Skip to main content

Get started

MDMongoose is master data management that runs entirely inside Databricks: one hierarchy stored in one Delta table, a web app on Databricks Apps, and incremental Spark Declarative Pipelines for loading and exporting. Everything deploys from one bundle — and it works on Databricks Free Edition.

The MDMongoose main screen

What is in this documentation​

PageRead it when you want to…
Overview of featuressee what MDMongoose does, feature by feature
Architectureunderstand how the app, the Delta table, jobs and pipelines fit together — with Ingestion, Data quality and Export
Deploymentinstall MDMongoose in your workspace and upgrade it later
Demo on Free Editiontry the full demo on a free Databricks account, step by step
User guidelearn the main screen: hierarchy, properties, actions and dialogs
Enum, Link and Arrayuse properties that point to other nodes
Edit Childrenedit many nodes at once in the Excel-like grid
Importing dataload a source table: staging table, record type, Table node, refresh, schedule
Creating master databuild golden records: matching, survivorship rules, master data records
Modifying dataadd, move, link and delete nodes and build hierarchies
Exporting to CSV and Exceldownload a branch from the app
Exporting with materialized viewsserve master data to other Databricks consumers
Securityconnect Unity Catalog groups and set permission levels
Monitoringcheck pipelines, jobs, data and data quality

The five-minute version​

  1. Deploy the bundle from a Databricks Git folder (Deployments panel → Deploy), then run the mdm_init job once. It creates the table, the system definitions and — if you keep the demo on — demo data with customers, suppliers, employees and products. → Deployment, or follow Demo on Free Edition for a free account
  2. Open the app (Compute → Apps → mdm). You land on the root of the hierarchy, MDM.
  3. Browse: System holds the definitions, Model your NodeTypes, tables and rules, Reference Data the loaded source records and Master Data the golden records. → User guide
  4. Load your own data: put a source table with change data feed into mdm_staging, create a Table node and click Load Childs. → Importing data
  5. Build golden records with a survivorship rule and Refresh Golden Records. → Creating master data

Key ideas​

  • Everything is a node. Reference values, source records, golden records, NodeTypes, PropertyTypes, actions, rules and permissions are all rows of the same table and all appear in the same tree.
  • Types are interfaces. A node can have several types; it shows the union of their properties and actions.
  • People win over automation. When someone edits a value that a load or a golden-record refresh produced, the value is marked UserDefined and automatic updates leave it alone.
  • Permissions follow the tree. The first Steward list from the root grants access; lists further down can only narrow it.
Try it without Databricks

The repository runs locally with an in-memory demo: pip install -e ".[app,dev]" and scripts/dev.sh, then open http://localhost:5173. Handy for evaluating the UI before you deploy.