kubernetes_horizontal_pod_autoscaler: scaling based on cpu and memory produces updates each run

[hydrapolic]

Terraform Version

0.14.10 and 0.15.1

Terraform Configuration Files

...
resource "kubernetes_horizontal_pod_autoscaler" "hpa" {
  metadata {
    name = "demo"
  }
  spec {
    min_replicas = 2
    max_replicas = 30
    scale_target_ref {
      api_version = "apps/v1"
      kind = "Deployment"
      name = "demo"
    }
    metric {   
      type = "Resource"
      resource {
        name = "cpu"
        target {
          type = "Utilization"
          average_utilization = "80"
        }
      }
    }
    metric {
      type = "Resource"
      resource {
        name = "memory"
        target {
          type = "Utilization"
          average_utilization = "80"
        }
      }
    }
  }
}

Debug Output

Crash Output

Expected Behavior

HPA should be applied only once.

Actual Behavior

Each time terraform apply is run a diff is shown.

Steps to Reproduce

1. terraform init
2. terraform apply

  # module.demo.kubernetes_horizontal_pod_autoscaler.hpa will be created
  + resource "kubernetes_horizontal_pod_autoscaler" "hpa" {
      + id = (known after apply)

      + metadata {
          + generation       = (known after apply)
          + name             = "demo"
          + namespace        = "default"
          + resource_version = (known after apply)
          + self_link        = (known after apply)
          + uid              = (known after apply)
        }

      + spec {
          + max_replicas                      = 30
          + min_replicas                      = 2
          + target_cpu_utilization_percentage = (known after apply)

          + metric {
              + type = "Resource"

              + resource {
                  + name = "cpu"

                  + target {
                      + average_utilization = 80
                      + type                = "Utilization"
                    }
                }
            }
          + metric {
              + type = "Resource"

              + resource {
                  + name = "memory"

                  + target {
                      + average_utilization = 80
                      + type                = "Utilization"
                    }
                }
            }

          + scale_target_ref {
              + api_version = "apps/v1"
              + kind        = "Deployment"
              + name        = "demo"
            }
        }
    }

3. terraform apply

  # module.demo.kubernetes_horizontal_pod_autoscaler.hpa will be updated in-place
  ~ resource "kubernetes_horizontal_pod_autoscaler" "hpa" {
        id = "default/demo"


      ~ spec {
            # (3 unchanged attributes hidden)

          ~ metric {
                # (1 unchanged attribute hidden)

              ~ resource {
                  ~ name = "memory" -> "cpu"

                    # (1 unchanged block hidden)
                }
            }
          ~ metric {
                # (1 unchanged attribute hidden)

              ~ resource {
                  ~ name = "cpu" -> "memory"

                    # (1 unchanged block hidden)
                }
            }

            # (1 unchanged block hidden)
        }
        # (1 unchanged block hidden)
    }

Additional Context

When only using cpu autoscaling, once applied, no changes are printed by running terraform apply. However when both
cpu and memory autoscaling is defined a diff is printed each time terraform apply is run swapping cpu for memory and vice
versa.

References

[jbardin]

Hi @hydrapolic,

Thanks for filing the issue. I have a feeling that the provider is changing some combination of values not shown in the diff which is triggering the output you see here. In order to work with legacy providers, terraform will hide some trivial changes that normally don't effect the result, like empty values being swapped for null values.

You may find more clues looking at the logs. I suspect you will find warnings about the provider producing an inconsistent plan for this resource.

Is there an easy way to setup a standalone reproduction of this configuration so we can troubleshoot it?

Thanks!

[jrhouston]

It looks like this is actually a provider bug – duplicate issue here: hashicorp/terraform-provider-kubernetes#1188

[hydrapolic]

Thanks @jrhouston for the linked bug, it's exactly the same issue.

[github-actions]

I'm going to lock this issue because it has been closed for 30 days ⏳. This helps our maintainers find and focus on the active issues.
If you have found a problem that seems similar to this, please open a new issue and complete the issue template so we can capture all the details necessary to investigate further.