Does Google have a native PostgreSQL to BigQuery connector?

Yes, two. BigQuery Data Transfer Service has a PostgreSQL connector for scheduled transfers, with incremental transfers in Preview that cannot sync deletes. Datastream replicates changes using PostgreSQL logical decoding and is priced per GiB processed. For Cloud SQL, BigQuery federated queries can also read the database without copying it.

Does Weld support PostgreSQL CDC into BigQuery?

Yes. Weld reads inserts, updates and deletes from the write-ahead log through a logical replication slot and a publication you create, and applies them to BigQuery. Each table needs a primary key or unique index, and wal_level must be set to logical.

Are PostgreSQL syncs to BigQuery incremental?

Yes. After the first full load, Weld syncs new and changed rows using PostgreSQL's xmin system column, so your tables do not need an updated_at column. You can also choose CDC or a full sync at midnight per table.

How are deleted rows handled?

Incremental syncs using xmin cannot detect deletes. A full sync at midnight removes deleted rows from the destination once a day, and CDC propagates deletes as they happen.

Can Weld keep a history of changes to PostgreSQL rows?

Not for this connector. History tables are not available for PostgreSQL sources in Weld, so each BigQuery table holds the current state of the source table.

Will syncing slow down my production database?

Scheduled syncs run queries against the database, so Weld recommends a read replica for busy systems. CDC reads changes from the WAL instead of scanning tables.

PostgreSQL to BigQuery

Copy your PostgreSQL tables into BigQuery and keep them current, either with incremental syncs on a schedule or with CDC that reads the write-ahead log and carries deletes. Weld connects directly or through an SSH tunnel, and lets you drop or hash columns before anything lands.

Start for free View setup guide

4.8 on G2

Source

Weld

Warehouse

Checked against Weld and vendor docs on 10 October 2026

Weld · New sync

SourcePostgreSQLConnected

Tables to sync

Your own tables, picked when you set up the sync.

Extract
Load
Keep in sync

DestinationGoogle BigQueryReady

postgresql dataset

postgresql.Synced
postgresql.Syncing
postgresql.Synced
postgresql.Synced

Trusted by data and AI teams

Options

Four ways to get PostgreSQL data into BigQuery

Google has three options of its own: a scheduled transfer, a CDC service and federated queries for Cloud SQL. They fit well if your database already lives in Google Cloud. The questions that decide it are deletes, freshness and whether Postgres sits next to your other sources.

BigQuery Data Transfer Service

Best for: Scheduled batch copies inside Google Cloud
Effort: Low to medium: a transfer config, plus a network attachment if the database has no public IP
Freshness: Hourly, daily or custom schedules, or on demand
Good to know: Full transfers reload every run. Incremental transfers are in Preview, need a TIMESTAMP watermark column and cannot sync deletes. Billed by slot-hours as a GA connector.

Datastream

Best for: Log-based CDC from Postgres into BigQuery
Effort: Medium: logical decoding, a replication slot, a publication and a stream
Freshness: Continuous, merged into BigQuery on a staleness setting you choose
Good to know: Priced per GiB processed, and Google notes processed bytes can be 2 to 5 times the raw data. The first 500 GB of backfill each month is free. Streams are limited to 10,000 tables.

Federated queries (Cloud SQL only)

Best for: Occasional joins against a Cloud SQL database
Effort: Low: a BigQuery connection and EXTERNAL_QUERY
Freshness: Live at query time
Good to know: Queries Cloud SQL for PostgreSQL without copying data, so every query runs against your database. Not available for Postgres hosted outside Cloud SQL.

Weld

Best for: Postgres alongside your other sources in one pipeline
Effort: Low: connect, pick tables, choose a sync mode per table
Freshness: CDC in near real time, or incremental and full syncs on your schedule
Good to know: Incremental syncs use the xmin system column, so tables need no updated_at column. CDC reads the WAL and propagates deletes. Columns can be removed or hashed per table.

How it works

How to sync PostgreSQL to Google BigQuery with Weld.
In the app, with the API or an agent.

