No space left on device with VHD target, only with Ansible module

[Procsiab]

I am reporting this since I cannot reproduce it from composer-cli, and from the metadata produced by the compose it seems that the custom size I set for the image artefact is respected, however: every compose that I run with the infra.osbuild.start_compose module and VHD target fails at the step org.osbuild.copy with the following error:

cp: error writing '/run/osbuild[...]': No space left on device

If I run the same compose from the CLI, it succeeds (I use the same parameters declared as the module's arguments).

I am using the infra.osbuild collection version 2.2.0, on a Fedora 38 system with Python 3.9.16
I attach the metadata and log files for both the succeeding job run from CLI and the failing one run from Ansible, and also the blueprint file and the command/task I used to run the respective composes.

Successful CLI run

composer-cli compose start datalab_general vhd datalab_general template_static/az_storage_account.toml --size 5120

Failed Ansible run

hosts: localhost
tasks:
  - name: Start the compose job, with VHD target                                                       
    infra.osbuild.start_compose:                                                                       
      blueprint: "{{ blueprint_name }}"                                                                
      size: 5120                                                                                       
      profile: "{{ lookup('ansible.builtin.template', 'templates/' + storageprof_name + '.toml.j2') }}"
      image_name: "{{ blueprint_name }}_img"                                                           
      allow_duplicate: true  # not idempotent                                                          
      compose_type: 'vhd'                                                                              
      timeout: 150                                                                                     
    register: builder_compose_start_out 

infra_osbuild_issue.zip

[Procsiab]

I think that this issue may be related to the fact that the size parameter is handled differently between the module_utils/weldrapiv1 implementation and the one found inside the weldr-client (composer-cli) project:

In the first case, the value is passed directly inside the HTTP request body, assuming it's in MB (https://github.com/redhat-cop/infra.osbuild/blob/958dcaba25c9b3ec402cbccce5d920b303e93525/plugins/module_utils/weldrapiv1.py#L294C18-L294C34)

In the second instance, the size is assumed to be in Bytes, therefore it is multiplied by 1024² before being sent inside the HTTP request to the Weldr API (https://github.com/osbuild/weldr-client/blob/cddc652a655a6cbc5144a417edf8c1c73bf49e98/weldr/compose.go#L160)

My assumption is supported by the fact that the following task can correctly queue and finish a build for a VHD (while I could not achieve this using the start_compose module):

    - name: Start the compose job, with VHD target
      ansible.builtin.uri:
        unix_socket: /run/weldr/api.socket
        url: http://localhost/api/v1/compose
        method: POST
        timeout: 150
        body_format: json
        body:
          blueprint_name: "{{ blueprint_name }}"
          compose_type: 'vhd'
          branch: 'master'
          size: 5368709120
      register: builder_compose_start_out

TL;DR

It must be assumed that the size for the start_compose module is in fact in Bytes (not in MiB, as stated here)