Parquet And Lakehouse Sources

Use the lakehouse path when Parquet data is too large or too operationally important to copy into ShrikeDB graph storage. ShrikeDB can register the source, profile it, build reviewed mappings, and query it as a graph view.

Supported V1 Source Model

Item Status

Active format

Parquet.

Reserved future formats

Delta and Iceberg registration are unavailable in this workflow.

Source roots

Local Parquet files; object-store URI registration also accepts metadata, with runtime access requiring separate verification.

Credentials

References only: environment, named profile, external secret store, or none.

Follow Register, review, and query Parquet for the HTTP/MCP envelope, required fields, mapping review, and response interpretation. The names below are MCP tools, not shell commands.

Operator Flow

lakehouse_connect_source
  -> lakehouse_profile_connection
  -> review mapping draft
  -> lakehouse_validate_mapping_draft
  -> lakehouse_save_mapping
  -> lakehouse_publish_mapping
  -> lakehouse_query_execute or semantic_execute

What Profiling Reads

Parquet metadata includes columns and row groups. The current connection-profiling tool returns a workflow summary; prepare and review mappings against the actual file schema. See the workflow chapter for the distinction between metadata reports and source-backed query results.

What A Mapping Declares

A reviewed mapping connects Parquet columns to graph-view concepts:

  • node labels or concepts;

  • relationship types;

  • source and target columns;

  • projected properties;

  • confidence threshold;

  • lineage and artifact dependencies.

Mappings are database-scoped. One database’s lakehouse source should not be visible to another database unless registered there.

Querying Lakehouse Data

Lakehouse data can be queried through:

  • semantic natural language;

  • MCP lakehouse_query_execute;

  • the frontend lakehouse inspection panels.

For direct Cypher or SPARQL, import the desired data into ShrikeDB’s graph storage and use the regular graph query endpoint.

Semantic lakehouse use is suitability-gated. The active mapping must cover the question’s required concepts and constraints. If it does not, inspect the reported skip or fallback reason and correct the mapping or query.