Oct 16, 2026: Parquet Direct (Preview)¶

We’re introducing Parquet Direct in preview. With Parquet Direct, you can create read-only Apache Iceberg™ tables in Snowflake from Apache Parquet™ files in your cloud object storage. Snowflake registers your existing files in place, so you can query your data without copying it or rewriting it, and with better query performance than external tables.

This release includes the following capabilities:

• Schema inference and evolution. Snowflake automatically infers column definitions from your Parquet files. When you create the table with ENABLE_SCHEMA_EVOLUTION = TRUE, Snowflake also evolves the table schema as new columns and widened types appear.
• Hive partition inference. With a catalog integration that uses TABLE_FORMAT = HIVE, Snowflake infers partition columns from Hive-style key=value paths. For other layouts, use TABLE_FORMAT = NONE and define partition columns with expressions.
• Event-driven auto-refresh. Keep tables in sync with files that are added, changed, or deleted in Amazon S3, Azure Blob Storage, or Google Cloud Storage.
• Monitoring. Track auto-refresh with event tables, the SYSTEM$PARQUET_DIRECT_AUTO_REFRESH_STATUS function, and the ICEBERG_TABLE_FILE_REGISTRATION_HISTORY table function.

For more information, see the following topics:

• Parquet Direct
• Direct Iceberg tables
• CREATE ICEBERG TABLE (Parquet files in object storage)

See the following tutorials for step-by-step instructions:

• Tutorial: Auto-refresh a Parquet Direct table from Amazon S3
• Tutorial: Auto-refresh a Parquet Direct table from Azure Blob Storage
• Tutorial: Auto-refresh a Parquet Direct table from Google Cloud Storage