Terraform-provider-kubernetes: Changes to affinity label / values in Deployment / StatefulSet force a new resource.

[jaceq]

Terraform Version

tf: 0.12.12
kubernetes-provider: 1.9.0 and 1.10.0

Affected Resource(s)

kubernetes_deployment
kubernetes_statefulset

Terraform Configuration Files

    template {
      metadata {
        namespace = var.namespace
        labels = {
          app     = var.name
          app-ver = "${var.name}-${var.image_tag}"
        }
      }
affinity {
 pod_anti_affinity {
              required_during_scheduling_ignored_during_execution {
                label_selector {
                  match_expressions {
                    key      = "app-ver"
                    operator = "In"
                    values   = ["${var.name}-${var.image_tag}"]
                  }
                }
                topology_key = "kubernetes.io/hostname"
              }
            }
}
}

Expected Behavior

A change to pod affinity label or value should do an in-place replacement and resource should NOT be recreated.
In fact any update to pod template defined as part of deployment / statefulset should be a in-place update and should not force a recreation of deployment / statefulset.

Actual Behavior

Resource will get replaced:

                  ~ affinity {

                      ~ pod_anti_affinity {

                          ~ required_during_scheduling_ignored_during_execution {
                              - namespaces   = [] -> null
                                topology_key = "kubernetes.io/hostname"

                              ~ label_selector {
                                  - match_labels = {} -> null

                                  ~ match_expressions {
                                        key      = "app-ver"
                                        operator = "In"
                                      ~ values   = [ # forces replacement
                                          - "app1-v1",
                                          + "app1-v2",
                                        ]
                                    }
                                }
                            }
                        }
                    }

Steps to Reproduce

1.Define a deployment or statefulset with affinity rules

1. change label OR value

[shlomimatichin]

we are also experiencing same issue (just migrating from kubernetes provider 1.3...)
@jaceq have you found a solution?

[shlomimatichin]

fix in PR #755, compiled release for linux x64 here: https://github.com/shlomimatichin/terraform-provider-kubernetes/releases/tag/v1.10.0-custom

[jaceq]

@shlomimatichin I just don't change it for now, but your PR will help a lot, let's hope it will get merged soon.

[bpoland]

A workaround I've found is to manually edit the resource with the new affinity rules. Then a terraform apply won't try to change it and does not cause a recreate. Definitely not ideal though.

[jaceq]

@bpoland While this works, it's not possible to do that in CI pipelines (and on my side this is the case)

[shlomimatichin]

@bpoland if you please +1 the PR, it might helpto increase the chance someone might merge it.

[dchi-sa]

Hello @jaceq
i found this workaround to work.
add this to the end of you deployment
lifecycle { ignore_changes = [ # There's a bug with declaring a pod affinity rule in terraform # which causes the resources to be force replaced regardless of changes spec[0].template[0].spec[0].affinity ]

after testing this, these are the results for my particular deployment
before:
# kubernetes_deployment.ipintelsvc_deployment[0] must be replaced
after:
# kubernetes_deployment.ipintelsvc_deployment[0] will be updated in-place

of course your mileage may vary depending on whether you have other updates to your resource or not.

[jaceq]

This seems to be long resolved.

[hashibot[bot]]

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 feel this issue should be reopened, we encourage creating a new issue linking back to this one for added context. If you feel I made an error 🤖 🙉 , please reach out to my human friends 👉 [email protected]. Thanks!