Data Tool Guides

Databases

PostgreSQL ETL Tools: 10 Best for Postgres CDC in 2026

Compare the best PostgreSQL ETL tools for Postgres CDC, with notes on logical replication, replication slots, WAL retention and schema drift.

Team Airbyte

March 31, 2026

Summarize with AI:

TL;DR

Short answer: Postgres CDC is what separates these tools. Logical replication reads the write-ahead log through a replication slot, so inserts, updates and deletes all arrive without repeatedly scanning tables. Tools without it poll a timestamp column, which is simpler to set up and silently misses deletes. Ten PostgreSQL ETL tools, log-based options first:

Airbyte: open-source Postgres source with CDC through pgoutput, plus xmin and cursor modes, into any of 700+ connectors' destinations.
Fivetran: managed Postgres CDC with no upkeep, billed on monthly active rows.
Debezium: the open-source CDC engine many tools build on, streaming WAL changes as events through Kafka Connect or Debezium Server.
Stitch: 140+ connectors and the simplest setup, with log-based replication through a replication slot and the wal2json plugin. Now consolidating into Qlik Talend Cloud.
Hevo Data: no-code Postgres logical replication, cloud-only and billed on events.
Talend: integration with data quality attached, CDC through Qlik Replicate. No free tier since Open Studio retired in 2024.
Informatica PowerCenter: mature enterprise ETL, though 10.5 left standard support in March 2026 and Salesforce now owns Informatica.
Pentaho Data Integration: visual ETL over JDBC, suited to scheduled batch loads rather than replication.
Meltano: open-source and CLI-first, with a Singer Postgres tap that supports log-based replication.
Apache Airflow: an orchestrator. Its Postgres provider runs queries, but you write the extraction yourself.
One Postgres constraint: a replication slot holds back WAL until its consumer reads it. If a pipeline stops, WAL accumulates on the primary until the disk fills. Monitor slot lag from day one, and on Postgres 13 or later set max_slot_wal_keep_size, accepting that a capped slot will need a resync.

PostgreSQL ETL tools move data out of Postgres, whether self-managed or on Amazon RDS, Aurora, Cloud SQL or Azure, into a warehouse or lake. The decision that matters most is how a tool reads changes, because that determines whether deletes reach your warehouse and how much load each sync puts on a production database.

How do tools read changes from Postgres?

There are three ways, in increasing order of completeness:

Cursor column: the tool selects rows whose updated_at or ID is greater than the last run. Simple and needs no database changes, but deleted rows are never seen and rows updated without touching the cursor are missed.
xmin: uses Postgres's internal transaction ID on each row to find changed rows without a cursor column. It catches updates a timestamp would miss but still cannot see deletes.
Logical replication (CDC): reads every insert, update and delete from the write-ahead log through a replication slot and a publication, usually with the built-in pgoutput plugin. Complete and light on the database, at the cost of some setup and slot monitoring.

How do the PostgreSQL ETL tools compare?

Tool	How it reads Postgres	Deletes captured	Hosting	Pricing basis
Airbyte	CDC via pgoutput, xmin or cursor	Yes, with CDC	Cloud or self-hosted	Free self-hosted; Cloud volume-based; Pro and Flex capacity-based
Fivetran	Logical replication, xmin or alternatives	Yes, with logical replication	Cloud, hybrid	Monthly active rows
Debezium	Logical replication, streamed as events	Yes	Kafka Connect or Debezium Server	Free, Apache 2.0
Stitch	wal2json logical replication, or key-based	Yes, log-based	Cloud	Rows per month
Hevo Data	Logical replication	Yes	Cloud	Events
Talend	Queries; CDC via Qlik Replicate	With Qlik Replicate	Cloud or on-premises	Quoted subscription
Informatica PowerCenter	PowerExchange CDC	Yes	On-premises	Quoted licence
Pentaho Data Integration	JDBC queries	No	Self-hosted	Community and enterprise editions
Meltano	Singer tap, log-based via wal2json or key-based	With log-based	Self-hosted	Free, open source
Apache Airflow	Queries you write	Only if you build it	Self-hosted or managed	Free, you run it

Which PostgreSQL ETL tools should you consider?

