Skip to main content
PySparkSnowflakeDatabricks

Snowflake with Python, PySpark, Databricks — Enterprise Integration

🎬 Story Time — “We Need All Our Tools Talking to Each Other”​

Ritika, a lead data engineer at a fast-growing SaaS company, faces an integration challenge.

Her ecosystem is huge:

  • Python notebooks for analysts
  • PySpark pipelines on Databricks
  • Machine learning workflows
  • Batch + streaming
  • Snowflake as the central data warehouse

The CTO declares:

“Everything must flow into Snowflake and out of Snowflake, seamlessly.”

Now Ritika must connect Python, PySpark, and Databricks in a clean, scalable architecture.

🧊 1. Snowflake + Python — Your Data Engineering Power Duo​

Python integrates with Snowflake through:

  • Snowflake Connector for Python
  • Snowpark for Python
  • Pandas + Snowflake Native Connectors
  • Streamlit-in-Snowflake (SIS)

Ritika starts with the Python connector.

🔌 1.1 Snowflake Python Connector​

Install:

pip install snowflake-connector-python

Connect:

import snowflake.connector

conn = snowflake.connector.connect(
    user='RITIKA',
    password='xxxxxxx',
    account='AB12345.ap-south-1',
    warehouse='ANALYTICS_WH',
    database='SALES_DB',
    schema='PUBLIC'
)

cursor = conn.cursor()
cursor.execute("SELECT COUNT(*) FROM ORDERS")
print(cursor.fetchone())

This powers:

  • ad hoc scripts
  • ETL micro-jobs
  • Python automations
  • Airflow & Prefect pipelines

🧠 1.2 Snowpark for Python — Server-Side Python​

Ritika discovers Snowpark, allowing Python logic to run inside Snowflake compute.

Install:

pip install snowflake-snowpark-python

Example:

from snowflake.snowpark import Session

session = Session.builder.configs(connection_parameters).create()

df = session.table("ORDERS")
df_filtered = df.filter(df["REVENUE"] > 1000)

df_filtered.show()

Benefits:​

  • Pushdown compute to Snowflake
  • Distributed processing
  • Zero data movement
  • ML model execution inside Snowflake

🔥 2. Snowflake + PySpark Integration​

Snowflake integrates with PySpark via the Spark Snowflake Connector.

Perfect for:

  • Large-scale Spark transformations
  • Ingest from Delta Lake
  • ETL pipelines running on Databricks or EMR
  • Converting Spark DataFrames → Snowflake tables

🔌 2.1 Spark Snowflake Connector Setup​

Add dependencies:

--packages net.snowflake:snowflake-jdbc:3.13.28,net.snowflake:spark-snowflake_2.12:2.12.0-spark_3.3

Connection options:

sfOptions = {
  "sfURL": "AB12345.snowflakecomputing.com",
  "sfAccount": "AB12345",
  "sfUser": "RITIKA",
  "sfPassword": "xxxx",
  "sfDatabase": "SALES_DB",
  "sfSchema": "PUBLIC",
  "sfWarehouse": "SPARK_WH"
}

Write Spark DataFrame → Snowflake​

df.write \
  .format("snowflake") \
  .options(**sfOptions) \
  .option("dbtable", "ORDERS_CLEAN") \
  .save()

Read from Snowflake → Spark​

df_snow = spark.read \
  .format("snowflake") \
  .options(**sfOptions) \
  .option("query", "SELECT * FROM SALES_DB.PUBLIC.ORDERS") \
  .load()

🏔️ 3. Snowflake + Databricks — A Modern Lakehouse Integration​

Databricks teams often use:

  • Spark for heavy transformations
  • MLflow for experimentation
  • Delta Lake for raw zone
  • Snowflake for analytics, BI & governed modeling

Ritika builds a pipeline:

  1. Raw data → Delta Lake
  2. Transform in Databricks using PySpark
  3. Load curated data → Snowflake
  4. Analysts query Snowflake using BI tools

🔗 3.1 Databricks + Snowflake Connector Example​

In Databricks notebook:

options = {
  "sfUrl": "ab12345.ap-south-1.snowflakecomputing.com",
  "sfUser": dbutils.secrets.get("snowflake", "USER"),
  "sfPassword": dbutils.secrets.get("snowflake", "PASSWORD"),
  "sfDatabase": "REVENUE_DB",
  "sfSchema": "PUBLIC",
  "sfWarehouse": "DBRICKS_WH"
}

spark_df = spark.sql("SELECT * FROM unified_sales")

spark_df.write \
  .format("snowflake") \
  .options(**options) \
  .option("dbtable", "UNIFIED_SALES_SF") \
  .mode("overwrite") \
  .save()

Why Databricks integrates well with Snowflake:​

  • High-performance parallel load
  • Supports Delta → Snowflake
  • Easy credential management via Secrets
  • Handles large ETL pipelines

🤖 4. Machine Learning Workflows​

Ritika combines:

  • Snowpark for Python (feature engineering inside Snowflake)
  • Spark ML or Databricks MLflow
  • Snowflake UDFs & UDTFs
  • Model scoring inside Snowflake

Example: Deploy ML model using Snowpark UDF:

@udf
def score_model(amount: float) -> float:
    return amount * 0.98  # simplified example

Apply on Snowflake table:

session.table("ORDERS").select(score_model("REVENUE")).show()

This removes the need for exporting large datasets.

🧠 5. Architecture Patterns​

✔ Pattern 1 — Databricks as Transformation Layer, Snowflake as Analytics​

  • Spark cleans & enriches
  • Snowflake stores final models & tables

✔ Pattern 2 — Snowpark-First Architecture​

  • All transformations in Snowflake
  • Only ML training outside

✔ Pattern 3 — Hybrid Lakehouse​

  • Delta for raw + bronze
  • Snowflake for gold semantic layers

📦 6. Best Practices​

  1. Use Snowpark where possible to avoid data movement
  2. Use Spark Connector for large-scale batch loads
  3. Do not oversize Snowflake warehouses for Spark loads
  4. Use COPY INTO for bulk micro-batch ingestion
  5. Use Secrets Manager on Databricks for credentials
  6. Monitor connector jobs through Query History
  7. Keep transformations close to the compute engine (Spark or Snowflake)

🎉 Real-World Ending — “Everything Works Together Now”​

With her new integration setup:

  • Python automations sync instantly with Snowflake
  • Spark pipelines load cleaned data at scale
  • Databricks notebooks talk to Snowflake seamlessly
  • ML workloads run inside Snowflake using Snowpark
  • No messy data exports or CSV dumps

Her CTO smiles:

“This is a true modern data platform. Excellent work.”

📘 Summary​

Snowflake integrates deeply with:

✔ Python & Snowpark​

✔ PySpark​

✔ Databricks​

✔ ML & Feature Engineering​

✔ Modern Lakehouse Workflows​

Together they create a scalable, flexible, and enterprise-grade data ecosystem.