Real-time replication, straight from the transaction log
Change data capture streams every insert, update and delete from Postgres, MySQL, SQL Server and MongoDB into your warehouse, in seconds to minutes rather than hours.
- Captures deletes
- No table scans
- SOC 2 Type II
PostgresWAL →
SnowflakeStreaming- INSERTordersid 10284 · amount 149.00Applied
- UPDATEusersid 112 · verified → trueApplied
- DELETEcartsid 821 · removed at sourceApplied
- INSERTeventsid 55410 · type clickApplied
- UPDATEinvoicesid 3301 · sent → trueApplied
Trusted by data and AI teams
Every change, as it happens. Without querying production.
Change data capture is a replication mode in Weld's data pipelines, for databases where freshness and deletes matter.
Every change, including deletes
Inserts, updates and hard deletes all reach the warehouse, not just the new rows an incremental sync can see.
DELETEsessions · id 99012 appliedSeconds to minutes
Changes stream continuously instead of waiting for the next hourly or daily batch.
lagseconds to minutesLight on your database
Weld reads the database's own change log, so there are no repeated table scans against production.
readsWAL · no table scansIn the order they happened
Changes are applied in log order, so the warehouse ends up in the same state as the source.
appliedin log orderCDC, incremental or full. Pick per table.
Weld runs all three. CDC is the one that sees deletes and keeps up in near real time.
| Mode | How it works | Captures deletes | Low latency | Low source load | Best fit |
|---|---|---|---|---|---|
| Full table sync | Re-copies entire tables on a schedule | — | — | — | Small datasets |
| Incremental sync | Pulls rows by timestamp or ID | Moderate freshness needs | |||
| Change data capture | Streams changes from the database log | Real-time systems |
Use CDC when
- You need changes in seconds to minutes, not hours
- Deletes and updates have to arrive, not just new rows
- Large production tables cannot take repeated scans
- Downstream systems react to events
Batch is enough when
- Tables are small or change slowly
- Hourly or daily freshness is fine
- You want the simplest possible setup
- The source has no transaction log to read
Four steps to a live stream. Each database has its own guide.
- 01
Connect your database
Add the database in Weld with a dedicated user for replication.
- 02
Turn on CDC
Switch the connector to CDC mode. Each database's guide covers the settings it needs.
- 03
Pick tables
Select the tables or collections to capture.
- 04
Changes flow in
Weld reads each change from the log and applies it to your warehouse.
Four databases. Read from their own logs.
Each one has a setup guide that walks through enabling CDC on the database side.
One warehouse. Every step of the pipeline.
Go deeper on CDC. Docs, modes and history tables.
Stream your database.
Skip the batch window.
Turn on change data capture for Postgres, MySQL, SQL Server or MongoDB and keep your warehouse current. Free for 14 days, no credit card.