rivernews
commented
3 years ago Our approach:
- [x] Digitalocean may delay detach cluster after deleting cluster
- [x] Add random string to cluster name #36
- Even better: add random string to var project-name instead.
- Better modulization - move code-server, Postgres, Redis, ... etc all these micro service to other repo as its own tf module.
- This will avoid failure of any micro service update causes tfstate corruption that leads to entire cluster nuking.
Somethings to deal with
- [x] How to let other tf module (microservice.tf in iriversland-api and appl-tracky-api repo) know cluster name?
- Option 1: a 3rd platform storing such value #40
- Option 1: tf data query cluster(s) on DO. Perhaps having a common tag attached to DO cluster(s), but might need to resolve ambiguity.
Terraform kubernetes provider hasn't been happy for the version we are using.
Deployment already exist.
Cluster already exist.
Not to say failed terraform apply leaving unrecoverable resources, have to manually delete all and start over.
Possibly using "data" terraform type to deal with this.