Address MetalLB CRDs Race Condition

[shakir85]

Some ideas to give the API Server enough time to reigster the custom resources before start deploying CRDs

Option 1

In the metal-crd module, add this:

resource "null_resource" "wait_for_metallb_crds" {
  provisioner "local-exec" {
    command = <<EOT
    for crd in ipaddresspools.metallb.io l2advertisements.metallb.io; do
      echo "Waiting for CRD $crd to be registered..."
      until kubectl get crd $crd >/dev/null 2>&1; do
        sleep 5
      done
    done
    EOT
  }
}

Then make the resources depend on it

resource "kubernetes_manifest" "ip_address_pool" {
  manifest = { ... }
  depends_on = [null_resource.wait_for_metallb_crds]
}

Option 2

It looks like Terraform 1.13+ has a time_sleep resource
https://github.com/hashicorp/terraform-provider-time/blob/main/docs/resources/sleep.md

Maybe combine MetalLB chart and its CRDs in a single module and do something like this

resource "time_sleep" "wait_30s" {
  depends_on = [helm_release.metallb] # or wherever CRDs are applied
  create_duration = "30s"
}

Then depend on it

resource "kubernetes_manifest" "ip_address_pool" {
  depends_on = [time_sleep.wait_30s]
}

Option 3

Handle this outsitde Terraform/grunt, in GHA for example.