guzmonne
commented
3 years ago Thank you for your comment.
This diagram is an architecture I used in most of my projects but now adapted to work with CDK.
I'm using a CDKPipeline to deploy multiple pipelines. Each pipeline builds a single application in a given environment. Most times, two environments are enough: staging and production. Both pipelines are tracking the same GitHub repository and do the same thing. The only difference is that one triggers whenever a new commit is done to the master branch and only when the master branch is tagged with a version number—something like v1.0.0.
I try to decouple stateful stacks from stateless stacks. So, for now, I'm using the CDK Pipeline to configure the data layer on all the environments. I might end up creating additional pipelines if it gets too complex.
Then there's the dev environment. I prefer this environment to work as a "playground" for the developers. It basically holds services needed for the development of the app. These services can be configured and created using multiple methods, and developers can destroy them at any time. Since I'm using AWS Organizations with SSO, I can centrally handle permissions for the developers if I need to limit what they can do (this is usually not the case, I configure a policy that denies the creation of new users, groups, or policies, and that's enough).
Give me a shoutout if you want to take more about this infrastructure or cloud in general.
I'll be updating this repository soon with more documentation and fixes on some of the Constructs and Stacl.
jmjava
Just want to say thank you for pushing this to Github. I am learning quite a bit. Your work is going to help me greatly for some upcoming tasks we are planning!
I see you have a pipeline for each environment? Or is this just a visual representation? We had been thinking about one pipeline per environment instead of moving from stage to stage. We use pipelines to deploy and sometimes bring down part of stack and redeploy just for an environment. With multiple stage this looks more tightly coupled. Also with a single pipeline it is my understanding that the GUI links from the CodePipeline do not work. And you cannot use lookups.
-john