Schema generation omits primary key dimensions for PostgreSQL tables, causing compile errors when joins are generated

Environment

Cube Version: Latest (cloned from cube-js/cube) Deployment: Docker Database: PostgreSQL Sample Database: Chinook Schema Format: JavaScript

Problem

When using Generate Data Model from the Playground, Cube successfully generates JavaScript schema files but fails to compile them because the generated cubes do not include primary key dimensions.

The generated cubes contain joins, but no dimension is marked as primary_key: true, resulting in compilation errors.

Example error:

Compile errors:

• album cube: primary key for 'album' is required when join is defined
• invoice cube: primary key for 'invoice' is required when join is defined
• invoice_line cube: primary key for 'invoice_line' is required when join is defined
• playlist_track cube: primary key for 'playlist_track' is required when join is defined
• track cube: primary key for 'track' is required when join is defined

Steps to Reproduce

• Clone the latest Cube repository.
• Start Cube using Docker.
• Connect a PostgreSQL database (tested with the Chinook sample database).
• Open Playground.
• Click Generate Data Model.
• Select JavaScript format.
• Generate schema.

Expected Behavior

Generated cubes should include primary key dimensions.

Example:

dimensions: { album_id: { sql: album_id, type: number, primary_key: true },

title: { sql: title, type: string } }

Actual Behavior

Generated schema:

cube(album, { sql_table: public.album,

joins: { artist: { sql: ${CUBE}.artist_id = ${artist}.artist_id, relationship: many_to_one } },

dimensions: { title: { sql: title, type: string } } });

The primary key dimension (album_id) is missing.

Database Metadata

The PostgreSQL database correctly exposes primary keys.

Example query:

SELECT tc.table_name, kcu.column_name FROM information_schema.table_constraints tc JOIN information_schema.key_column_usage kcu ON tc.constraint_name = kcu.constraint_name WHERE tc.constraint_type = 'PRIMARY KEY';

Result:

album album_id artist artist_id customer customer_id employee employee_id genre genre_id invoice invoice_id invoice_line invoice_line_id media_type media_type_id playlist playlist_id playlist_track playlist_id playlist_track track_id track track_id

Investigation

While debugging the source code:

• driver.tablesSchema() is used as the source for schema generation.
• ScaffoldingSchema expects primary keys to be available via: column.attributes?.includes("primaryKey")

However, the generated schema does not contain any primary key dimensions, suggesting that the metadata passed into the scaffolding process may not contain the expected primaryKey attribute.

Additional Observation

I had previously worked around this issue by replacing the older schema introspection implementation with TableSchemaV2(), after which primary keys were generated correctly.

This suggests there may be a regression or inconsistency between the current schema introspection implementation and the scaffolding generator.

Questions

• Is this expected behavior?
• Has TableSchema been superseded by TableSchemaV2 for schema generation?
• Could this be a regression in the latest Playground schema generator?

Additional Information

• PostgreSQL correctly exposes primary key metadata.
• Foreign keys are generated correctly.
• Joins are generated correctly.
• Only the primary key dimensions are omitted, causing schema compilation to fail.