Combining the power of MinIO’s high-performance, scalable enterprise object storage with the fast, in-memory data processing capabilities of Polars, a lightning-fast DataFrame library, can significantly enhance the performance of your data pipelines. This is particularly true in AI workflows, where preprocessing large datasets and performing feature selection are critical steps. In this post, we will explore how integrating MinIO with Polars can streamline your data workflows and optimize performance, especially for complex analytical workloads.

Why Polars for AI Data Pre-Processing?

Polars is a DataFrame library designed for speed. Unlike traditional Python-based libraries like Pandas, Polars is built in Rust, allowing it to handle large datasets efficiently. Polars employs an eager execution model, providing fast results by performing operations immediately rather than waiting for delayed computations. This makes Polars particularly useful for real-time analytics and time-sensitive data processing.

Key Polars features:

• Speed: Built with Rust, Polars is extremely fast and can handle large datasets far beyond what Pandas can manage.
• Memory Efficiency: Utilizes a columnar memory layout, which makes operations faster and reduces memory consumption.
• Lazy Execution: Polars has a lazy API that optimizes query plans by reordering and combining operations for better performance.
• Multithreading: Polars leverages multithreading for parallel computations, allowing it to process data much faster than single-threaded solutions.

Key MinIO features:

• Performance: As the fastest object store on the market, MinIO’s high-performance perfectly complements Polars' speed, capable of retrieving and storing massive datasets.
• Scale: MinIO’s distributed architecture scales horizontally, keeping pace with your growing AI/ML workloads while Polars efficiently crunches through the data.
• Data Durability and Redundancy: MinIO’s erasure coding and object locking protect your data in a modern, truly efficient way.
• Integration with AI/ML Frameworks: Through MinIO’s strong compliance to the S3 API and a robust SDK, MinIO supports a wide variety of AI/ML frameworks like TensorFlow and PyTorch. Through these integrations, you can retrieve your pre-processed data flies with Polars straight to training and inference without a hitch.

Accelerating Polars Workflows with GPU (Optional)

For those looking for even higher performance, Polars offers a beta release for a GPU engine powered by RAPIDS cuDF, providing up to 13x faster processing on NVIDIA GPUs. This is particularly useful when dealing with hundreds of millions of rows, where even small performance boosts can significantly reduce processing time.

To access this GPU acceleration, you simply need to install Polars with GPU support and specify the GPU engine when collecting your data.

pip install polars[gpu] --extra-index-url=https://pypi.nvidia.com

The other integration information will be the same.

Integrating MinIO with Polars

Let’s explore how MinIO can be integrated into a cohesive data processing pipeline. Whether you’re dealing with large-scale time series data, log files, or AI/ML model training datasets, MinIO provides the storage foundation, while Polars processes this data quickly and efficiently.

Step 1: Ensure Docker is Installed

Install Docker (if not done already): Follow the official Docker installation guide.

Step 2: Deploy MinIO in a Rootless Docker Container

Run the MinIO Container: Next, start the MinIO container in rootless mode. You will specify the data directory and the access and secret keys. Adjust the port and directory as needed.

mkdir -p ${HOME}/minio/data
docker run \
   -p 9000:9000 \
   -p 9001:9001 \
   --user $(id -u):$(id -g) \
   --name minio1 \
   -e "MINIO_ROOT_USER=ROOTUSER" \
   -e "MINIO_ROOT_PASSWORD=CHANGEME123" \
   -v ${HOME}/minio/data:/data \
   quay.io/minio/minio server /data --console-address ":9001"

• -p 9000:9000: Exposes MinIO’s API on port 9000.
• -p 9001:9001: Exposes the web console on port 9001.
• -v ~/minio/data:/data: Mounts the ~/minio/data directory on the host to store the data.
• MINIO_ROOT_USER and MINIO_ROOT_PASSWORD are used for authentication.

Step 3: Access MinIO

Once the container is up, open a web browser and go to: http://localhost:9001

Log in using the MINIO_ROOT_USER and MINIO_ROOT_PASSWORD credentials.

Step 4: Create a Bucket and Upload a Parquet File

Create a bucket in MinIO according to these instructions: 

Next, add Parquet files to your bucket.

Step 2: Accessing Data from MinIO in Polars

To read data from MinIO into a Polars DataFrame, you can use MinIO’s S3-compatible API with the requests library in Python. Authenticate using your MinIO username (access key) and password (secret key).

Let’s say your data is stored in a Parquet file; you will first need to pip install MinIO and Polars.

pip install minio
pip install polars

here’s how you can read that data directly into Polars:

import polars as pl
from minio import Minio
import io

# Configure MinIO S3 access
minio_url = "localhost:9000" 
access_key = "ROOTUSER"
secret_key = "CHANGEME123"

# Initialize MinIO client
client = Minio(
    minio_url,
    access_key=access_key,
    secret_key=secret_key,
    secure=False  # Set to True if you're using HTTPS
)

# Retrieve the parquet file from the bucket
bucket_name = "ducknest"
object_name = "wild_animals.parquet"

# Download the object as a stream
response = client.get_object(bucket_name, object_name)

# Read the file content into a Polars DataFrame
data = io.BytesIO(response.read())
df = pl.read_parquet(data)

# Perform your data analysis
print(df.describe())

Step 3: Processing Large Datasets with Polars

Polars really shines when working with large datasets. Its memory efficiency and multithreading allow it to handle complex operations like filtering, grouping, and aggregation much faster than traditional libraries like Pandas. MinIO comes into play by providing the perfect performant storage layer to handle these massive datasets. No matter how large your datasets get, data retrieval can remain fast and efficient. This is because MinIO’s speed is throttled only by the underlying hardware. Polars and MinIO work together to make a powerful combination, enabling smooth data processing and minimizing bottlenecks in your AI/ML pipelines.

For instance, here’s how you can perform an aggregate operation on your Polars DataFrame:

# Group by the correct column names (as per the schema)
result = df.group_by("category").agg(
    [
        pl.col("value").count().alias("total_value"),  # Count the number of animals in each habitat
        pl.col("quantity").mean().alias("avg_quantity")  # Calculate the average species value (after casting to numeric)
    ]
)

# Print the result
print(result)

When You’re Ready to Deploy

When you're ready to deploy, MinIO’s scalability shines, effortlessly managing massive datasets, while Polars accelerates data processing, ensuring smooth, end-to-end performance. MinIO’s Enterprise Object Store (EOS) is not only cost-effective compared to traditional block storage solutions, but also wildly enhances performance.

For organizations looking for more control and insight, the MinIO Enterprise Console is a powerful tool. It offers a unified "single pane of glass" to manage all your MinIO deployments, whether on-prem, in the cloud, or at the edge. Another standout feature of MinIO Enterprise Object Store is the Enterprise Catalog, which enables real-time searching and querying of object metadata at exabyte scale. Using a GraphQL interface, administrators can perform compliance checks, operational audits, and manage space utilization with ease. These are just two in a full suite of enterprise tooling available to use specifically built for large scale deployments of MinIO. You’ll have what you need when you’re ready to deploy MinIO and Polars together.

Conclusion

By integrating MinIO Enterprise Object Store with Polars, you can build high-performance, scalable data pipelines capable of processing massive datasets with ease. Whether you’re working on real-time analytics, large-scale AI/ML workloads, or just handling huge data lakes, this combination delivers both speed and efficiency. As the demand for faster data processing and scalable storage grows, leveraging technologies like MinIO and Polars will become increasingly important for modern data infrastructures. Let us know if you have any questions while you integrate at hello@min.io or on our Slack channel.