Iceberg Tables
Snowflake · Advanced Features
Theory
An Iceberg table in Snowflake stores its data as Parquet files plus Apache Iceberg metadata in cloud storage that you own, connected through an external volume. Snowflake can be the catalog (Snowflake-managed Iceberg tables, with full read and write from Snowflake) or it can read tables whose catalog lives elsewhere, such as AWS Glue, through a catalog integration.
Because the format is open, other engines such as Spark, Flink or Trino can read the same files without copying them into Snowflake. You give up some native features and conveniences in return; for example, storage is billed by your cloud provider and some table features differ from native tables.
Why we use it
Many companies run more than one engine on the same data lake. Iceberg tables let Snowflake work on that shared data while avoiding lock-in to a single storage format and avoiding duplicate copies.
When to use it
Use Iceberg tables when data must be readable by several engines or kept in your own storage in an open format. Use native Snowflake tables when Snowflake is the only engine, since they are simpler and fully managed.
Why an Open Format Matters
Because the underlying files are plain Parquet with an open Iceberg catalog, the exact same table can also be read by Spark, Flink, or Trino — no export step, no vendor lock-in on the data itself, only on where it's cataloged.
What You Give Up vs a Native Table
Iceberg tables support ACID transactions and most standard SQL, but some Snowflake-native conveniences (certain clustering and Time Travel behaviors) work a little differently than on a native permanent table.
When It's the Right Choice
Reach for Iceberg specifically when other engines outside Snowflake genuinely need to read the same physical files — a multi-engine data lake strategy — not as a default replacement for ordinary tables.
How it works
- Data must be shared with other engines
- Create an external volume on your cloud storage
- Should Snowflake manage the Iceberg catalog? Yes: CREATE ICEBERG TABLE with CATALOG = 'SNOWFLAKE'. No: Use a catalog integration to an external catalog.
- Query the table with normal SQL in Snowflake
- Other engines read the same Parquet and metadata files
Example
CREATE OR REPLACE ICEBERG TABLE lake_events (
event_id INT, event_type VARCHAR, amount NUMBER(10,2), event_time TIMESTAMP_NTZ(6))
CATALOG = 'SNOWFLAKE'
EXTERNAL_VOLUME = 'my_external_volume'
BASE_LOCATION = 'events_iceberg/';
SELECT * FROM lake_events WHERE event_type = 'purchase';
-- Open format: readable by Spark/Flink/Trino too, full DML, Time Travel worksKey points
- Same data readable by Snowflake, Spark, Flink and Trino
- Files are open Parquet, not a Snowflake-proprietary format
- Needs an external volume pointing at your cloud storage
- Catalog can be Snowflake or an external one via a catalog integration
- Best for multi-engine data lake strategies, not a default table type
- Horizon Catalog's Iceberg REST API lets external engines read Snowflake-managed tables