Import Parquet

Apache Parquet is an open source, column-oriented persistent storage format designed for efficient data storage and retrieval. Kùzu supports bulk data import from Parquet files using the COPY FROM command. You can use COPY FROM to import data into an empty table or to append data to an existing table.

Notes

Parquet files store schema information in their metadata, so you don’t need to explicitly handle columns based on type, unlike in CSV. However, the same rules apply:

• Copy nodes before relationships: In order to copy a relationship table R from a Parquet file RFile, the nodes that appear in RFile need to already exist in the database (either imported in bulk or inserted through Cypher data manipulation commands).

Import to node table

Similar to CSV import, the order of columns in a Parquet file need to match the order of predefined properties for node tables in the catalog, i.e. the order used when defining the schema of a node table.

The following example is for a file named user.parquet. The output is obtained by using print(pyarrow.Table).

pyarrow.Table
name: string
age: int64
----
name: [["Adam","Karissa","Zhang","Noura"]]
age: [[30,40,50,25]]

To load this Parquet file into a User table, simply run:

COPY User FROM "user.Parquet";

Import to relationship table

Similar to CSV import, the first two columns for a relationship file should the from and the to columns that represent existing nodes’ primary keys.

The following example is for a file named follows.parquet. The output is obtained by using print(pyarrow.Table).

pyarrow.Table
from: string
to: string
since: int64
----
from: [["Adam","Adam","Karissa","Zhang"]]
to: [["Karissa","Zhang","Zhang","Noura"]]
since: [[2020,2020,2021,2022]]

To load this Parquet file into a Follows table, simply run:

COPY Follows FROM "follows.Parquet";

Import multiple files to a single table

It is common practice to divide large Parquet files into several smaller files for cleaner data management. Kùzu can read multiple files with the same structure, consolidating their data into a single node or relationship table. You can specify that multiple files are loaded in the following ways:

Glob pattern

This is similar to the Unix glob pattern, where you specify file paths that match a given pattern. The following wildcard characters are supported:

Wildcard	Description
*	match any number of any characters (including none)
?	match any single character
[abc]	match any one of the characters enclosed within the brackets
[a-z]	match any one of the characters within the range

COPY User FROM "User*.parquet"

List of files

Alternatively, you can just specify a list of files to be loaded.

COPY User FROM ["User0.parquet", "User1.parquet", "User2.parquet"]