PostgreSQL Change Data Capture PlatformsNov 14, 2026By Toolverly Editorial

PostgreSQL Change Data Capture Platforms in 2026: AWS DMS vs Google Cloud Datastream vs Fivetran vs Estuary vs Debezium

Compare five PostgreSQL CDC options by how they backfill existing rows, represent deletes and recover when a downstream consumer falls behind.

Source code displayed on a computer monitor.

Start with Datastream for a supported BigQuery target, or AWS DMS when your team already operates an AWS migration or replication estate. Shortlist Fivetran for managed warehouse tables and Estuary when you want captured changes separated from downstream materializations. Choose Debezium when the event stream itself is the product and your team can run the rest. In every case, settle what a delete means before trusting the first dashboard.

Define the record your destination needs

Compare data integration platforms first if your real buying problem is onboarding many unrelated data sources. A PostgreSQL specialist decision begins when snapshot recovery and delete semantics can change a report.

Use data observability platforms to detect a stale or inconsistent downstream table. Separately, assign a person to inspect source replication lag and decide whether recovery requires a backfill.

Product First load and continuing changes Destination responsibility Main unit to cost
AWS DMS Full load with continuing CDC Validate source and target separately Replication instances or capacity, storage and transfer
Google Cloud Datastream Managed backfill and CDC BigQuery merge or append-only write mode Backfill and change volume; destination charges
Fivetran Managed incremental PostgreSQL sync Warehouse tables; inspect deletion flags Monthly active rows, connection terms and plan
Estuary PostgreSQL capture with separately managed materialization Collection can feed distinct downstream targets Data moved and active connector instances
Debezium Snapshot plus PostgreSQL change events Operate the stream and selected sink Infrastructure, retention and support labor

AWS DMS

AWS DMS can use logical replication for ongoing changes from a supported PostgreSQL source. It documents source slot requirements and explains that WAL is retained while the slot still needs it. A full load plus CDC is useful when transferring an initial state and maintaining it during a migration, but this does not make every destination an audited change log. Match source version and target type, document source failover behavior, and price provisioned or serverless capacity, plus applicable storage and transfer. Official references include Product and scope, Plan or supporting details.

Google Cloud Datastream

Datastream captures changes from PostgreSQL and documents both merge and append-only modes for BigQuery. Choose merge when readers need current state and the table is eligible for the intended semantics; choose append-only if you want changes retained and have logic to read them correctly. Check source replica identity and BigQuery limitations before assuming that one setting fits every table. Use the backfill and ongoing data processing categories when estimating spend, alongside the target's own storage and queries. Official references include Product and scope, Plan or supporting details, Additional product details.

Fivetran

Fivetran documents logical replication for PostgreSQL and alternatives with separate boundaries. For warehouse reporting, check the connector's soft-delete behavior and whether your model excludes marked rows. Its monthly active row approach generally counts a row once after it becomes active in a month, rather than counting every successive update as a new row. That cost profile differs from moving change bytes for a high-churn table, so estimate from distinct active keys and verify the current connection and plan terms. Official references include Product and scope, Plan or supporting details.

Estuary

Estuary's PostgreSQL capture documentation exposes backfill controls and notes that unchanged large values stored out of line may be absent from some WAL change records. Check the target materialization's behavior before interpreting every event as a complete row. The Cloud pay-as-you-go offer lists $0.50 per GB and $100 per month for each connector instance through the first six, with instances prorated by time and a lower listed rate thereafter. Count both capture and each destination you operate rather than treating the source connector as the whole pipeline. Official references include Product and scope, Plan or supporting details.

Debezium

Debezium documents PostgreSQL snapshot modes, change events and delete/tombstone behavior. It makes sense when the event record is an interface for several consumers and a team can own connectors, broker retention and sink semantics. Do not assume a downstream consumer will interpret tombstones or replay idempotently. Pay particular attention to replication slots and publications when more than one connector reads the source, and include the operations work when comparing it with managed services. Official references include Product and scope.

Choose who will own recovery

Choose the destination's delete contract before shopping for latency. A customer who disappears from a current-state report may still need to appear in audited event history. Decide whether reports should filter a marked row, see a removed row or reconstruct state from a change log. Include a high-churn table and one downstream failure in evaluation. Record who watches replication lag, what happens when the retained PostgreSQL write-ahead log grows, and who authorizes a new backfill if the original position is lost. A shorter advertised sync interval does not answer any of those questions.

By Toolverly Editorial, with AI-assisted research and drafting.

Photo by Rob Wingate on Unsplash. Read about our publication or send a correction.