If deployment happens without updating source control, chaclie can get stuck.

When moving from chalice-manged roles to manually, for example, chalice deletes the role and updates the deployment metadata.

When this deployment happens in an environment that does not update the stored (in github, outside a docker container, etc) `.chalice/deployed/.json` file, the change succeeds in AWS, but isn't reflected in the stored metadata in the chalice directory.

Then, the metadata in the file doesn't match reality and it attempts to delete every time you deploy, resulting in errors.

To me it seems very common that CI environments won't be able to update the `.chalice/deployed` directory, and so this must be a widespread issue.

Start by tracing deployment handling around the .chalice/deployed/.json metadata and reproduce a deployment where the environment cannot update that file. Verify how a later deployment compares stored metadata with AWS state. Done means deployments no longer repeatedly attempt to delete resources after the original deployment succeeds.

Written by the indexing model from the issue text.