Redshift to Snowflake Migration

Point this agent at your warehouse codebase (SQL, dbt, orchestration) and ask it to plan your migration to Snowflake. It inventories the Redshift footprint, plans an UNLOAD → S3 Parquet → external stage → COPY INTO pipeline, translates DDL and dialect differences, and validates a sample slice in a Snowflake sandbox database with paired row counts and checksums.

What it does

1. Map the Redshift footprint: table DDL with dist/sort keys and encodings, views (including late-binding), stored procedures, UDFs, WLM queues, scheduled queries, Spectrum external tables, and downstream BI connections.
2. Plan the migration: schema-by-schema order, the UNLOAD → Parquet → stage → COPY INTO load pipeline, clustering-key decisions driven by profiling (not by copying sort keys), warehouse/resource-monitor sizing to replace WLM, and a cutover plan with a dual-run window.
3. Generate DDL: translated CREATE TABLE/VIEW statements with the type mapping applied (SUPER→VARIANT, IDENTITY→AUTOINCREMENT, encodings dropped), dialect rewrites for common patterns, and procedures flagged for Snowflake Scripting rewrite.
4. Sandbox — create the schema in the sandbox database, load a sample slice from the staging bucket, run paired queries on both engines, compare row counts and checksums, and report PASS/PARTIAL/FAIL evidence.

Prerequisites

• SNOWFLAKE_ACCOUNT — test account identifier
• SNOWFLAKE_USER — user for the sandbox
• SNOWFLAKE_PASSWORD — that user's password (test account only)
• SNOWFLAKE_WAREHOUSE — sandbox warehouse
• SNOWFLAKE_DATABASE — sandbox database the agent may create schemas in
• AWS_READONLY_ACCESS_KEY_ID / AWS_READONLY_SECRET_ACCESS_KEY — optional; read access to the S3 bucket where you run UNLOAD

Current scope

This solution translates schemas, plans loads, and validates sample slices. It does not execute UNLOAD on your Redshift cluster (you run the generated UNLOAD statements yourself), does not rewrite PL/pgSQL stored procedure bodies (it flags and inventories them for Snowflake Scripting rewrite), and does not migrate Spectrum external tables, users/grants, or BI dashboards — those each get a planned follow-up lane.