AAP-36732: ModelPipelines: Configuration improvements: Remove env vars

[manstis]

Jira Issue: https://issues.redhat.com/browse/AAP-36732

Description

This PR removes the discrete (legacy) environment variables previously used to configure LLM services (WCA, Ollama, GRPC etc). All configuration is now controlled by a JSON (or YAML) block, stored in a ANSIBLE_AI_MODEL_MESH_CONFIG environment. Various examples are given in this PR.

I have included support for the legacy use of environment variables.

These are used if ANSIBLE_AI_MODEL_MESH_CONFIG is not set.

This is predominantly for the use by the Operator; where an older version uses the latest service version.

Testing

N/A. Internal refactoring of existing code.

Steps to test

N/A. Internal refactoring of existing code.

Scenarios tested

N/A. Internal refactoring of existing code.

Production deployment

• This code change is ready for production on its own
• This code change requires the following considerations before going to production:

[goneri]

Work as expected with the following configuration:

export ANSIBLE_AI_MODEL_MESH_CONFIG="
---
ModelPipelineCompletions:
  provider: dummy
ModelPipelineContentMatch:
  provider: dummy
ModelPipelinePlaybookGeneration:
  provider: dummy
ModelPipelineRoleGeneration:
  provider: dummy
ModelPipelinePlaybookExplanation:
  provider: dummy
ModelPipelineChatBot:
  provider: dummy
"

The part I found a bit annoying is that if my YAML syntax is incorrect, the exception is silently ignored.

[manstis]

@goneri

The part is dislike I found a bit annoying is that if my YAML syntax is incorrect, the exception is silently ignored.

That'll be caused by this where exceptions are effectively ignored.

Let me tinker a little before this is merged.

[goneri]

@goneri

The part is dislike I found a bit annoying is that if my YAML syntax is incorrect, the exception is silently ignored.

That'll be caused by this where exceptions are effectively ignored.

Let me tinker a little before this is merged.

Actually my configuration doesn't work because the config keys are missing:

[django]     |   File "/var/www/ansible-ai-connect-service/ansible_ai_connect/ai/api/model_pipelines/config_serializers.py", line 38, in to_internal_value
[django]     |     serializer = REGISTRY[provider_part["provider"]][Serializer](data=data["config"])
[django]     |                                                                       ~~~~^^^^^^^^^^
[django]     | KeyError: 'config'

[manstis]

@goneri

Actually my configuration doesn't work because the config keys are missing.

Excellent find. Thanks.

[manstis]

@goneri PR updated to better handle exceptions (and fixed KeyError: 'config').

[manstis]

This aligns the Python version with the other GHA definition.

3.11 is needed to support ExceptionGroup (used in this PR). It is also the version used to build the container.

[sonarqubecloud]

Quality Gate passed

Issues
1 New issue
0 Accepted issues

Measures
0 Security Hotspots
100.0% Coverage on New Code
0.0% Duplication on New Code

See analysis details on SonarQube Cloud

[jameswnl]

PR Deployment created in namespace wisdom-pr-1487

Look up the wisdom-service url in the cluster

[manstis]

See https://issues.redhat.com/browse/AAP-38734 re: failing "PR Deployment Testing" check.

I have however deployed the container built by this PR to stage2-west manually:

• namespace: manstis
• instance: my-aiconnect
• I won't paste routes etc on Git Hub.