The first six can read Postgres's write-ahead log; the last four are batch, transformation or orchestration tools that suit Postgres when replication is not the main job.

1. Airbyte

Airbyte is an open-source data integration platform with 700+ connectors and a community of 25,000+ data engineers. Its Postgres source offers three update methods: CDC through logical replication with pgoutput, which captures deletes; xmin, for databases where replication cannot be enabled; and a standard cursor column. It loads into any Airbyte destination and works with dbt for transformation.

What's unique about Airbyte?

You get log-based CDC without running Kafka, and Postgres moves through the same pipeline as your SaaS and other database sources. Syncs checkpoint so a failed run resumes rather than restarts. It runs as self-hosted open source, on Airbyte Cloud, or as Enterprise Flex, which keeps the data plane in your own environment when Postgres holds regulated records.

Pros	Cons
Log-based CDC without running Kafka yourself	Transformation relies on dbt rather than being built in
700+ connectors, so Postgres joins the rest of your stack	Self-managed deployment needs Kubernetes or Docker skills
Checkpoint-based recovery resumes after a failed sync	Not an orchestrator, so pair it with Airflow or Dagster
Enterprise Flex keeps data and keys in your own environment	Flex pricing is custom and needs a sales conversation

2. Fivetran

Fivetran is a closed-source, managed ELT service with 700+ connectors. Its Postgres connector uses logical replication for CDC, with xmin and other options where replication is unavailable, and applies schema changes automatically. It bills on monthly active rows, so high-churn tables can drive cost up.

Here is a closer look at Airbyte vs Fivetran.

Pros	Cons
Automatic schema drift handling on Postgres tables	Monthly active row pricing is hard to forecast
Several update methods for databases without replication	Closed source, so connectors cannot be edited
SOC 2 Type II compliance	Data is processed on vendor infrastructure unless you use hybrid deployment
Very low ongoing maintenance	High-churn Postgres tables inflate the bill

3. Debezium

Debezium is the reference open-source platform for change data capture, and it is the engine behind Postgres CDC in several of the tools here. It reads the write-ahead log through logical decoding with pgoutput, so inserts, updates and deletes stream out in commit order. It runs on Kafka Connect, or without Kafka through Debezium Server, which publishes to other systems such as Kinesis, Pub/Sub or Redis. Red Hat offers a commercially supported build.

Pros	Cons
The most mature Postgres CDC implementation available	You build the load into your warehouse yourself
Reads the write-ahead log, so no polling or added load	Databases only, so SaaS sources need another tool
Free and open source, with commercial support from Red Hat	Kafka Connect is significant infrastructure if you use it
Runs on Kafka Connect or standalone with Debezium Server	You own hosting, monitoring and upgrades

4. Stitch Data

Stitch is a cloud extract-and-load platform with 140+ connectors, owned by Qlik. Its Postgres integration supports log-based replication through a replication slot and the wal2json plugin, as well as key-based incremental replication, and it can hash sensitive columns on the way through. It has no user-defined transformations, and Qlik now publishes a migration path from Stitch to Qlik Talend Cloud.

Here is a closer look at Airbyte vs Stitch.

Pros	Cons
Configurable replication methods per table	140+ connectors, narrower than the leaders
Column hashing for sensitive Postgres fields	No user-defined transformations
Among the lowest-cost managed options	Cloud only, with no hybrid deployment
Log-based replication through wal2json	Being consolidated into Qlik Talend Cloud

5. Hevo Data

Hevo Data is a no-code cloud ELT platform with 150+ connectors and Postgres logical replication, with transformation applied before data reaches the destination. It is cloud-only and billed on events, which suits teams without engineering capacity who do not need self-hosting.

