Skip to content

DrToller.Framework

A machine-readable execution contract for data workloads. Humans extend primitives; humans and AI compose pipelines.

DrToller.Framework is a declarative execution framework for data and ML workloads. DRTML expresses the logical shape of a workload as a machine-readable execution contract; the Framework validates, compiles, and executes it across interchangeable backends.

The framework owns:

  • loading and compiling DRTML (*.drtml + fragments);
  • storage (parquet / PostgreSQL / Qdrant / neo4j stub) via a single adapter registry;
  • processing paths: partition_loop, mapping, reduce_merge, dataset_metrics, evaluation, embed_train;
  • telemetry (Prometheus emit, Grafana codegen);
  • integrations (Streamlit, Airflow, Ray connect, local runner).

Step canon: domain logic interacts through explicit contracts — receive data → transform → return data. Infrastructure (I/O, Prometheus, Airflow/Ray APIs) stays in the Framework. How this layer relates to Spark, Ray, Airflow, and agent stacks: Compare.

drtoller/framework/
contracts/ # layer-neutral DTOs (saturation, worker delta)
drtml/ # load / validate / compile
runtime/ # RunContext, import_symbol, pattern_runners
storage/ # session, feed, backends, registry adapters
processing/ # dispatch + modes; job_map / job_session
mapping/ # merge / remap algorithms
algorithms/ # embed, cluster, dataset metrics
db/ # postgres / qdrant transport
telemetry/ # prometheus + grafana codegen
integration/ # UI, Airflow, Ray, env
ext/ # Extension SDK (register_*, load_plugins)
examples/ # hello-step (dev smoke, not installed)

Dependency canon:

drtml → contracts / storage.plan → db|storage → processing → telemetry|integration

Reverse edges are forbidden. Full rules: Principles.

The documentation covers DRTML v4, framework boundaries, storage backends, processing modes, algorithms, observability, and deployment. Recent changes: What’s new. Start and poll a run over HTTP: Run API.