Data Hub troubleshooting

These are roughly in the order they come up when you connect a source and put it to work.

A Windows file path loses its backslashes

Typing C:\data\app.db into a file path field stores C:dataapp.db. The backslashes are stripped as you type, so the connection then fails with “Database not found” even though the path you entered was correct.

Write the path with forward slashes instead — C:/data/app.db. Node accepts them on Windows, and they survive the field.

This affects the file-path connectors, SQLite and DuckDB. It has been reported.

The connection test fails

Read the message rather than assuming it is credentials.

“Database not found” means the path or database name is wrong. For file-based sources, check the backslash problem above. Remember the path is resolved by the server, not your browser, so it has to exist wherever the platform runs.

Authentication failures name the user. Network failures mention the host or a timeout, and usually mean a firewall or a database not listening where you expect.

I saved a connection whose test failed

Nothing stops you. The wizard allows Create & Link regardless of the test result, so a connection can exist and be permanently offline.

Select it in the sidebar and use Edit to fix the details, then test again. The status dot on the connection tells you where you stand.

The connection is online but has no tables

The credentials work and the database is reachable, but the schema is empty from this user’s point of view.

Usually that is a schema setting — the connector is looking at public and your tables are elsewhere — or a permissions problem on the database side, where the user can connect but cannot see the tables. Both are fixed in the connection settings or on the database, not in Data Hub.

Tables are listed but other modules see nothing

They are not enabled. Every table arrives hidden and stays hidden until you enable it.

Select the connection, find the tables, and use Enable on each one, or Show All. Choose which tables to expose explains why the default is what it is.

This is the single most common reason a healthy connection appears to do nothing.

A view shows as DRAFT on the module card but LIVE in the workspace

Both describe the same view. The Views tab shows saved views as LIVE with a version number, while the summary card on the service’s Data Hub module page labels the same views DRAFT.

The workspace is the one to trust. The card’s label is a display inconsistency and has been reported.

Auto-Detect found nothing

Detection needs evidence — a declared foreign key, or consistent naming it can infer from. A database with neither will produce no suggestions, which is honest rather than broken.

It also only looks at exposed tables. A relationship between an enabled table and a hidden one will not be found; enable both and run it again.

You can always record a relationship by hand on the canvas.

My relationship changes disappeared

They were not committed. Detected relationships and card positions are pending until you use Commit, which shows a count of what is waiting.

Leaving the tab with changes pending prompts you to save or discard. Choosing discard is the usual way people lose them.

The data is out of date

Check the cache settings for that dataset. The explorer states the current behaviour in plain language above the grid.

Refresh re-reads immediately, ignoring the cache. If people need that regularly, turn on the manual refresh button for the dataset rather than explaining the workaround each time.

For Airtable, Coda, and Notion the answer is different: those are synced into landed tables rather than queried live, so the data is as fresh as the last sync, and no amount of cache tuning changes that.

I cannot create a connection

Creating a data source needs the platform admin role. Service membership is not enough.

If the database is already connected somewhere on the platform, you do not need to create anything — check Link Existing in the wizard and link it instead. That path is open to service members.

The Writeback Lab says it is not supported

Not every connector allows writeback. Those that do not declare it rather than failing when you try, which is why the badge is there.

Read-only sources are the usual case. There is no way to enable writeback for a connector that does not support it.

A file upload has gone stale

An uploaded CSV or Excel file is a snapshot taken when you uploaded it. Nothing re-reads the original file, because there is no connection to re-read.

Upload the new version. For anything that changes regularly, connect the underlying database instead.