Getting started with PyIceberg

PyIceberg is a Python implementation for accessing Iceberg tables, without the need of a JVM.

Installation

Before installing PyIceberg, make sure that you're on an up-to-date version of pip:

pip install --upgrade pip

You can install the latest release version from pypi:

pip install "pyiceberg[s3fs,hive]"

You can mix and match optional dependencies depending on your needs:

Key	Description:
hive	Support for the Hive metastore
glue	Support for AWS Glue
dynamodb	Support for AWS DynamoDB
sql-postgres	Support for SQL Catalog backed by Postgresql
sql-sqlite	Support for SQL Catalog backed by SQLite
pyarrow	PyArrow as a FileIO implementation to interact with the object store
pandas	Installs both PyArrow and Pandas
duckdb	Installs both PyArrow and DuckDB
ray	Installs PyArrow, Pandas, and Ray
daft	Installs Daft
s3fs	S3FS as a FileIO implementation to interact with the object store
adlfs	ADLFS as a FileIO implementation to interact with the object store
snappy	Support for snappy Avro compression
gcs	GCS as the FileIO implementation to interact with the object store

You either need to install s3fs, adlfs, gcs, or pyarrow to be able to fetch files from an object store.

Connecting to a catalog

Iceberg leverages the catalog to have one centralized place to organize the tables. This can be a traditional Hive catalog to store your Iceberg tables next to the rest, a vendor solution like the AWS Glue catalog, or an implementation of Icebergs' own REST protocol. Checkout the configuration page to find all the configuration details.

For the sake of demonstration, we‘ll configure the catalog to use the SqlCatalog implementation, which will store information in a local sqlite database. We’ll also configure the catalog to store data files in the local filesystem instead of an object store. This should not be used in production due to the limited scalability.

Create a temporary location for Iceberg:

mkdir /tmp/warehouse

Open a Python 3 REPL to set up the catalog:

from pyiceberg.catalog.sql import SqlCatalog

warehouse_path = "/tmp/warehouse"
catalog = SqlCatalog(
    "default",
    **{
        "uri": f"sqlite:///{warehouse_path}/pyiceberg_catalog.db",
        "warehouse": f"file://{warehouse_path}",
    },
)

Write a PyArrow dataframe

Let's take the Taxi dataset, and write this to an Iceberg table.

First download one month of data:

curl https://d37ci6vzurychx.cloudfront.net/trip-data/yellow_tripdata_2023-01.parquet -o /tmp/yellow_tripdata_2023-01.parquet

Load it into your PyArrow dataframe:

import pyarrow.parquet as pq

df = pq.read_table("/tmp/yellow_tripdata_2023-01.parquet")

Create a new Iceberg table:

catalog.create_namespace("default")

table = catalog.create_table(
    "default.taxi_dataset",
    schema=df.schema,
)

Append the dataframe to the table:

table.append(df)
len(table.scan().to_arrow())

3066766 rows have been written to the table.

Now generate a tip-per-mile feature to train the model on:

import pyarrow.compute as pc

df = df.append_column("tip_per_mile", pc.divide(df["tip_amount"], df["trip_distance"]))

Evolve the schema of the table with the new column:

with table.update_schema() as update_schema:
    update_schema.union_by_name(df.schema)

And now we can write the new dataframe to the Iceberg table:

table.overwrite(df)
print(table.scan().to_arrow())

And the new column is there:

taxi_dataset(
  1: VendorID: optional long,
  2: tpep_pickup_datetime: optional timestamp,
  3: tpep_dropoff_datetime: optional timestamp,
  4: passenger_count: optional double,
  5: trip_distance: optional double,
  6: RatecodeID: optional double,
  7: store_and_fwd_flag: optional string,
  8: PULocationID: optional long,
  9: DOLocationID: optional long,
  10: payment_type: optional long,
  11: fare_amount: optional double,
  12: extra: optional double,
  13: mta_tax: optional double,
  14: tip_amount: optional double,
  15: tolls_amount: optional double,
  16: improvement_surcharge: optional double,
  17: total_amount: optional double,
  18: congestion_surcharge: optional double,
  19: airport_fee: optional double,
  20: tip_per_mile: optional double
),

And we can see that 2371784 rows have a tip-per-mile:

df = table.scan(row_filter="tip_per_mile > 0").to_arrow()
len(df)

Explore Iceberg data and metadata files

Since the catalog was configured to use the local filesystem, we can explore how Iceberg saved data and metadata files from the above operations.

find /tmp/warehouse/

More details

For the details, please check the CLI or Python API page.