Question 1

Do MongoDB documents become real nested columns?

Accepted Answer

Sometimes, and the distinction matters more here than on other warehouses. When the shape is typed, embedded documents become STRUCT and arrays become ARRAY, which is the outcome you want and lets you query with dot notation or explode a list directly. When the shape is generic JSON, an untyped collection or an empty object, it falls back to a STRING containing JSON that you parse at query time. Given how varied Mongo documents are, sample a collection before assuming which you will get.

Question 2

What kind of MongoDB deployment does this need?

Accepted Answer

One that supports change streams, so a replica set or a sharded cluster rather than a standalone server. Polytomic opens a cluster level change stream filtered to your namespaces and operations, using resume tokens so a run continues from where it stopped, which also means the oplog has to be retained long enough to resume after a pause. The connecting user needs to list databases and collections, read the selected collections, and open that cluster level stream.

Question 3

Why did a field not appear in my table?

Accepted Answer

Almost certainly because it was not in the sample. Schema inference reads the hundred most recent documents in a collection with recursion off, so a field that only shows up in older documents, or appears rarely, may not make the initial schema at all. You can add it afterwards, and anything discovered later is added with an ALTER TABLE rather than needing a rebuild. Worth testing against a collection you know is irregular before trusting the output.

Question 4

What happens if a sync pauses for too long?

Accepted Answer

If the pause outlasts your oplog retention, the change stream reports lost history and that collection needs a resync, so size retention against the worst outage you can imagine rather than a typical one. On deletes: a change stream delete carries only the document key, so a key only tombstone is emitted and the row marked, or removed outright with hard deletes enabled. Neither fallback path, a full scan or a tracking field query, captures deletes at all.

Question 5

Why move document data into a lakehouse?

Accepted Answer

Because a document store is awkward for analysis and competes with your application when you try. Aggregating across collections is difficult, and the joins you want reach into billing or CRM data Mongo never had. ETL into a data warehouse handles the semi structured shape natively where it can, keeps JSON where it cannot, and puts the whole thing where SQL and notebooks reach the same tables.

Sync MongoDB to Databricks

Sync MongoDB to Databricks with Polytomic

Typed shapes stay native

Deletes arrive as tombstones

Schema from a sample

Or sync Databricks to MongoDB

Trusted by businesses worldwide

Step-by-step integration guide

Connect MongoDB in Polytomic

Connect Databricks in Polytomic

Create a sync between MongoDB and Databricks

Map fields and enable the sync

Frequently asked questions

The sync and ETL platform that does it all

Powerful data sync

Advanced analytics

Enterprise security

Easy setup

Ready to connect MongoDB with Databricks?