How ai-solution-eng/frameworks works
Project type and purpose
• This is not a full-stack application but an open-source “packaging and deployment” tool: a repository of Helm charts for multiple frameworks, each adapted (“ported”) to deploy on HPE Ezmeral Container Platform (PCAI).
• Its goal is to provide versioned charts, porting documentation, and packaged artifacts (tar.gz) so that end users or CI pipelines can import these frameworks into PCAI.
File structure
• Top-level entries are framework folders (argo-cd, grafana, langflow, minio, open-webui, postgresql, triton-inference-server, weaviate).
• Inside each framework folder:
• A logo file (used by PCAI import UI)
• A porting.md document (describing how customizations were made)
• A versioned subfolder (x.y.z) containing a Helm chart directory structure (Chart.yaml, values.yaml, templates/, ci/)
• A packaged tar.gz of the chart (FRAMEWORK-x.y.z.tar.gz)
• The uniform pattern implies a templated pipeline that: checks out source → builds Helm charts → packages artifacts → publishes to PCAI.
README for additional insights
• Confirms that each framework is ported to PCAI, with pointers to HPE docs and tutorial references.
• No runtime code here: it’s all chart definitions and porting guides.
Components
a. Source repository (Git)
b. Continuous Integration/Delivery pipeline
– Chart linting & validation (helm lint, tests/*)
– Packaging (helm package)
– Artifact storage (Helm chart repository or OCI registry)
c. PCAI Import Service
– UI or API that ingests tar.gz + logo + porting.md
– Persists chart into the PCAI catalog
d. Target PCAI Kubernetes Cluster(s)
– When a user selects a framework, the chart is deployed via Helm, creating Deployments, StatefulSets, Services, CRDs, etc.
e. Framework components (inside each chart)
– Kubernetes resources in templates/ (Deployments, Services, Ingress, ConfigMaps, RBAC, PodMonitors, etc.)
– CI overrides for different scenarios (ci/*.yaml)
Relationships and interactions
• Developers push chart source → CI pipeline runs → produces FRAMEWORK-x.y.z.tar.gz + index update → publishes to chart repo.
• PCAI Import Service polls or receives webhook from chart repo → imports chart metadata + logo + porting.md.
• End users deploy charts from PCAI catalog → Helm client in PCAI invokes Kubernetes API to create resources on target cluster.
• Dependencies between components: CI depends on repo, PCAI depends on chart repo, Helm depends on Kubernetes.
Architectural patterns and design principles
• Template-driven configuration (Helm’s values + templates)
• Versioned artifacts (x.y.z folder + tar.gz)
• Separation of concerns: chart authorship vs. chart packaging vs. chart import vs. chart deployment
• Extensibility: adding a new framework simply means adding a new top-level folder with the same structure.
Technologies, frameworks, or libraries
• Helm v3
• Kubernetes API objects (Deployments, StatefulSets, Services, Ingresses, RBAC, NetworkPolicies, etc.)
• PCAI Import API/UI (HPE Ezmeral)
• CI tools (e.g., GitHub Actions, Jenkins)—referenced by the presence of ci/*.yaml
Detail without over-segmentation
• logical layer (Source → Build → Store → Import → Deploy)