Core concepts

Understand what a graph model maps, what publishing changes, and which data a query reads.

On this page

Sources, tables, and models

A source is a connection or collection of files that GraphLake can read. A table is a named input within that source. A graph model maps rows from one or more tables into labels and relationships.

For example, an accounts table in PostgreSQL can define Account nodes while a Parquet transfer table defines Transfer edges. The relationship mapping uses account keys to connect those two inputs.

See Connect data for direct source access and Local tables for scheduled Parquet copies.

Keys, endpoints, and properties

Node label
A named type such as Account. The business key identifies one node within that label.
Relationship type
A directed link such as Transfer. Its source and destination columns map to the complete keys of the two endpoint nodes, in order.
Property
A mapped column such as amount that you can display, filter, or aggregate. Distinct transfer IDs keep parallel transfers separate.

When account IDs repeat across tenants, use a composite key such as [tenant_id, account_id]. Both endpoint mappings must include the tenant column and use compatible types. Try this workflow in your first graph.

Drafts, published revisions, and data freshness

A draft is the editable configuration. Publishing makes a validated revision available to clients. Queries use the published revision and report that revision alongside source versions. Rolling back a model switches its mapping and leaves source data under the source system’s control.

A local table uses the last successfully published copy. A CSR index represents the graph scope and source state captured at build time. If you need the latest records, check the local table and CSR freshness before choosing which data to query.

See Model a graph for publishing and Graph algorithms for CSR selection.