Zero-ETL and ELT describe different things. ELT is an ordering: load raw, transform in the warehouse. Zero-ETL is an operating model in which the cloud provider runs extraction and loading. It produces an ELT architecture, since transformation happens only after the data lands.
Back to glossary
Zero-ETL
Zero-ETL is a managed replication path between an operational data store and an analytics store, configured rather than built.
What is Zero-ETL?
AWS introduced the term at re:Invent in November 2022, announcing Aurora zero-ETL integration with Amazon Redshift in preview. The motivation it gave was the cost of hand-built pipelines: one per source cluster, each written, tested, and operated by the data team.
The term describes who operates the pipeline rather than the absence of one. Extraction and loading still happen, run by the provider; transformation moves into the destination. Microsoft ships the same pattern as Mirroring in Microsoft Fabric. Google applies the label to BigQuery federated queries over Bigtable, where nothing is copied at all.
How a Zero-ETL integration works
A zero-ETL integration is log-based change data capture with the operating burden moved to the provider. Aurora MySQL and RDS for MySQL read the binary log (binlog), MySQL's ordered record of committed row changes. Aurora PostgreSQL uses a logical replication slot, a server-side marker that retains write-ahead log (WAL) until the consumer confirms it has read past it.
The sequence is the same across implementations:
- Initial snapshot. A consistent read loads the selected tables and records the log position where the snapshot ends.
- Continuous apply. The service tails the log from that position and applies inserts, updates, and deletes. Source tables require a primary key; tables without one never synchronize.
- Managed storage. Data lands read-only in Redshift managed storage or a SageMaker Lakehouse catalog, with a REFRESH_INTERVAL setting controlling how often changes are applied.
Where Zero-ETL fits and where it runs out
Zero-ETL works inside one vendor's boundary. AWS zero-ETL integrations cover Aurora MySQL and PostgreSQL, RDS for MySQL, DynamoDB, DocumentDB, and SaaS sources such as Salesforce, landing in Redshift, SageMaker Lakehouse, or OpenSearch. Anything outside that list is out of scope.
The constraints that decide viability sit at the table level. AWS documents that replication applies no transformations; rows arrive as they exist in the source, which is why ELT remains the dominant warehouse pattern. Filters apply at database and table level. Column and row filtering are unavailable, so restricted columns reach the target unless excluded at the source. Unsupported types break individual tables. Aurora PostgreSQL rejects custom types and types created by extensions; Aurora MySQL supports InnoDB only. Adding a column anywhere but the end, or combining column operations in one statement, can force a reload that leaves the table unqueryable in the target.
Zero-ETL compared with managed CDC replication
| Dimension | Zero-ETL | Managed CDC replication |
|---|---|---|
| Coverage | Managed engines and analytics services inside one provider | Managed or self-hosted Postgres, MySQL, SQL Server, MongoDB into Snowflake, BigQuery, Databricks, Redshift |
| Control | Table-level filters only; no in-flight transformation; read-only target | Column selection, typing rules, soft-delete handling, per-table backfill |
| Latency | Near real time for database sources; a one-hour minimum for SaaS sources on AWS | Set by flush interval and destination write pattern |
Zero-ETL is the shorter path when source and destination sit inside one provider and the warehouse can absorb the raw schema. A managed CDC tool earns its line item when sources span clouds or engines, or the destination is Snowflake, BigQuery, or Databricks. Artie sits in the second category, reading the Postgres WAL, MySQL binlog, and MongoDB oplog and writing to those warehouses with per-table control over backfills and schema handling. It adds a vendor and a bill that a native integration does not, the trade to weigh when comparing CDC tools for a cloud data warehouse.