search

opendatahub-io / ai-edge

ODH integration with AI at the Edge usecases
Apache License 2.0
8 stars 17 forks source link

Investigate the use of the Multicloud Validated Pattern to deploy the Infrastructure dependencies for our use cases #98

Open LaVLaS opened 11 months ago

LaVLaS commented 11 months ago

Currently we are relying on providing the setup instructions for deploying the required infrastructure for our Edge PoC using ACM, GitEa, OpenShift Pipelines. Since all of those are stan

https://validatedpatterns.io/patterns/multicloud-gitops/

piotrpdev commented 11 months ago

Shouldn't we use the "Industrial Edge" pattern instead? That looks much more similar to what we're currently doing.

image

LaVLaS commented 10 months ago

Shouldn't we use the "Industrial Edge" pattern instead? I created this story when we were moving further into owning ACM infrastructure rollout as part of the Proof of Concept demo when this could responsibilty be offloaded to existing projects that already provide that functionality.

The Industrial Edge pattern is a combination of Multicloud Pattern + MLOps Pipelines + Industrial Edge intelligent application. Once these AI Edge features are integrated into ODH, we can work with the validated pattern team to update the Industrial Edge pattern to use our ODH Edge features for Jupter Notebooks, MLOps pipelines and Model Observability to deploy and monitor their anomaly detection application.

As a long term goal, we could provide a Multi Cloud AI Edge pattern that can deploy an ACM Hub + ODH to support someone to have a "one-click" install of a AI/ML development environment for developing and delivering models to the edge

adelton commented 10 months ago

I took a look at the https://validatedpatterns.io/patterns/multicloud-gitops/ and tried to follow the documentation and set it up. I found out that the documentation is lacking in many places (I filed 15+ issues in https://github.com/validatedpatterns/docs/issues for those findings) and in the end I got blocked by https://github.com/validatedpatterns/docs/issues/351 -- no applications got deployed on the managed cluster.

I tried to debug the issue but gave up after two hours.

So the problem that I see with this validated pattern is -- while it is a nice theoretical concept based on a reasonable architecture, the promise of "With a pattern, you can automatically deploy a full application stack through a GitOps-based framework" listed at https://validatedpatterns.io/ does not seem to be currently realized. The actual artifacts provided (documentation and the YAMLs) are not a solid, stable baseline for building further solutions. The steps that we currently have in this repo in the README.mds seem to be reasonably easy to follow and copy-n-paste without being dragged into a fairly convoluted structure of the tooling used by the pattern.

Now I'm sure those issues can and will be fixed. But I worry that using it might be a long-term maintenance liability which would not be outweighted by the benefits of being "pattern-compatible". :-)

LaVLaS commented 10 months ago

Now I'm sure those issues can and will be fixed. But I worry that using it might be a long-term maintenance liability which would not be outweighted by the benefits of being "pattern-compatible". :-)

Ok. The priority of this was always low at best since our expectations would be that a hub cluster would have ACM + OpenShift Pipelines available prior to the installation of OpenShift AI. We'll have to see if the Validated Patterns team responds to the issue and how quickly, or if, they respond

beekhof commented 9 months ago

I'd suggest the sepsis pattern actually, much more focused. Industrial Edge is a bit of a beast