Hi team, This week I participated in a Java conference ( https://2022autumn.javacro.hr/eng) and I demonstrated how Apache Brooklyn can be used to develop applications for the cloud locally, that get deployed to little or no change to a cloud provider. I used the brooklyn-terraform module to deploy a three tier application to a local Kubernetes cluster and then migrate it to EKS.
Here are a few things a few things that we must/should do that I figured out while preparing the demo and presenting at #javacro2022: 1. When we have a frontend (or anything really) depending on a cluster that exposes some sensors the frontend needs, deleting the app causes the dependent component to hang, because the cluster is deleted first. Not sure if this is caused by the fact that the cluster is a software component, or maybe we have a race condition between deleting tasks. I can reproduce it easily, so this is a bug that we can and we should fix. 2. We need to add custom terraform types: dbs, servers, for most used cloud providers etc. 3. All terraform entities under the same parent - when deployed on Kubernetes - should be deployed under the same namespace - it helps with cleaning. Also.. maybe in the case of an application that had only terraform entities deployed on the same Kubernetes cluster as children, a successful deletion of the namespace should suffice and Apache Brooklyn should back off instead of trying to delete the entities itself and failing to do so. We could use the power of Kubernetes when possible. 4. At this conference there were a lot of young people just dipping their toes in cloud development and they are interested in Apache Brooklyn for its ability to help them get used to working on the cloud without an actual cloud. Maybe a collaboration with Universities for workshops should be considered. Also we could use some fresh eyes on Brooklyn to get some feedback, use them as a testing resource and a source of ideas for things missing in Apache Brooklyn. 5. When terraform deploys over an unstable network, the application deploy fails in Brooklyn. Is there anything we can do about this? Maybe a retry? 6. One observation from a fellow speaker is that in order to get things done, we need to write a lot of YAML. I don't particularly disagree, nor have any idea how to fix it, unless we consider trying to enrich our catalog with out of the box types for most recent technologies. That is pretty much it, if anything else comes to mind, I'll make another list and continue this email thread. Observation: You might ask why I used Apache Brooklyn to drive a Terraform deployment on a Kubernetes cluster? Because it provided me the opportunity to try to create a CAMP blueprint agnostic towards where it got deployed. -- Best regards, Iuliana Cosmina <http://rpx.kicks-ass.net>
