Load CSV into Snowflake from your Mac.

The CSV-to-Snowflake workflow today

Standard Snowflake CSV loading requires multiple SQL commands: CREATE FILE FORMAT, PUT to upload to internal stage, COPY INTO to load from stage to table, optionally CREATE TABLE first if the destination doesn't exist. Each step has its own quirks (file format escaping, COPY error handling, NULL value mapping). For developers loading CSVs occasionally, this is a lot of friction.

QueryFlow's pipeline approach

Visual ETL pipeline: drag CSV source card onto canvas, point at file. Drag Snowflake destination, pick database/schema/table (or create new). Click the connecting line, AI Map auto-detects column matches. Run. Behind the scenes QueryFlow handles file format detection, type inference, internal stage upload via Snowpipe-style PUT, and COPY INTO with appropriate error handling. The complexity is hidden, not removed.

Schedule recurring loads from a folder

Common workflow: a vendor delivers a CSV daily to an SFTP folder, you need to load it into Snowflake every morning. QueryFlow setup: configure SFTP source pointing at the folder, configure Snowflake destination, schedule daily at 7 AM. Each morning the pipeline grabs new files, loads them, archives the original. No infrastructure required beyond your Mac.

Type handling and edge cases

QueryFlow's CSV source auto-detects types (string, integer, decimal, date, timestamp, boolean) from sample rows. You can override types per column. Common edge cases handled: quoted strings with embedded commas, newlines within quoted fields, different date formats per column, NULL representations like 'NULL' string or empty fields, BOM markers, character encoding (UTF-8, UTF-16, Latin-1). For pathological CSV files, you can use a Flow Book to clean the file with pandas before loading.

Handling load failures gracefully

Snowflake COPY INTO can fail on bad rows. QueryFlow's pipeline lets you configure failure behavior: stop on first error (default for small loads), continue on errors with bad rows captured for review (for large loads where you accept some loss), or use ON_ERROR=SKIP_FILE to skip files entirely. Failed rows surface in the Observatory dashboard.

Performance benchmarks

On an M2 MacBook Pro with a Snowflake X-Small warehouse: 100K row CSV load: ~30 seconds end-to-end. 1M row CSV load: ~3-4 minutes. 10M row CSV load: ~30-40 minutes. The bottleneck is rarely your Mac — Snowflake warehouse size and concurrent operations on the warehouse matter more.

Bulk patterns

For very large CSV loads (100M+ rows), split the file using shell tools (split -l 10000000 huge.csv) and load chunks in parallel by running multiple QueryFlow pipelines simultaneously. For truly massive workloads, Snowpipe (Snowflake's continuous ingestion) remains the right tool, but for the 95% of CSV loads under 10M rows, QueryFlow is dramatically simpler.