legacy `Applications.*` recipes generate credentials; consumers use recipe outputs, with warnings when explicit AppHost parameters are replaced
Redis	The pinned recipe deploys without authentication; generated passwords are discarded with `ASPIRERADIUS075`, while explicit passwords fail with `ASPIRERADIUS085`

Don’t copy run-mode connection strings into deployment configuration. Host and port values come from the backing resource’s properties, and legacy recipe passwords use listSecrets(). URI credential values are encoded before composition. Radius also injects CONNECTION_<NAME>_<PROPERTY> variables for its connections block; these don’t replace Aspire’s ConnectionStrings__* contract.

Connection strings use portable aliases. Connection-property prefixes retain their own rules: db__primary produces DB__PRIMARY_PASSWORD, while its connection-string alias is ConnectionStrings__db_primary.

Protect credential-bearing values

Aspire publishes credential-bearing environment values through valueFrom.secretKeyRef backed by Radius.Security/secrets, rather than putting them directly in the Kubernetes Deployment specification. A recipe-generated listSecrets() expression avoids writing the resolved password in local publish artifacts, but it can still expose the resolved value in deployment records. Radius’s own connection variables can also contain plaintext credentials. Restrict access to the cluster, deployment records, and secrets.

For workloads that require authenticated Redis, provision a suitable service yourself instead of assuming the pinned unauthenticated recipe applies the local password.

Resolve publish diagnostics

These runtime diagnostics identify unsupported projections or inconsistent output. Fix the application model or callback rather than suppressing them.

Diagnostic	Remediation
`ASPIRERADIUS070`	Review the warning that a recipe-generated credential replaces a referenced parameter
`ASPIRERADIUS072`	Reference only the database supported by the recipe; multiple child databases on one server aren’t independently provisioned
`ASPIRERADIUS073`, `ASPIRERADIUS078`, `ASPIRERADIUS086`	Remove unsupported formatting, deploy-time conditions, or run-only values from published connection expressions
`ASPIRERADIUS074`, `ASPIRERADIUS084`	Keep backing constructs and generated environment secrets that consumers still reference
`ASPIRERADIUS075`, `ASPIRERADIUS085`	Review the unauthenticated Redis recipe; don’t assume a supplied password will be installed
`ASPIRERADIUS076`	Supply the connection properties required by the mapped Radius resource type
`ASPIRERADIUS077`	Reference the primary endpoint instead of a secondary endpoint the recipe doesn’t deploy
`ASPIRERADIUS080`	Review the warning for a named SQL Server child database the legacy recipe doesn’t create
`ASPIRERADIUS081`	Don’t project local TLS scheme, URL, or TLS-enabled values when the backing recipe has no transport-security output
`ASPIRERADIUS082`	Give RabbitMQ an explicit username rather than the loopback-only `guest` account
`ASPIRERADIUS083`, `ASPIRERADIUS087`, `ASPIRERADIUS088`	Use valid Kubernetes names and secret keys, and set either a literal environment value or a complete secret reference
`ASPIRERADIUS089`	Keep the broker’s credential aligned with its consumers when customizing infrastructure
`ASPIRERADIUS090`	Rename colliding Kubernetes secrets, including collisions with generated `<container>-env-secret` names
`ASPIRERADIUS091`	Upgrade the target Radius control plane to v0.60 or later

ASPIRERADIUS071 and ASPIRERADIUS079 indicate an internal resource-to-connection-schema mismatch. Report them with the package version and a minimal reproduction instead of substituting a local endpoint.