improving control flow for lb_v2 resources

[jrperritt]

The control flow of waiting for LB v2 resources to become ready to create/update could be improved. Currently, for example, it just immediately retries if it gets a 409, which can lead to what appear to be race conditions in OpenStack. It seems like the problem should be fixed in OpenStack, but Terraform should handle this better by, for example, waiting until the LB is ACTIVE before continuing to try to create the LB pool after receiving a 409.

Terraform Version

0.9.8.

Affected Resource(s)

• openstack_lb_*_v2

Terraform Configuration Files

You'll need to provide image_id and flavor_id:

variable "subnet_id" {
  type = "string"
  default = "4409be68-f1d6-44a5-a54a-bc22e790b237"
}

variable "image_id" {
  type = "string"
}

variable "flavor_id" {
  type = "string"
}

resource "openstack_lb_loadbalancer_v2" "lb" {
  vip_subnet_id = "${var.subnet_id}"
  name = "lb"
}

resource "openstack_lb_listener_v2" "lb_listener" {
  loadbalancer_id = "${openstack_lb_loadbalancer_v2.lb.id}"
  protocol = "HTTPS"
  protocol_port = 443
}

resource "openstack_lb_pool_v2" "lb_pool" {
  lb_method = "ROUND_ROBIN"
  protocol = "HTTPS"
  listener_id = "${openstack_lb_listener_v2.lb_listener.id}"
}

resource "openstack_lb_member_v2" "lb_member" {
  address = "${openstack_compute_instance_v2.server.access_ip_v4}"
  pool_id = "${openstack_lb_pool_v2.lb_pool.id}"
  protocol_port = 443
  subnet_id = "${var.subnet_id}"
}

resource "openstack_lb_monitor_v2" "lb_monitor" {
  delay = 30
  max_retries = 3
  pool_id = "${openstack_lb_pool_v2.lb_pool.id}"
  timeout = 5
  type = "PING"
}

resource "openstack_compute_instance_v2" "server" {
  name      = "server"
  image_id  = "${var.image_id}"
  flavor_id = "${var.flavor_id}"
  config_drive = false
}

Expected Behavior

1 LB pool gets created

Actual Behavior

4 LB pools get created

Steps to Reproduce

1. terraform apply

[jtopjian]

@jrperritt Is this similar to hashicorp/terraform#15225 or is a 409 being hit elsewhere?

I should probably just blanket all LB resources with 409 checks.

[jrperritt]

Yes, similar to that, but the fix that was implemented is not sufficient. Instead of immediately retrying the request, waitForLoadBalancerActive (or similar) should be called to wait for an ACTIVE state.

[jrperritt]

I'll submit a fix for this.

[jtopjian]

#25 has been merged. I'm sorry for the delay on that one.

[jtopjian]

Fixed by #29