Does Snowflake have a native PostgreSQL connector?

Yes. The Openflow Connector for PostgreSQL is generally available and replicates selected tables in near real time or on a schedule, for PostgreSQL 11 to 18. Each table needs a primary key or a suitable replica identity, and it runs on an Openflow deployment that you set up in Snowflake.

Does Weld support PostgreSQL CDC into Snowflake?

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 Snowflake. Each table needs a primary key or unique index, and wal_level must be set to logical.

Are PostgreSQL syncs to Snowflake incremental?

Yes. After the first full load, Weld syncs new and changed rows using PostgreSQL's xmin system column. You can also choose CDC or a full sync at midnight per table.

Why do my queries fail unless I quote table names?

Snowflake treats unquoted names as upper case, and Weld creates lower case names by default. Quote the names, or create the sync with upper case names turned on so you can query without quotes.

What happens if a full sync into Snowflake fails?

Nothing changes in the destination. Weld stages the data and loads it with a single COPY INTO statement that aborts entirely on error, so the table stays as it was.

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 Snowflake table holds the current state of the source table.

PostgreSQL to Snowflake

Replicate your PostgreSQL tables into Snowflake and keep them current, with incremental syncs on a schedule or CDC that reads the write-ahead log and carries deletes. Weld connects directly or through an SSH tunnel, and you choose per table how it syncs.

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

DestinationSnowflakeReady

postgresql schema

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

Trusted by data and AI teams

Options

Three ways to get PostgreSQL data into Snowflake

Snowflake now has its own PostgreSQL connector, built on Openflow and generally available. It suits teams that want replication managed inside Snowflake. The deciding questions are how much you want to run yourself and whether Postgres is one source among many.

Openflow Connector for PostgreSQL

Best for: CDC replication managed inside Snowflake
Effort: Medium: an Openflow deployment and runtime, plus Postgres replication settings
Freshness: Near real time or on a schedule
Good to know: Generally available for PostgreSQL 11 to 18. Tables without a primary key or suitable replica identity replicate inserts only, TRUNCATE is ignored, and only username and password authentication is supported. Billed as Snowflake usage.

Export files and COPY INTO

Best for: One-off loads or a pipeline you script yourself
Effort: High: export, stage, load, and your own change tracking
Freshness: Whatever you schedule
Good to know: COPY INTO loads staged files such as CSV, JSON or Parquet on a warehouse you size, and Snowpipe can load files as they arrive. Updates and deletes are yours to reconcile.

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. Full syncs into Snowflake are atomic.

How it works

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

Setup guide API guide MCP docs

Weld · Data warehouse

Data warehouseSnowflakeConnected

Data locationChosen when you connect

CredentialsService account or key, stored encrypted

Data

What lands in Snowflake
From PostgreSQL to Snowflake.

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

Your selected tables
One Snowflake table per PostgreSQL table you pick. Weld quotes every name it writes, so names stay lower case unless you turn on upper case for the sync.
_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 Snowflake SQL.

Runs in Snowflake against a schema named postgres in the database Weld writes to. Names are double quoted because Weld creates them in lower case by default. Replace orders and customers with your own tables.

Snowflake · query

1select2  'orders' as table_name,3  count(*) as row_count,4  max("_weld_synced") as last_synced_at,5  datediff('minute', max("_weld_synced"), current_timestamp()) as minutes_since_sync6from "postgres"."orders"7union all8select9  'customers',10  count(*),11  max("_weld_synced"),12  datediff('minute', max("_weld_synced"), current_timestamp())13from "postgres"."customers"14order by minutes_since_sync desc

Weld MCP server

Ask your agent.
It queries Snowflake 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 Snowflake.

Set up the MCP server

Claude CodeOpus 5

ConnectedWeld MCP

Which PostgreSQL tables do we have in Snowflake, 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 Snowflake

>Ask about your PostgreSQL dataMCP

Under the hood

Things to know.
Before you sync PostgreSQL.

Lower case names need quotes in Snowflake: Snowflake resolves unquoted names to upper case, and Weld quotes every name it writes, so a table created as "orders" must be queried with quotes. Turn on upper case names when you create the sync if you want to query without quotes; the setting cannot be changed afterwards.
Use key-pair authentication: Snowflake is phasing out single-factor password sign-in, so Weld recommends key-pair authentication for its user. Encrypted private keys are the safer choice.
Incremental syncs do not see deletes: The xmin method finds new and changed rows but cannot tell which rows were deleted. Use a full sync at midnight for tables with frequent deletes, or CDC, which reads deletes from the WAL.
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. Drop slots you no longer use and cap retention with max_slot_wal_keep_size.

Sources: Weld PostgreSQL connector docs · Weld PostgreSQL CDC docs · Weld Snowflake destination docs · Weld managed fields · Snowflake: About Openflow Connector for PostgreSQL · Snowflake: Openflow cost · Snowflake: Overview of data loading