Pros	Cons
Log-based Postgres CDC with a no-code interface	Around 150 integrations, well short of the leaders
Transformations and enrichment before data lands	Event-based pricing is hard to forecast at scale
Data activation syncs results back to source APIs	Cloud only, so no self-hosted or hybrid option
Automated error recovery on failed syncs	Connector gaps cannot be filled by your own team

6. Talend

Talend pairs integration with data quality and governance, reading Postgres through dedicated components, with log-based CDC available through Qlik Replicate. Two things to check before shortlisting it: Qlik acquired Talend in 2023 and now sells it as Qlik Talend Cloud, and Talend Open Studio, the free open-source edition, was retired on 31 January 2024, so there is no free tier or self-serve route in.

Pros	Cons
Integration, quality and governance in one platform	Implementation is involved and rarely self-serve
Log-based CDC through Qlik Replicate	No free tier since Open Studio was retired in 2024
Very broad connector catalogue	Steep learning curve and long onboarding
Strong governance for regulated Postgres estates	Roadmap now tied to Qlik after acquisition

7. Informatica PowerCenter

Informatica PowerCenter is long-established enterprise ETL with profiling, cleansing and governance, and PowerExchange CDC for Postgres. Two things to weigh: Salesforce completed its acquisition of Informatica in November 2025, and PowerCenter 10.5 left standard support in March 2026, so new investment is going into Informatica's cloud platform rather than PowerCenter.

Pros	Cons
Data profiling alongside cleansing and transformation	Enterprise product with no self-serve entry point
Proven at very large enterprise scale	PowerCenter 10.5 out of standard support since March 2026
Strong governance, lineage and metadata support	High licence cost and long implementation
Handles complex legacy sources alongside Postgres	Overpowered for a single Postgres pipeline

8. Pentaho Data Integration

Pentaho Data Integration, long known as Kettle, is visual ETL that reads Postgres over JDBC with a large library of transformation steps. It has no change capture of its own, so it suits scheduled batch loads and heavy reshaping in transit rather than replication. Check the licensing of the edition you need, as the community and enterprise editions differ.

Pros	Cons
Large library of visual transformation steps	No log-based change capture, so deletes are missed
Combines ETL with reporting in the wider Pentaho suite	Transforms before loading rather than in the warehouse
Broad source and destination coverage over JDBC	Interface feels dated against modern platforms
Runs on your own infrastructure	Edition licensing needs checking before you commit

9. Meltano

Meltano is an open-source, CLI-first platform spun out of GitLab that manages Singer taps and targets as a version-controlled project. Its Postgres tap supports log-based replication through wal2json as well as key-based incremental sync. Maintenance depends on the tap's maintainers, since the wider Singer community is now largely inactive.

Pros	Cons
Free, open source and Git-native	CLI-first, so no interface for non-technical users
Strong DataOps and CI/CD orientation	Connector maintenance is left to the Singer community
SDK available for building connectors	No support package with an SLA
Works well alongside dbt and Airflow	wal2json must be installed on the database server

10. Apache Airflow

Apache Airflow is an open-source orchestrator, not an ETL tool. Its Postgres provider lets you run queries and move results, but you write the incremental logic and loading yourself, and there is no change capture. Its real role is scheduling a connector-based tool such as Airbyte, or dbt models on top of replicated Postgres data.

Here is a closer look at Airbyte vs Airflow.

Pros	Cons
Free, open source and very widely adopted	Not an ETL tool, so extraction logic is yours to build
Postgres provider for running queries and transfers	No change capture, so deletes are missed unless you build it
Handles dependencies, retries and scheduling well	Production operation takes real engineering effort
Pairs well with a connector-based tool for movement	Steep learning curve for teams new to DAGs

How do you set up Postgres CDC?

