Summary

Fixes JSON serialization issues with the Vultr cloud provider's startup script handling.

Problem

The startup_script module was failing with:

Object of type 'bytes' is not JSON serializable by the 'tagless' profile.

This occurred due to complex interactions between:

• Nested template lookups in cloud-init/base.yml
• The "tagless" JSON serialization profile in Ansible 2.19+
• The vultr.cloud.startup_script module's base64 encoding logic

Solutions

1. Two-Step Fact Assignment

Implemented the Linode provider pattern: evaluate the template into a fact first, then pass that fact to the module. This ensures the template is fully evaluated before JSON serialization.

Change:

- name: Set cloud-init script as fact
  set_fact:
    algo_cloud_init_script: "{{ lookup('template', 'files/cloud-init/base.yml') }}"

- name: Upload the startup script
  vultr.cloud.startup_script:
    name: algo-startup
    script: "{{ algo_cloud_init_script }}"

2. Cloud-Init Template String Filters

Added | string filters to nested lookups in files/cloud-init/base.yml to ensure consistent string handling:

# SSH config template
{{ lookup('template', 'files/cloud-init/sshd_config') | string | indent(width=6, first=True) }}

# SSH public key
- "{{ lookup('file', SSH_keys.public) | string }}"

Known Issue - Dependency Bug

IMPORTANT: There is a bug in vultr.cloud collection v1.13.0 that prevents this from working completely.

The module's configure() method calls:

self.module.params["script"] = base64.b64encode(self.module.params["script"].encode())

The base64.b64encode() returns bytes, which cannot be JSON serialized. It should be:

self.module.params["script"] = base64.b64encode(self.module.params["script"].encode()).decode('ascii')

Upstream bug report: Will be filed at https://github.com/vultr/ansible-collection-vultr/issues/

Until this is fixed upstream, users will need to manually patch the module or wait for a new release.

Testing

• ✅ Cloud-init template evaluates without undefined markers
• ✅ All nested template lookups return strings
• ⚠️ Startup script upload requires patched vultr.cloud module

Impact

• Improves robustness of cloud-init template for all providers
• Follows established patterns from other cloud providers (Linode)
• Backward compatible - all changes are defensive improvements
• Blocked by upstream bug in vultr.cloud collection

Files Changed

• roles/cloud-vultr/tasks/main.yml - Two-step fact assignment for startup script
• files/cloud-init/base.yml - String filters on nested lookups (shared by all providers)

Technical Details

The vultr.cloud.startup_script module (v1.13.0) calls base64.b64encode(script.encode()) internally. When combined with Ansible 2.19's "tagless" JSON serialization profile, certain template evaluation patterns can result in bytes objects that fail serialization. The two-step fact assignment ensures the template is fully evaluated to a string before entering the module's processing pipeline.

However, the module itself has a bug that creates bytes that cannot be serialized. This PR contains the Algo-side fixes, but full functionality depends on the upstream fix.