Dependency errors during undeploy process

[rhoboat]

Hi,

We are working on moving our resources to a new region (region-A) and removing the ones created in the old one (region-B).

As part of this process we’re destroying the resources from CLI using terragrunt destroy and then deleting the folders/files and creating a PR and merging it using GW Ref Arch pipeline.

Everything was working as expected until we went to remove the sample-app folders, since we had removed other resources used by the sample-app like the load balancers, now we’re getting a dependency error like this one:

[ecs-deploy-runner][2022-02-10T11:41:37+0000] [INFO] [infrastructure-deploy-script] 2022-02-10 11:41:37  Successfully
 checked out origin/master
[ecs-deploy-runner][2022-02-10T11:41:37+0000] [INFO] [infrastructure-deploy-script] 2022-02-10 11:41:37  Running 
command "terragrunt plan -input=false -destroy"
[ecs-deploy-runner][2022-02-10T11:41:39+0000] time=2022-02-10T11:41:39Z level=error msg=Could not convert include
 to the execution context to evaluate additional locals
[ecs-deploy-runner][2022-02-10T11:41:39+0000] time=2022-02-10T11:41:39Z level=error msg=Encountered error while
 evaluating locals.
[ecs-deploy-runner][2022-02-10T11:41:39+0000] time=2022-02-10T11:41:39Z level=error msg=Error reading file at path
 /tmp/tmp9vl1acw3/dev/us-east-2/dev/networking/alb: open 
/tmp/tmp9vl1acw3/dev/us-east-2/dev/networking/alb: no such file or directory

We already deleted these resources from the CLI (sample-app and ecs-cluster), but now we can’t push the changes to master using the pipeline. Is there are a recommended way to fix this issue?

Thanks

[rhoboat]

The Reference Architecture can be destroyed/undeployed following reverse dependency order, to avoid this issue where existing modules depend on already destroyed resources.

Here's an example undeploy documentation that you would have received in your Ref Arch deployment (look under /docs).

First recommendation, if you still have the code in your version control for the old region (region-B), restore the already deleted module and redeploy. Then undeploy using the reverse dependency order in that undeploy doc. This is a known path.

If that's not feasible because there's too many to restore, you can use the skip_outputs flag in the dependency blocks for the resources that have already been destroyed. Also you'll need to update the mock_outputs_allowed_terraform_commands to include plan and apply. By default we only specify validate. If you then run terragrunt plan -destroy or terragrunt apply -destroy, those dependencies will be ignored and you should be able to complete the destroy. I was able to test this out on a real Ref Arch to confirm this, but please let me know if this doesn't work.

[fsanzdev]

This suggestion worked, you can mark the response as answered