application placement to any machine/container/kvm doesn't work if machine is not started

[arif-ali]

Description

When you create tf plan with physical + lxd + application, then this will not apply in the first instance.

We now have to wait for the lxd to be active for the application to be added.

This was working with juju 2.9

Urgency

Annoying bug in our test suite

Terraform Juju Provider version

0.12.0

Terraform version

1.9.0

Juju version

3.4.3

Terraform Configuration(s)

resource "juju_machine" "machine" {
  model       = "test"
}

resource "juju_machine" "cinder" {
  model       = "test"
  placement   = "${join(":",["lxd",juju_machine.machine.machine_id])}"
  constraints = "spaces=oam"
}

resource "juju_application" "cinder" {
  name = "cinder"

  model = "test"

  charm {
    name     = "cinder"
    channel  = "yoga/stable"
    base     = "[email protected]"
  }

  units = 1

  placement = juju_machine.cinder.machine_id

  endpoint_bindings = [{
    space    = "oam"
  },{
    endpoint = "public"
    space    = "oam"
  },{
    endpoint = "admin"
    space    = "oam"
  },{
    endpoint = "internal"
    space    = "oam"
  },{
    endpoint = "shared-db"
    space    = "oam"
  }]

}

Reproduce / Test

* Create a fresh model called `test`
* Then run terraform apply with the above plan, and we will get the following issue


juju_machine.machine: Creating...
juju_machine.machine: Creation complete after 5s [id=test:0:machine-0]
juju_machine.cinder: Creating...
juju_machine.cinder: Creation complete after 5s [id=test:0/lxd/0:machine-0/lxd/0]
juju_application.cinder: Creating...
╷
│ Error: Client Error
│ 
│   with juju_application.cinder,
│   on 05-cinder.tf line 11, in resource "juju_application" "cinder":
│   11: resource "juju_application" "cinder" {
│ 
│ Unable to create application, got error: machine "0/lxd/0" not started, please retry when started
╵

### Debug/Panic Output

_No response_

### Notes & References

_No response_

[arif-ali]

after doing some other work, this is not just an LXD issue, but any machine. My simple plan below had to be run in 2 stages

resource "juju_model" "microk8s" {
  name = var.mk8s-model-name

  cloud {
    name = var.cloud
  }

  config = {
    apt-mirror = "http://${var.apt_mirror}/archive.ubuntu.com/ubuntu"
    lxd-snap-channel = var.lxd-snap-channel

    container-image-metadata-url = "http://${var.apt_mirror}/lxd/"
    container-image-stream = "released"

    agent-metadata-url = "http://${var.apt_mirror}/juju/tools/"
    agent-stream = "released"
  }
}


resource "juju_machine" "mk8s" {
  for_each    = {
  for index, machine in var.mk8s-machines:
  machine.machine_id => machine
  }
#  model       = var.mk8s-model-name
  model       = juju_model.microk8s.name
  name        = each.value.machine_id
  constraints = each.value.constraints
  #base        = each.value.base
}

resource "juju_application" "microk8s" {
  name = "microk8s"

  #model = var.mk8s-model-name
  model = juju_model.microk8s.name

  charm {
    name     = "microk8s"
    channel  = "1.28/stable"
    base     = "[email protected]"
  }

  units = 4

  placement = "${join(",", sort([
    for res in juju_machine.mk8s :
        res.machine_id
  ]))}"

  config = {
  }
}

The error below

juju_model.microk8s: Creating...
juju_model.microk8s: Creation complete after 2s [id=9c93b2ac-8cbb-4eba-8d94-aeaec1f1d1d1]
juju_machine.mk8s["2002"]: Creating...
juju_machine.mk8s["2000"]: Creating...
juju_machine.mk8s["2003"]: Creating...
juju_machine.mk8s["2001"]: Creating...
juju_machine.mk8s["2001"]: Creation complete after 8s [id=microk8s:0:2001]
juju_machine.mk8s["2002"]: Creation complete after 8s [id=microk8s:1:2002]
juju_machine.mk8s["2000"]: Still creating... [10s elapsed]
juju_machine.mk8s["2003"]: Still creating... [10s elapsed]
juju_machine.mk8s["2000"]: Creation complete after 13s [id=microk8s:2:2000]
juju_machine.mk8s["2003"]: Creation complete after 20s [id=microk8s:3:2003]
juju_application.microk8s: Creating...
╷
│ Error: Client Error
│ 
│   with juju_application.microk8s,
│   on 20-microk8s.tf line 33, in resource "juju_application" "microk8s":
│   33: resource "juju_application" "microk8s" {
│ 
│ Unable to create application, got error: machine "0" not started, please retry when started
╵

[hmlanigan]

This is juju behavior and not specific to the terraform provider. To deploy an application juju must know the operating system to use. If an actual machine is used in deployment, the operating system of the machine must be known. It is used to deploy the application, or to validate against the operating system specified for the application.

The fix is for the terraform provider to wait for the machine to finish provisioning before returning from creating the machine resource. Thus delaying the application deployment.

[hmlanigan]

See #116 for additional info