Snowflake V2

Snowflake is the default choice for enterprise cloud data warehousing — its separation of compute and storage means queries scale independently of data volume, and virtual warehouse auto-suspend eliminates idle compute cost. SQL-standard with Snowflake extensions (VARIANT for semi-structured data, time travel, data sharing). Snowpark enables agents to push Python/pandas computation directly into the warehouse without data movement. Cortex AI adds built-in LLM functions (COMPLETE, CLASSIFY_TEXT, EXTRACT_ANSWER) that run inside Snowflake. For agents: the Python connector, Snowflake SQL API (REST), and Snowpark SDK are the primary integration paths. Confidence is docs-derived.

SQL API (REST): POST https://{account}.snowflakecomputing.com/api/v2/statements with {statement: "SELECT ...", timeout, role, warehouse, database, schema}. Response includes statementHandle; poll GET /api/v2/statements/{handle} for async result retrieval. Python connector: conn = snowflake.connector.connect({user, password, account, warehouse, database}); cursor = conn.cursor(); cursor.execute("SELECT ..."); rows = cursor.fetchall(). Snowpark: session = Session.builder.configs({...}).create(); session.sql("SELECT ...").show(); df = session.table("my_table"); df.filter(col("sales") > 100).show(). JDBC/ODBC: standard interfaces for BI tools.

Username + password (basic, deprecated for new projects). Key pair authentication: RSA private key for the connector (recommended). OAuth: service account OAuth for enterprise SSO environments. MFA: TOTP MFA enforced per-user via Snowflake admin. Role-based access control (RBAC): GRANT USAGE ON DATABASE to ROLE; per-object permissions. HTTPS enforced. Key pair rotation: generate new RSA key pair, assign to user, rotate. IP allowlisting configurable per-network policy.

Synchronous (small results) and async (large results) modes. Time-based query timeout configurable per-statement. Warehouse auto-suspend: warehouse spins down when idle (avoids cost); cold start adds 2-5 seconds to first query after idle. Snowflake credits: charged per virtual warehouse-second; monitor via QUERY_HISTORY view. Query failures: detailed error messages in SQL API response / connector exceptions. Snowflake uptime at status.snowflake.com (SLA 99.9%). Time travel: query historical snapshots up to 90 days (Enterprise). Data sharing: provider shares → consumer can query shared data without copying.

docs.snowflake.com is extensive — SQL reference, Python connector guide, Snowpark documentation, SQL API reference, Cortex AI functions guide, and performance tuning guides. Getting started: free 30-day trial ($400 credits), create virtual warehouse, first SQL query in under 5 minutes. Python connector: pip install snowflake-connector-python. Snowpark: pip install snowflake-snowpark-python. SnowSQL CLI for interactive SQL. Community via Snowflake Community Forum, Snowflake Summit content on YouTube, and GitHub. Enterprise-grade documentation.