argocd-gitops-ai-workloads

Production GitOps patterns for AI/ML workloads: ArgoCD ApplicationSet fleet management, repo separation strategy, progressive canary delivery, and GPU resource templates.

Patterns

ApplicationSet (patterns/application_set.py)

• ApplicationSetGenerator — LIST/GIT/MATRIX/CLUSTER generators with to_dict()
• ApplicationSet — full ApplicationSet resource with sync policy
• SyncPolicy — MANUAL/AUTO/AUTO_PRUNE/AUTO_SELF_HEAL
• build_applicationset() — convenience builder for AI workload fleets

Repo Separation (patterns/repo_separation.py)

• RepoSeparationStrategy — app-repo vs config-repo separation with promotion_pipeline()
• RepoLayout — generates all expected directory paths for a config repo
• GitOpsRepo — typed repo descriptor with short_name and is_config_repo
• validate_repo_layout() — warns on missing prod, empty apps, conflicting tooling

Progressive Delivery (patterns/progressive_delivery.py)

• ProgressiveDelivery — canary rollout state machine (start/advance/abort)
• CanaryConfig — steps with weight/pause/analysis_template per step
• RolloutStatus — phase tracking (RUNNING/PAUSED/COMPLETED/FAILED)
• to_argo_rollout_dict() — generates Argo Rollouts resource YAML

Resource Templates (patterns/resource_templates.py)

• ResourceTemplateBuilder — generates Deployment + ResourceQuota dicts
• AIWorkloadConfig — per-workload GPU/CPU/memory config with effective_profile()
• GpuType — T4/A10G/A100/H100/H200 with VRAM properties
• WorkloadType — LLM_INFERENCE/BATCH/TRAINING/FINE_TUNING/EMBEDDING

Installation

pip install -e ".[dev]"
pytest -q