Attaching to PostgreSQL databases

The PostgreSQL extension can be installed and loaded by running the following commands using the Kuzu CLI or your preferred language client API:

INSTALL postgres;
LOAD postgres;

Set up a PostgreSQL server via Docker

During local development, it’s convenient to set up a PostgreSQL server using Docker. Run the following command to start a PostgreSQL server locally:

docker run --name kuzu-postgres \
       -e POSTGRES_PASSWORD=testpassword \
       -p 5432:5432 \
       --rm postgres:latest

Note that the storage volume for this database is not persistent and will be deleted once the container is stopped. Moreover, the password in this case is provided via plain text, which is not recommended in a real use case, so the below example is for testing purposes only.

1. Create a sample PostgreSQL database

To illustrate the usage of the extension, we create a sample Postgres database of university students. We will use asyncpg, an asynchronous PostgreSQL client library for Python, to create the database and insert some sample data via a Python script. This step assumes you have run pip install asyncpg to install the library, and are using Python 3.10+.

import asyncio
import asyncpg

async def main():
    conn = await asyncpg.connect('postgresql://postgres:testpassword@localhost:5432/postgres')
    # Create and insert data to a new table
    try:
        await conn.execute("CREATE TABLE IF NOT EXISTS person (name VARCHAR, age INTEGER);")
        await conn.execute("INSERT INTO person (name, age) VALUES ('Alice', 30)")
        await conn.execute("INSERT INTO person (name, age) VALUES ('Bob', 27)")
        await conn.execute("INSERT INTO person (name, age) VALUES ('Carol', 19)")
        await conn.execute("INSERT INTO person (name, age) VALUES ('Dan', 25)")
    except asyncpg.exceptions.DuplicateTableError:
        print(f"Table already exists, skipping creation and insertion...")
    # Check results
    print(await conn.fetch("SELECT * FROM person"))

asyncio.run(main())

2. Attach database

ATTACH [PG_CONNECTION_STRING] AS [alias] (dbtype postgres, skip_unsupported_table = OPTION, schema = SCHEMA_NAME)

skip_unsupported_table: Whether Kuzu should throw an exception or skip when encountering a table with unsupported datatype. Supported options are: TRUE or FALSE. Default is FALSE.
schema: The name of the schema in Postgres to attach. Kuzu attaches to the public schema of postgres by default.

The below example shows how the university PostgreSQL database can be attached to Kuzu using the alias uw:

ATTACH 'dbname=university user=postgres host=localhost password=testpassword port=5432' AS uw (dbtype postgres);

The ATTACH statement requires the following parameters:

PG_CONNECTION_STRING: PostgreSQL connection string with the necessary parameters
alias: Database alias to use in Kuzu - If not provided, the database name from PostgreSQL will be used. When attaching multiple databases, it’s recommended to use aliasing.

The below table lists some common connection string parameters:

Parameter	Description	Default
`dbname`	Database name	[user defined]
`host`	Host IP address	`localhost`
`user`	Postgres username	`postgres`
`password`	Postgres password	[empty]
`port`	Port number	5432

3. Data type mapping from PostgreSQL to Kuzu

The table below shows the mapping from PostgreSQL’s type to Kuzu’s type:

PostgreSQL Data Type	Corresponding Data Type in Kuzu
bigint (int8)	INT64
bigserial (serial8)	INT64
bit [ (n) ]	STRING
bit varying [ (n) ] (varbit [ (n) ])	STRING
boolean (bool)	BOOL
box	DOUBLE[]
bytea	BLOB
character [ (n) ] (char [ (n) ])	STRING
character varying [ (n) ] (varchar [ (n)])	STRING
cidr	STRING
circle	DOUBLE[]
date	DATE
double precision (float8)	DOUBLE
inet	STRING
integer (int, int4)	INT32
interval [ fields ] [ (p) ]	INTERVAL
json	JSON
line	DOUBLE[]
lseg	DOUBLE[]
macaddr	STRING
macaddr8	STRING
money	STRING
numeric [ (p, s) ] (decimal [ (p, s) ])	DECIMAL
path	DOUBLE[]
pg_lsn	STRING
pg_snapshot	STRING
point	STRUCT(x DOUBLE, y DOUBLE)
polygon	DOUBLE[]
real (float4)	FLOAT
smallint (int2)	INT16
smallserial (serial2)	INT16
serial (serial4)	INT32
text	STRING
time [ (p) ] [ without time zone ]	UNSUPPORTED
time [ (p) ] with time zone (timetz)	UNSUPPORTED
timestamp [ (p) ] [ without time zone ]	TIMESTAMP
timestamp [ (p) ] with time zone (timestamptz)	UNSUPPORTED
tsquery	STRING
tsvector	STRING
txid_snapshot	STRING
uuid	UUID
xml	STRING

