-1

Creating reusable, containerised microservices is great for EDA but networking and data sharing can be costly compared to a monolithic app. A great solution would be to run a workflow on a single pod composed of these microservice containers, injecting workflow specific configuration for each one. The data can be shared via local emptyDir on RAM or storage.

Argo workflows achieve something similar, but it would be great see it in Knative too.

I would expect faster workflows and more efficient use of k8s resources

Martin Golding
  • 1

1 Answers1

0

I'm at assuming you're referring to this story which has been hot for the last week or so.

It would be possible to build a system like the one described by the Prime Video team using multi-container support and a controller container much like the one described by the Prime Video team. It's even possible to build a multi-container workflow orchestrator like this so that it works with both Knative Serving and standard Kubernetes Deployments (including ones managed by KEDA), which really suggests it would make sense as an independent project.

It's also possible to use Argo Workflows or Tekton for this; you don't need Knative to do everything. One of the goals of the project is to provide a tool that works well with other Kubernetes tools, rather than replacing all other tools.

E. Anderson
  • 3,405
  • 1
  • 16
  • 19
  • Yes, that Prime Video post prompted my interest. Do you mean to use Argo / Tekton instead of Knative for multi-container workflows? I like the idea of using Kogito to separate the business logic from the rest of the app, and use camel for integrations. I think they are designed to work with Knative workflows so there would be a benefit to a multi-container pod in that scenario. – Martin Golding May 11 '23 at 10:38
  • My point was that Argo and Tekton implement ordered execution of containers with shared storage, which would be similar to the original Prime Video architecture. I suspect using them at the described scale would run into the same scaling issues that Step Functions had. – E. Anderson May 12 '23 at 13:05