Scale a "daemon" application not supported

[mcarvalhor]

Description

When we deploy a "daemon" juju application using terraform without explicit unit count, every terraform apply command fails with "Unable to update application resource, got error: scale a "daemon" application not supported".

We can see from the plan that juju terraform provider is trying to change the number of units, and it shouldn't.

The current workaround is: ignore the error.

Urgency

Annoying bug in our test suite

Terraform Juju Provider version

0.11.0

Terraform version

v1.7.2-dev

Terraform Configuration(s)

# This plan should be deployed on a Juju KUBERNETES model.
# Please notice that I am NOT specifying "units = X" for the applications.

resource "juju_application" "metallb-controller" {
  name  = "metallb-controller"
  model = "my-model-name"

  charm {
    name     = "metallb-controller"
    channel  = "latest/stable"
    revision = 46
  }

  config = {
    iprange  = "10.1.2.3/32" # This is my Kubernetes worker ingress IP list.
    protocol = "layer2"
  }
}

resource "juju_application" "metallb-speaker" {
  name  = "metallb-speaker"
  model = "my-model-name"

  charm {
    name     = "metallb-speaker"
    channel  = "latest/stable"
    revision = 41
  }
}

Reproduce / Test

$ # Juju Status is empty. This should be an empty kubernetes model.
$
$ terraform init && terraform apply -auto-approve

$ # Now Juju Status shows 1 unit for "metallb-controller" and X units for "metallb-speaker".
$ #  where X is dynamic and depends on the number of units on the underlying Kubernetes cloud - an information abstracted from this model.

$ # Now let's try to run `terraform apply` again. No changes should be expected.
$ terraform apply
$ # Hm. Actually we see changes (eg.: https://pastebin.canonical.com/p/WHgMTWBy2P/). This is wrong, and even when trying to apply it fails.

Debug/Panic Output

https://private-fileshare.canonical.com/~mcarvalhor/2024-04-16_terraform-provider-daemon-application-issue/.XWe1p1qVPiQQktN6nALn/tf-debug.log

Notes & References

I have a live environment on Canonical private cloud (Prodstack 6) where this issue is seen. Feel free to ping me on Mattermost if you'd like to troubleshoot this using that environment.

[hmlanigan]

@mcarvalhor What version if juju are the controller and model running?

[mcarvalhor]

Hi!

Controller: v3.1.8
Model: v3.1.8

[mcarvalhor]

Another workaround is adding:

  lifecycle{
    ignore_changes = [ units, ]
  }

This one is better as we don't even see the changes on the plan. But I guess the bug still needs a permanent fix.