kaos doesn't rollback correctly from failed cloud deployments #80
Description
What is the current behaviour?
When a kaos build deploy command fails, it results in a .kaos/<CLOUD>/None file, which then would make kaos think that a backend was already build for that cloud. When the user tries to truly build the backend for that same cloud, the following gets reported back:
➜ kaos build deploy -c GCP -vy
Aborting - kaos backend is already built.
Even though the .kaos directory doesn't contain a valid backend and the list of backends reports only one:
➜ kaos build list
+-------+---------+----------+-------+
| index | context | provider | env |
+-------+---------+----------+-------+
| 0 | DOCKER | DOCKER | local |
+-------+---------+----------+-------+
➜ tree .kaos
.kaos
├── config
└── state
├── AWS
│ └── None
├── DOCKER
│ ├── config.json
│ └── terraform.tfstate
└── GCP
└── None
6 directories, 3 files
Trying to destroy the GCP backend would result in the following error (valid scenario):
➜ kaos destroy -c GCP -vy
Warning - kaos [prod] backend in GCP has not been deployed!
What is the expected behaviour?
When kaos fails to build a backend, the None file should be removed, the .kaos/<CLOUD> should be correctly removed.
How to reproduce? (e.g. logs, minimal example, etc...)
- Move your GCP credentials file somewhere else or just empty the
GOOGLE_APPLICATION_CREDENTIALSenvironment variable - Try to build kaos backend on GCP:
kaos build deploy -c GCP -vy - Make sure that the failure led to the creation of
.kaos/GCP/Nonefile - Re-run GCP build command:
kaos build deploy -c GCP -vy - Get the
Aborting - kaos backend is already built.error back
Context (Environment)
- kaos Cloud Provider: LOCAL, AWS and GCP
- Operating System: mac OS
Additional Details
- The
kaos build deploycommand could search for theterraform.tfstateinside the cloud directory rather than asserting that the directory exists