Streamlit Development

Streamlit Development

Build interactive data applications using Streamlit, test them locally, and deploy to Snowflake's native Streamlit environment.

Quick Start

Execution Modes:

• Local Development - PyPI packages, full environment
• Snowflake Deployment - Snowflake Anaconda packages, managed environment

Connection Pattern

Critical: Support both local development and Snowflake deployment:

import streamlit as st
from snowflake.snowpark.context import get_active_session
from snowflake.snowpark import Session

@st.cache_resource
def get_snowpark_session():
    try:
        return get_active_session()  # Snowflake
    except:
        return Session.builder.config('connection_name', 'default').create()  # Local

For connection setup, see the snowflake-connections skill for:

• Creating ~/.snowflake/connections.toml
• Authentication methods (SSO, key pair, username/password)
• Multiple environment configurations
• Environment variable overrides

Local Development

Setup

# Install dependencies (pin Streamlit version to match Snowflake)
uv pip install --system -r requirements.txt

# Example requirements.txt
streamlit==1.46.0  # Must match Snowflake version
snowflake-snowpark-python
pandas

Run Locally

streamlit run app.py
# Or with environment overrides (see snowflake-connections skill)
SNOWFLAKE_DATABASE=MY_DB streamlit run app.py

Testing

Unit/Integration Tests

pytest streamlit_app/tests/ -v

Browser Testing

Use Playwright MCP for interactive testing:

• Verify pages load without errors
• Test forms and navigation
• Check responsive design

See TESTING_GUIDE.md for patterns.

Deployment

Option 1: Snowflake CLI (Recommended)

snow streamlit deploy --replace -c default

Option 2: Schemachange

Include Streamlit deployment in migration scripts.

Key Patterns

• Separate data access from UI logic
• Cache Snowpark sessions
• Use forms for multi-field input
• Handle errors gracefully
• Limit column nesting (max 2 levels)

See TROUBLESHOOTING.md for common issues.

Related Skills

Complementary Testing:

• playwright-mcp skill - Automate browser testing for Streamlit apps

Use playwright-mcp for visual testing, form validation, responsive design testing, and accessibility checks of your Streamlit applications.

Code Organization Best Practices

1. Separate Data Access from UI

✅ DO: Modular data access

# utils/data_loader.py
class DataQueries:
    def __init__(self, session):
        self.session = session

    def get_sales(self, start_date, end_date):
        return self.session.sql(f"""
            SELECT * FROM sales
            WHERE date BETWEEN '{start_date}' AND '{end_date}'
        """).to_pandas()

# app.py
from utils.data_loader import DataQueries
session = get_snowpark_session()
queries = DataQueries(session)
df = queries.get_sales('2024-01-01', '2024-12-31')
st.dataframe(df)

❌ DON'T: Mix SQL with UI code

2. Cache Snowpark Session

Always cache your session to avoid reconnection overhead:

@st.cache_resource
def get_snowpark_session():
    """Get or create Snowpark session (cached)"""
    try:
        return get_active_session()  # When running in Snowflake
    except:
        from snowflake.snowpark import Session
        return Session.builder.config('connection_name', 'default').create()

3. Use Forms for Multi-Field Input

✅ DO: Group inputs in forms

with st.form("customer_form"):
    name = st.text_input("Name")
    email = st.text_input("Email")
    phone = st.text_input("Phone")

    if st.form_submit_button("Save"):
        save_customer(name, email, phone)
        st.success("Customer saved!")

❌ DON'T: Trigger rerun on every input (causes rerun on every keystroke)

4. Handle Errors Gracefully

Provide helpful feedback:

try:
    save_customer(name, email)
    st.success("✅ Customer saved successfully!")
except ValueError as e:
    st.error(f"❌ Invalid input: {e}")
    st.info("💡 Tip: Check that email format is correct")
except Exception as e:
    st.error(f"❌ Unexpected error: {e}")
    st.info("💡 Please contact support if this persists")

5. Limit Column Nesting (Max 2 Levels)

✅ DO: 2 levels maximum

col1, col2 = st.columns(2)
with col1:
    label_col, input_col = st.columns([1, 3])
    with label_col:
        st.markdown("**Name:**")
    with input_col:
        name = st.text_input("Name", label_visibility="collapsed")