Setup guide API guide MCP docs

Weld · Data warehouse

Data warehouseGoogle BigQueryConnected

Data locationChosen when you connect

CredentialsService account or key, stored encrypted

Data

What lands in Google BigQuery
From PostgreSQL to Google BigQuery.

Weld does not impose a schema on a database source: what lands in BigQuery is the tables you select, with the columns you keep.

Your selected tables
One BigQuery table per PostgreSQL table you pick, under the destination name you set. Removed columns are left out and hashed columns arrive hashed.
_weld_synced
A timestamp Weld adds to every row with the time it was last synced. Use it to check freshness without touching your own columns.
Tables in your publication
For CDC, only tables in the publication you name can stream changes, and each needs a primary key or unique index and a replica identity other than NONE.

Example query

How fresh is each replicated table?
In Google BigQuery SQL.

Runs in BigQuery against a dataset named postgres. Replace orders and customers with your own tables; _weld_synced is the column Weld adds to every row.

Google BigQuery · query

1select2  'orders' as table_name,3  count(*) as row_count,4  max(_weld_synced) as last_synced_at,5  timestamp_diff(current_timestamp(), max(_weld_synced), minute) as minutes_since_sync6from postgres.orders7union all8select9  'customers',10  count(*),11  max(_weld_synced),12  timestamp_diff(current_timestamp(), max(_weld_synced), minute)13from postgres.customers14order by minutes_since_sync desc

Weld MCP server

Ask your agent.
It queries Google BigQuery for you.

Connect Claude, Cursor, ChatGPT or your own agent to Weld over MCP. It reads the PostgreSQL context, finds the tables it needs and runs the query in Google BigQuery.

Set up the MCP server

Claude CodeOpus 5

ConnectedWeld MCP

Which PostgreSQL tables do we have in Google BigQuery, and which is the largest?

Looking up what Weld has synced from PostgreSQL.

●weld·search_available_references

⎿raw.postgresql.*

tables, columns and freshness

●weld·get_reference_schema

⎿columns, types and nullability

●weld·run_query

⎿select count(*) from {{raw.postgresql.…}}

submitted

●weld·get_query_status

⎿DONE

●weld·get_query_result

⎿rows returned

✓Answered from PostgreSQL data in Google BigQuery

>Ask about your PostgreSQL dataMCP

Under the hood

Things to know.
Before you sync PostgreSQL.

Incremental syncs do not see deletes: The xmin method finds new and changed rows but cannot tell which rows were deleted. Set tables with frequent deletes to a full sync at midnight, which removes them, or use CDC, which reads deletes from the WAL.
xmin changes on any column update: xmin is updated per row, so a change to any column resyncs the row, even a column you excluded. The Weld user also needs permission to read xmin on each table.
CDC has database prerequisites: CDC needs wal_level set to logical, a non-temporary logical replication slot, a publication, a primary key or unique index on every table, and a replica identity other than NONE. On PostgreSQL 13 or below, add a small heartbeat table so idle periods do not build up WAL.
Unused slots hold on to WAL: A replication slot left behind after you stop a sync makes PostgreSQL keep WAL indefinitely, and WAL is shared by every database on the server. Drop slots you no longer use and cap retention with max_slot_wal_keep_size.
Read replicas can cancel long queries: On a replica, sync queries can fail with a conflict with recovery error when the primary has moved on. Raising max_standby_streaming_delay fixes it; Weld's docs suggest 5 to 30 minutes depending on size and load.

Sources: Weld PostgreSQL connector docs · Weld PostgreSQL CDC docs · Weld BigQuery setup guide · Weld managed fields · Google Cloud: Introduction to PostgreSQL data transfers · Google Cloud: Load PostgreSQL data into BigQuery · Google Cloud: BigQuery pricing (Data Transfer Service) · Google Cloud: Datastream PostgreSQL source · Google Cloud: Datastream BigQuery destination · Google Cloud: Datastream pricing · Google Cloud: Cloud SQL federated queries