4. Scan from PostgreSQL tables with Cypher

Finally, we can utilize the LOAD FROM statement to scan the Person table.

LOAD FROM uw.person
RETURN *

Result:

┌────────┬───────┐
│ name   │ age   │
│ STRING │ INT64 │
├────────┼───────┤
│ Alice  │ 30    │
│ Bob    │ 27    │
│ Carol  │ 19    │
│ Dan    │ 25    │
└────────┴───────┘

5. Scan from PostgreSQL tables with SQL

Starting from v0.9.0, Kuzu provides a sql_query table function allows you to execute arbitrary read-only SQL queries on any attached Postgres database and retrieve its results for use in Kuzu. The function takes in two parameters:

ATTACHED_PG_NAME: the name of the attached postgres database to execute SQL queries.
SQL_QUERY: the read SQL query to execute.

The example below shows how to scan from a SQL query which filters out all people who are younger than 20:

CALL SQL_QUERY('uw', 'SELECT * FROM person WHERE age > 20')
RETURN *

Result:

┌────────┬───────┐
│ name   │ age   │
│ STRING │ INT64 │
├────────┼───────┤
│ Alice  │ 30    │
│ Bob    │ 27    │
│ Dan    │ 25    │
└────────┴───────┘

6. USE: Reference database without alias

You can use the USE statement for attached databases to use a default database name for future operations. This can be used when reading from an attached database to avoid specifying the full database name as a prefix to the table name.

Consider the same attached database as above:

ATTACH 'university' AS uw (dbtype postgres);

Instead of defining the database name for each subsequent clause like this:

LOAD FROM uw.person
RETURN *

You can do:

USE uw;
LOAD FROM person
RETURN *

7. Copy data from PostgreSQL tables

One important use case of the external RDBMS extensions is to facilitate seamless data transfer from the external RDBMS to Kuzu. In this example, we continue using the university.db database created in the last step, but this time, we copy the data and persist it to Kuzu. This is done with the COPY FROM statement. Here is an example:

We first create a Person table in Kuzu. In this example we will make Person have the same schema as the one defined in the external RDBMS.

CREATE NODE TABLE Person (name STRING, age INT32, PRIMARY KEY(name));

When the schemas are the same, we can copy the data from the external DBMS table to the Kuzu table simply as follows.

COPY Person FROM uw.person;

If the schemas are not the same, e.g., Person contains only name property while the external uw.person contains name and age, we can still use COPY FROM but with a subquery as follows:

COPY Person FROM (LOAD FROM uw.person RETURN name);

Note that you can use the SQL_QUERY function as a subquery as well.

COPY Person FROM SQL_QUERY('uw', 'SELECT name FROM person');

8. Query the data in Kuzu

Finally, we can verify the data in the Person table in Kuzu.

MATCH (p:Person) RETURN p.*;

Result:

┌────────┬───────┐
│ name   │ age   │
│ STRING │ INT64 │
├────────┼───────┤
│ Alice  │ 30    │
│ Bob    │ 27    │
│ Carol  │ 19    │
│ Dan    │ 25    │
└────────┴───────┘

9. Clear attached database schema cache

To avoid redundantly retrieving schema information from attached databases, Kuzu maintains a schema cache including table names and their respective columns and types. Should modifications occur in the schema via an alternate connection to attached RDBMSs, such as creation or deletion of tables, the cached schema data may become obsolete. You can use the clear_attached_db_cache() function to refresh cached schema information in such cases.

CALL clear_attached_db_cache()

Note: If you have attached to databases from different RDBMSs, say Postgres, DuckDB, and Sqlite, this call will clear the cache for all of them.

10. Detach database

To detach a database, use DETACH [ALIAS] as follows:

DETACH uw