[Bug]: aws_sagemaker_image_version doesn't keep version status

[esgrano-redeploy]

Terraform Core Version

1.9.8

AWS Provider Version

5.73.0

Affected Resource(s)

aws_sagemaker_image_version

Expected Behavior

I am writing a module to create a Sagemaker image with multiple versions.

resource "aws_sagemaker_image_version" "latest" {
  image_name = aws_sagemaker_image.this.id
  base_image = "XXXXXXXXXXX.dkr.ecr.eu-central-1.amazonaws.com/sagemaker-custom-ce:latest"
}

resource "aws_sagemaker_image_version" "v0-1-0" {
  image_name = aws_sagemaker_image.this.id
  base_image = "XXXXXXXXXXX.dkr.ecr.eu-central-1.amazonaws.com/sagemaker-custom-ce:v0.1.0"
}

I am expecting these version to be created and if I run TF again to not be changed.

Actual Behavior

Every time I run terraform, TF wants to delete the previous version and the next one it looks picking un a random base-image value. When I create one version looks fine, however if I try to create a version for the same image the first time it wants to delete the previous version (which it tries every time but fails) and then replace the resource.
This is the first run after I added a new version:

Second run

and then on the console:

Relevant Error/Panic Output Snippet

No response

Terraform Configuration Files

terraform {
  required_version = ">= 1.3.6"
    
  required_providers {
    aws = {
      source  = "hashicorp/aws"
      version = ">= 5.73.0"
    }
  }
}

resource "aws_sagemaker_image" "this" {
  image_name   = "sagemaker-custom-ce-test-2"
  role_arn     = "<ROLE_ARN>
  display_name = "sagemaker-custom-ce"
  description  = "Our custom base image"
}

resource "aws_sagemaker_image_version" "latest" {
  image_name = aws_sagemaker_image.this.id
  base_image = "XXXXXXXXXX.dkr.ecr.eu-central-1.amazonaws.com/sagemaker-custom-ce:latest"
}

resource "aws_sagemaker_image_version" "v0-1-0" {
  image_name = aws_sagemaker_image.this.id
  base_image = "XXXXXXXXXX.dkr.ecr.eu-central-1.amazonaws.com/sagemaker-custom-ce:v0.1.0"
}

Steps to Reproduce

terraform plan
terraform apply

multiple times

Debug Output

No response

Panic Output

No response

Important Factoids

No response

References

No response

Would you like to implement a fix?

None

[github-actions]

Community Note

Voting for Prioritization

• Please vote on this issue by adding a 👍 reaction to the original post to help the community and maintainers prioritize this request.
• Please see our prioritization guide for information on how we prioritize.
• Please do not leave "+1" or other comments that do not add relevant new information or questions, they generate extra noise for issue followers and do not help prioritize the request.

Volunteering to Work on This Issue

• If you are interested in working on this issue, please leave a comment.
• If this would be your first contribution, please review the contribution guide.

[mlnrt]

I would add to this that if you want to delete an image version, for example, go from:

resource "aws_sagemaker_image_version" "v2" {
  image_name = aws_sagemaker_image.this.id
  base_image = "XXXXXXXXXX.dkr.ecr.eu-central-1.amazonaws.com/sagemaker-custom-ce:2.0"
}

resource "aws_sagemaker_image_version" "v1" {
  image_name = aws_sagemaker_image.this.id
  base_image = "XXXXXXXXXX.dkr.ecr.eu-central-1.amazonaws.com/sagemaker-custom-ce:1.0"
}

to

resource "aws_sagemaker_image_version" "v2" {
  image_name = aws_sagemaker_image.this.id
  base_image = "XXXXXXXXXX.dkr.ecr.eu-central-1.amazonaws.com/sagemaker-custom-ce:2.0"
}

Terraform will delete the latest image (v2 in the example above) instead of deleting the specific version corresponding to the deleted resource. And this creates issues if the latest image is the one which is now being used.

[goncalocapelalopes]

Bump. This makes the image version attachments to a sagemaker domain through Terraform unreliable, I currently have to do it manually through the console everytime.

[timschill]

This should be prioritized since it basically renders the resource useless.

[DrFaust92]

I think the gap is not passing version as part of delete (possible in API but is not done). will take a look. I hope there is no change needed in resource ID structure for it

[jar-b]

This will indeed require an update to the resource ID as both image_name and version will now be required for the read operation to fetch the appropriate version (rather than always retrieving the latest).