bigquery-graph

Use the GRAPH clause followed by MATCH patterns with node and edge labels, then RETURN the results. BigQuery supports only the GoogleSQL GQL standard, never Cypher, and path traversals should bind patterns to path variables like MATCH p = (a)-[e]->(b).

Query semantic graphs with the GRAPH_EXPAND table-valued function, passing the full graph name as a string. Never query the graph as a regular table, and wrap measure columns in the AGG() function instead of standard aggregations like SUM or COUNT.

Use GRAPH_TABLE only when you need SQL integration such as aggregations, joins with relational tables, or advanced filtering on graph results. For pure graph pattern matching, prefer standalone GQL with native RETURN statements.

GRAPH_EXPAND requires a tree structure and rejects cyclic graphs, self-loops, disconnected components, and convergent diamond paths. For such topologies, use standard GQL pattern matching or split the graph into separate GRAPH_EXPAND calls.

Measure columns cannot be selected directly or used in native GQL queries. Wrap them in the AGG() function within a GRAPH_EXPAND query, and avoid SELECT * since it implicitly includes measures.

GRAPH_EXPAND prefixes properties with node or edge aliases, so combinations like node N with property a_b and node N_a with property b collide. Use distinct safe aliases with AS and design property names to prevent prefix-induced duplicates.