❌ DON'T: 3+ levels of nested columns (causes Streamlit errors)

Performance Optimization

Cache Data Queries

@st.cache_data(ttl=600)  # Cache for 10 minutes
def load_sales_data(start_date, end_date):
    return session.sql(f"""
        SELECT * FROM sales
        WHERE date BETWEEN '{start_date}' AND '{end_date}'
    """).to_pandas()

Use Session State

# Initialize state
if 'data' not in st.session_state:
    st.session_state.data = load_data()

# Access throughout app
df = st.session_state.data

Lazy Load Heavy Computations

if st.button("Run Analysis"):
    with st.spinner("Analyzing..."):
        result = expensive_computation()
        st.session_state.result = result

if 'result' in st.session_state:
    st.write(st.session_state.result)

Snowflake-Specific Considerations

API Limitations

Some Streamlit features don't work in Snowflake:

Package Availability

Only Snowflake Anaconda packages available:

# environment.yml
name: streamlit_env
channels:
  - snowflake
dependencies:
  - pandas
  - plotly
  # ❌ DON'T include:
  # - streamlit  (already provided)
  # - snowflake-snowpark-python  (already provided)

Check package availability: https://repo.anaconda.com/pkgs/snowflake/

Python Version Support

Don't specify Python version - Snowflake controls this:

# ❌ DON'T DO THIS
dependencies:
  - python=3.11  # Wrong!

# ✅ DO THIS
dependencies:
  - pandas
  - plotly

Common Pitfalls

DuplicateWidgetID Error

Problem: Two widgets with same implicit key

Solution: Add explicit keys

st.text_input("Name", key="customer_name")
st.text_input("Name", key="product_name")

IndentationError

Check before deploying:

python -c "import ast; ast.parse(open('streamlit_app/app.py').read())"

Session Not Found (Local Development)

Ensure proper fallback:

def get_snowpark_session():
    try:
        return get_active_session()  # Snowflake
    except:
        from snowflake.snowpark import Session
        return Session.builder.config('connection_name', 'default').create()

Form Parameter Errors

Some parameters not supported in Snowflake:

• ❌ border=False in st.form()
• ❌ border=True in st.container()
• ❌ hide_index=True in st.dataframe() (older versions)

Pre-Deployment Checklist

Before Deploying

• ✅ Fix all indentation errors
• ✅ Run unit tests: pytest streamlit_app/tests/ -v
• ✅ Test locally: streamlit run streamlit_app/app.py
• ✅ Verify forms and navigation with browser testing
• ✅ Check environment.yml only has non-default packages
• ✅ Remove development dependencies
• ✅ Test with different user roles/permissions
• ✅ Verify data access controls

Deployment Commands

# Method 1: Snowflake CLI (Recommended)
snow streamlit deploy --replace --connection default

# Method 2: Schemachange
schemachange deploy --config-folder . --connection-name default

Post-Deployment

• ✅ Verify app appears in Snowflake: Data → Databases → Schema → Streamlit
• ✅ Test all features in Snowflake UI
• ✅ Check permissions for different roles
• ✅ Monitor for errors in Streamlit logs

Security Best Practices

1. Never Hardcode Credentials

✅ DO:

session = get_active_session()  # Uses Snowflake auth

❌ DON'T:

password = "secret123"  # Never do this!

2. Use Role-Based Access Control

# Check user role
current_role = session.sql("SELECT CURRENT_ROLE()").collect()[0][0]

if current_role == "ADMIN":
    st.write("Admin features visible")
else:
    st.info("Admin access required")

3. Validate User Inputs

def save_customer(name, email):
    if not name or len(name) < 2:
        raise ValueError("Name must be at least 2 characters")
    if "@" not in email:
        raise ValueError("Invalid email format")

    # Proceed with save
    ...

Resources

• connections.py - Required session pattern for local/Snowflake compatibility
• snowflake-connections skill - Connection setup, authentication, and multi-environment configuration
• playwright-mcp skill - Browser testing automation for Streamlit apps

Goal: Transform AI agents into expert Streamlit developers who build production-ready data applications with proper code organization, performance optimization, and Snowflake-specific best practices.