Terraform cycle when altering a count

Question

I have some resources whose count is parameterised by a variable. This is used to create VM resources as well as null_resources for e.g. running deployment scripts on them. When I reduce the value of the count from 2 to 1 and apply, I get an error.

Terraform executes plan with no complaints. But when I apply, it tells me there is a cycle:

Error: Cycle: null_resource.network_connection_configuration[7] (destroy), null_resource.network_connection_configuration[8] (destroy), null_resource.network_connection_configuration[3] (destroy), null_resource.network_connection_configuration[4] (destroy), null_resource.network_connection_configuration[0] (destroy), null_resource.network_connection_configuration[6] (destroy), null_resource.network_connection_configuration[1] (destroy), null_resource.network_connection_configuration[9] (destroy), null_resource.network_connection_configuration[2] (destroy), null_resource.network_connection_configuration[10] (destroy), hcloud_server.kafka[2] (destroy), local.all_machine_ips, null_resource.network_connection_configuration (prepare state), null_resource.network_connection_configuration[5] (destroy)

Here is the relevant part of the file:

variable kafka_count {
  default = 3
}

resource "hcloud_server" "kafka" {
  count       = "${var.kafka_count}"
  name        = "kafka-${count.index}"
  image       = "ubuntu-18.04"
  server_type = "cx21"
}

locals {
  all_machine_ips = "${hcloud_server.kafka.*.ipv4_address)}"
}

resource "null_resource" "network_connection_configuration" {
  count = "${length(local.all_machine_ips)}"

  triggers = {
    ips = "${join(",", local.all_machine_ips)}"
  }

  depends_on = [
    "hcloud_server.kafka"
  ]

  connection {
    type        = "ssh"
    user        = "deploy"
    host        = "${element(local.all_machine_ips, count.index)}"
    port        = 22
  }

// ... some file provisioners
}

When I try to find the cycle using the visualisation:

terraform graph -verbose -draw-cycles

There are no cycles visible.

When I use TF_LOG=1 the debug log doesn't show any errors

So the issue is that I can increase the count but not decrease it. I don't want to manally hack the file as it means I won't be able to scale down in future! I'm using Terraform v0.12.1.

Are there any strategies for debugging this situation?

Answer 1

I had a similar issue with 0.12.x - I was calling a provisioner within an aws_instance resource, which was giving the same error you had when increasing the count for the resource.

I got around it by using the self object (self.private_ip) to reference the resource rather than using count.index or element().