Whichever log-based tool you choose, the database needs the same preparation:

Enable logical decoding. Set wal_level to logical and restart. On Amazon RDS and Aurora that means setting rds.logical_replication to 1 in the parameter group; on Cloud SQL, the cloudsql.logical_decoding flag; on Azure Database for PostgreSQL, the wal_level server parameter.
Create a replication user. Grant it the REPLICATION attribute (rds_replication on RDS) plus SELECT on the tables you want.
Create a publication and a replication slot. The publication lists the tables to replicate; the slot, usually with the pgoutput plugin, tracks how far the consumer has read. Some tools create these for you.
Check replica identity. Tables need a primary key for updates and deletes to replicate cleanly. For tables without one, set REPLICA IDENTITY FULL, at the cost of larger WAL.
Monitor the slot. Alert on replication slot lag. An abandoned slot keeps WAL indefinitely, so drop slots you no longer use.

How do you start syncing Postgres data with Airbyte?

Step 1: Set up Postgres as a source

Prepare the database as above if you want CDC. In Airbyte, open Sources, choose New source and select Postgres, then enter the host, port (5432 by default), database, schemas and credentials, and choose an update method: read changes using write-ahead log (CDC) with your replication slot and publication names, detect changes with the xmin system column, or scan changes with a user-defined cursor. Set SSL mode or an SSH tunnel as needed and run the connection test.

Step 2: Set up a destination

Choose where the data lands: a cloud warehouse such as Snowflake, BigQuery or Redshift, a data lake, a database or cloud storage.

Step 3: Configure the connection

Select the tables to replicate, use incremental sync with deduplication so the destination mirrors the source, and set a frequency frequent enough to keep slot lag low.

The process is the same in Airbyte Open Source, which you can deploy yourself, and in Airbyte Cloud, which you can try free for 14 days.

Which PostgreSQL ETL tool should you choose?

If deletes matter, choose a log-based tool and prepare the database once. Airbyte gives you CDC into any warehouse alongside your other sources, self-hosted or in Cloud; Fivetran does the same with no upkeep if row-based pricing works for you; Debezium suits teams who want real-time events and will run the infrastructure. Stitch and Hevo are the simpler managed options. If you cannot enable logical replication, xmin is the next best method. Keep Talend, Informatica or Pentaho where they already run, and use Airflow to orchestrate rather than extract.

For tools that cover more than Postgres, see our guide to the top data integration tools.

Replicate Postgres with Airbyte

Move Postgres tables into Snowflake, BigQuery, Databricks or any Airbyte destination with log-based CDC, self-hosted or in Airbyte Cloud.

Try Airbyte Cloud free for 14 days or talk to our team.

Frequently Asked Questions (FAQs)

1. What is Postgres CDC?

Change data capture in PostgreSQL means reading inserts, updates and deletes from the write-ahead log through logical replication, rather than querying tables for changed rows. It captures deletes and puts little load on the database, which is why most ETL tools prefer it when it is available.

2. Why is my Postgres disk filling up after setting up CDC?

Almost always an inactive or lagging replication slot. Postgres keeps WAL until every slot has consumed it, so a stopped pipeline makes WAL accumulate. Restart the consumer or drop the unused slot, and on Postgres 13 or later consider max_slot_wal_keep_size to cap how much WAL a slot can hold.

3. Can you use CDC on Amazon RDS or Aurora PostgreSQL?

Yes. Set rds.logical_replication to 1 in the parameter group and reboot, then grant the rds_replication role to the replication user. Cloud SQL and Azure Database for PostgreSQL have equivalent settings.

4. What if I cannot enable logical replication?

Use xmin-based or cursor-based incremental sync, which Airbyte and Fivetran both support. Neither captures deletes, so use soft deletes or periodic full refreshes for tables where removed rows matter.

About the Author

Team Airbyte

Engineering Team of Airbyte.

PostgreSQL ETL Tools: 10 Best for Postgres CDC in 2026

About the Author

Related posts