infrastructure_adapters

---

iso_doc_type: "Description" iso_viewpoint: "ComponentView" type: "architecture" title: "Infrastructure Adapters & Platform Integrations" description: "ISO 42010 ComponentView / ISO 15289 Description documentation detailing external platform adapters in io/ and controller/." tags: ["iso42010", "okf", "component_view", "infrastructure", "adapters"] timestamp: "2026-10-01T17:50:00Z" generated: "agent:wiki-iso-documentation-standard" verified: "true" last_verified_commit: "073b5fd"

Infrastructure Adapters & Platform Integrations

Purpose: Document external connectivity adapters, storage systems, message brokers, and telemetry backends.

---

1. External Infrastructure Topology

graph LR
    subgraph CoreApp["MLOps Python Package"]
        IO_SRV["io/services.py"]
        IO_REG["io/registries.py"]
        IO_DATA["io/datasets.py"]
        CTRL["controller/kafka_app.py"]
    end

    subgraph ExternalServices["External Infrastructure Platforms"]
        MLFLOW[("MLflow Tracking & Registry")]
        KAFKA{"Confluent Kafka Broker"}
        S3[("MinIO / S3 Object Store")]
        DVC["Data Version Control (DVC)"]
        NOTIF["Notification Gateway (Alerts)"]
    end

    IO_SRV -->|Tracking URI & REST API| MLFLOW
    IO_REG -->|Model Registry API| MLFLOW
    IO_DATA -->|Parquet Read/Write| S3
    IO_DATA -->|Pointers & Metadata| DVC
    CTRL -->|TCP 9092 & Consumer Groups| KAFKA
    IO_SRV -->|Desktop / Webhook Alerts| NOTIF

    classDef core fill:#1e293b,stroke:#38bdf8,stroke-width:2px,color:#f8fafc;
    classDef ext fill:#111827,stroke:#34d399,stroke-width:2px,color:#f8fafc;
    class IO_SRV,IO_REG,IO_DATA,CTRL core;
    class MLFLOW,KAFKA,S3,DVC,NOTIF ext;

Loading

---

2. Infrastructure Adapters Breakdown

1. MLflow Tracking & Registry (io/services.py, io/registries.py)

• Protocol: HTTP/REST and Python MLflow Client (mlflow.tracking.MlflowClient).
• Configuration: Injected via osvariables.py (MLFLOW_TRACKING_URI, MLFLOW_REGISTRY_URI).
• Features: Experiment isolation, metric logging (MSE, RMSE, R2), artifact versioning, and champion-challenger alias assignment.

2. Parquet Storage & DVC Integration (io/datasets.py)

• Format: Apache Parquet columnar binary with snappy compression.
• Implementations:
  • ParquetReader: Enforces optional row limits and extracts dataset lineage metadata.
  • ParquetWriter: Atomically writes DataFrames to persistent storage.
• DVC Sync: Datasets in data/ are tracked using .dvc sidecar files committed to Git.

3. Kafka Streaming Broker (controller/kafka_app.py)

• Client: confluent-kafka (C-optimized librdkafka bindings).
• Topology: Input topic for incoming JSON requests; output topic for prediction responses.
• Resilience: Consumer poll loop with automatic partition assignment, graceful shutdown handlers, and error recovery.

4. Logging & Alerting Services (io/services.py)

• LoggerService: Configures Loguru sinks with structured serialization and level filtering.
• AlertsService: Dispatches system notifications upon pipeline failures.

---

Related: Tactical Design · Production Operations · Master Index