Data sources are connected in the “Data Sources” section of the web interface: creating, connection checks, source details. Three database engines are supported — SQLite, PostgreSQL and ClickHouse — plus your own sources built on Python scripts.
Source types
- SQLite — path to the database file, timeout (30 s by default), read-only mode;
- PostgreSQL — host, port (5432), database, login, password, SSL, timeout;
- ClickHouse — host, HTTP port (8123), database, login, password, SSL, timeout;
- Python script — a source for any service that has a Python client: the source settings name the script key and the constant arguments.
A source from your own script
When there is no ready connector for your service, your own Python script can become the source. In the source settings you pick the script and set the constant arguments, and the script interprets the query text in its own way: it can be an address, a file path or an expression of your own — an agreement inside the script.
Such a source is the equal of the others: charts and tables are built on it, it passes validation, takes part in alert rules and lands in the PDF export. If a query to the service fails, the report shows a clear error instead of an empty chart. How to set up scripts — in the Python Scripts section.
Schema and result caching
- schema auto-detection (
autoDetect) — tables and columns are picked up automatically; - periodic schema refresh — by default once every 3600 seconds;
- table caching (
cacheTables) — the schema is not requested anew for every report; - caching of query results with statistics and cleanup — details in the Queries section.
Validation
DataSourceValidationService checks source availability: connection errors are visible before reports are built, not on a broken dashboard. A non-existent source in a block query is caught by the validator (DATA_SOURCE_NOT_FOUND).
File data without setup
If your data lives in files — Parquet, SQLite, Markdown — you don’t need to configure a source: put the files into the Lakehouse folder and query them right from the file browser. Details in the Lakehouse section.
A SQLite file in the root of the Lakehouse folder is discovered automatically and shown as a ready source: it is marked with a “lakehouse” icon in the list, and the connection is read-only. You don’t need to configure anything; if you want your own variant, create a regular source with the same name — your setting takes priority over the discovered one.
In every deployment the sources list shows sources named ml-demo_… — the demo databases from the template library. They are read-only and exist so that any template can be opened and studied before you have connected your own data. Create your own source with the same name, and it replaces the demo one.