Data engineering terms, defined without a sales pitch.

Why a plain glossary, not a feature glossary

A lot of vendor glossaries define a term in a way that happens to describe their own product perfectly, which is backwards. CDC does not mean "real-time streaming with sub-second latency" just because some CDC vendors sell that. It means tracking row-level changes, full stop, and there are batch and log-based ways to do it. These pages define the term first, honestly, and only then say what QueryFlow does with it.

Some of these terms also get used loosely by vendors to make a marketing claim sound more precise than it is. Fivetran markets "schema drift handling" without ever defining schema drift on its own terms, and "idempotent delivery" shows up in pricing pages with no explanation of what idempotent actually means for a load. Untangling the plain definition from the marketing claim is most of the value of a page like this.

The eighteen terms

• Change Data Capture (CDC) — CDC is a way of tracking row-level changes, not a specific tool.
• ELT vs. ETL — The difference is where the transform happens, not which tool you use.
• Data Sync vs. ETL — Data Sync is a smaller, more specific job than ETL.
• Reverse ETL — Reverse ETL moves data out of the warehouse, not into it.
• Schema Drift — Schema drift is when a source's structure changes under you.
• Idempotent Loads — An idempotent load gives the same result no matter how many times it runs.
• Insert vs. Update vs. Upsert — Three different write behaviors, and picking the wrong one duplicates or loses rows.
• Monthly Active Rows (MAR) — MAR is the unit Fivetran bills you on.
• Fivetran Pricing Explained — How Fivetran's MAR-based pricing actually works, with a dated source.
• Job Scheduler (ETL) — What a job scheduler actually does versus a plain cron job.
• The .qfpipe File — What's inside QueryFlow's own pipeline definition file.
• AI SQL Assistant — What an AI SQL assistant actually does, and where it can be wrong.
• Service Account vs. OAuth — Two different ways to authenticate a connection, and when each fits.
• BigQuery Jobs API — How BigQuery actually runs your query behind the scenes.
• Databricks Catalogs and Schemas — How Unity Catalog's three-level namespace actually works.
• Snowflake Alert Object — What Snowflake's native ALERT object does and doesn't cover.
• Connection Diagnostics — How to read a connection error like an engineer, not by guessing.
• Output Destinations — Every place QueryFlow can send a query's results.

Where to go next

If you're trying to actually move data rather than define a term, the integrations hub lists every source-to-destination how-to, and the free tools page has a CSV-to-CREATE-TABLE generator and a MERGE-key checker that turn a couple of these definitions into something you can run.