Radius sandbox submission review

[joshgav]

trafficstars

Radius is a framework for building cloud-native app platforms proposed to CNCF sandbox in https://github.com/cncf/sandbox/issues/65. This issue tracks discussions and reviews of Radius to help with accepting it in CNCF sandbox.

We've asked @jonvsm and team to join an upcoming TAG meeting to discuss the following about the project. Please plan to present for about 20 minutes, followed by 5-10 minutes of questions and answers from TAG members.

• the values this project proposes to end users
• the project's high-level technical architecture
• a brief demo of experiences and UIs
• the project's near-term roadmap
• state of the project's community and governance
• comparison with existing projects

[joshgav]

Thank you to the team for presenting Radius to us at Kubecon in Chicago.

The recording starts here: https://youtu.be/NZCmYRVziGY?t=3996 The presentation is here: https://drive.google.com/file/d/1bhrVWwwT9Si-1NVFQhr3gtqMyQF4zrVR/view Project repo is here: https://github.com/radius-project/radius

Here's a summary of what was presented.

Project

• part of Azure Incubations, charged with contributing to open source community, contributed KEDA, Dapr and others in the past
• goal is to show potential, get community feedback
• currently developed by Azure team only, seeking more contributors now

Problems to solve

• Kubernetes is a standard across clouds, but more than K8s is needed for cloud-native applications, such as managed services in clouds, or capabilities provided by platform teams
• Cloud-native applications involve complex architectures and are difficult to understand and troubleshoot
• Requires cross-platform deployments - multiple clouds and datacenters
• Need to enforce good practices in a central way
• today's apps are viewed as a flat collection of pieces, no relationships

Values project offers

• enable collaboration between app developers and infrastructure operators
• infrastructure recipes - descriptions of what needs to be done following best practices
• application graph - show interconnections between services and components
• cloud neutral - a toolset that works across any cloud or datacenter

Architecture of Recipes

• a recipe defines an interface for a collection of dependencies, developed by a platform team
• several implementations of the same recipe can be provided, appropriate for different service providers and datacenters
• run rad init CLI to configure Radius on a K8s cluster
• add a recipe to an existing application when it's time to deploy
• bind recipe service to workload with custom annotation - consistent env vars are added to application
  • opportunity to create an industry standard for writing to specific env vars etc. for secrets - if there's momentum around Open Service Broker could use that
  • some services are relatively standard, like Redis, but not so easy for custom services
  • recipes have a way to find and publish connection details to app
• recipe controller creates implementation and publishes bindings
• recipes can discover some parameters from context - recipe authors can use them via variables - allows automation in different environments
• recipes declare their output
• recipes can include components described in bicep, Kubernetes YAML, Terraform modules, etc. Top level of recipe is bicep.
• once deployed can display full app graph with recipes connected to applications, including actual implementation of recipe, which differs based on environment (e.g., AWS or Azure)

Questions

• Is this related to OAM? A: Yes, came from same team, but OAM was just a spec.
• Can we learn from Radius and apply to OAM?
  • "It should feel magical" that components connect together. We'd like to do it automatically.
  • Specs only go so far - till an implementation is made hard to determine how it should work.
• Will bicep be required? A: That's an area to work together, don't intend for bicep to be required.
• Are you built on Kubernetes? A: intend to be able to target all kinds of runtime services, including things like Azure Web Apps or AWS Lambda.
• How much bicep do you have to use? Operators (platform engineers) will need to write bicep at least for now; developers can simply use "recipe" Kubernetes resources.
• Can you use Kubernetes YAML directly? You can embed K8s resources in a recipe, but Kubernetes YAML/APIs aren't the top-level API. Don't believe K8s is an ideal base for framework.

[abangser]

Hi @dims, can you expand a bit more on what you are asking for here? Are you asking for more details from Radius submission team or more details from the TAG leadership? Thanks!

[dims]

Whooops! wrong repo. sorry :)

deleted a couple of comments earlier to make sure i don't confuse folks other than myself!

[abangser]

No worries @dims! That is why I wanted to ask 😄

I think if you ask https://github.com/cncf/sandbox/issues/65 with specific questions you may get the best response possible 🤞

[thschue]

Presentation is done and currently no open action items. Therefore closing this ticket.