Snowflake Iceberg Tables Complete Guide: Building and Operating Open Table Format

Iceberg Tables let you create and manage tables in the Apache Iceberg open table format on Snowflake. The data is stored as Parquet files in cloud storage on an External Volume, and the Iceberg metadata (manifests and snapshots) delivers ACID transactions, schema evolution, and time travel. Other engines such as Spark, Trino, and Flink can also read the same data, making this the foundation of an open data lakehouse.

Apache Iceberg Architecture Overview

Iceberg Table
  │
  ├─ Catalog (metadata management)
  │    ├─ Snowflake Catalog (managed by Snowflake)
  │    └─ External Catalog (AWS Glue, Hive Metastore, etc.)
  │
  ├─ Metadata Layer
  │    ├─ metadata file (JSON) ← table schema, partition spec
  │    ├─ manifest list        ← list of manifests per snapshot
  │    └─ manifest file        ← list of data files and statistics
  │
  └─ Data Layer
       └─ Parquet files        ← the actual data (columnar, compressed)

Snowflake-Catalog Iceberg Tables

Snowflake manages both metadata and data. All DML operations (INSERT/UPDATE/DELETE/MERGE) are supported, so the experience is nearly identical to a regular Snowflake table.

-- Prerequisite: an External Volume must already exist
-- (See the External Volumes article for CREATE EXTERNAL VOLUME details)

-- Create a Snowflake-catalog Iceberg Table
CREATE OR REPLACE ICEBERG TABLE analytics.lakehouse.web_events (
  event_id STRING NOT NULL,
  user_id INT,
  event_type STRING,
  page_url STRING,
  event_timestamp TIMESTAMP_NTZ,
  properties OBJECT,
  session_id STRING
)
  CATALOG = 'SNOWFLAKE'
  EXTERNAL_VOLUME = 'iceberg_s3_vol'
  BASE_LOCATION = 'analytics/web_events/';

-- DML is supported
INSERT INTO analytics.lakehouse.web_events
VALUES (
  'evt-001', 12345, 'page_view', '/products/123',
  '2026-03-27 14:30:00',
  OBJECT_CONSTRUCT('referrer', 'google.com', 'device', 'mobile'),
  'sess-abc'
);

-- UPDATE
UPDATE analytics.lakehouse.web_events
SET properties = OBJECT_INSERT(properties, 'processed', TRUE)
WHERE event_id = 'evt-001';

-- DELETE
DELETE FROM analytics.lakehouse.web_events
WHERE event_timestamp < DATEADD(YEAR, -2, CURRENT_TIMESTAMP());

External-Catalog Iceberg Tables

-- Integration with AWS Glue Data Catalog

-- Step 1: Create the Catalog Integration
CREATE OR REPLACE CATALOG INTEGRATION glue_catalog_int
  CATALOG_SOURCE = GLUE
  CATALOG_NAMESPACE = 'analytics_db'
  TABLE_FORMAT = ICEBERG
  GLUE_AWS_ROLE_ARN = 'arn:aws:iam::123456789012:role/snowflake-glue-role'
  GLUE_CATALOG_ID = '123456789012'
  GLUE_REGION = 'us-east-1'
  ENABLED = TRUE;

-- Step 2: Create the external-catalog Iceberg Table (read-only)
CREATE OR REPLACE ICEBERG TABLE analytics.lakehouse.spark_events
  EXTERNAL_VOLUME = 'iceberg_s3_vol'
  CATALOG = 'glue_catalog_int'
  CATALOG_TABLE_NAME = 'web_events';

-- Read queries run normally
SELECT event_type, COUNT(*) AS cnt
FROM analytics.lakehouse.spark_events
WHERE event_timestamp >= '2026-03-01'
GROUP BY event_type
ORDER BY cnt DESC;

-- Refresh metadata (when updated externally)
ALTER ICEBERG TABLE analytics.lakehouse.spark_events REFRESH;

Catalog Management: Detailed Comparison

Schema Evolution

-- Add a column (Snowflake catalog)
ALTER ICEBERG TABLE analytics.lakehouse.web_events
  ADD COLUMN country_code STRING;

-- Rename a column
ALTER ICEBERG TABLE analytics.lakehouse.web_events
  RENAME COLUMN page_url TO url;

-- Drop a column
ALTER ICEBERG TABLE analytics.lakehouse.web_events
  DROP COLUMN session_id;

Iceberg schema evolution preserves backward compatibility. After adding a column, data from pre-add snapshots is read with the new column as NULL. Column renames are tracked internally by field ID, so existing Parquet files are never rewritten.

Time Travel

-- Query data at a specific point in time
SELECT * FROM analytics.lakehouse.web_events
AT (TIMESTAMP => '2026-03-26 00:00:00'::TIMESTAMP_NTZ);

-- List snapshots
SELECT *
FROM TABLE(INFORMATION_SCHEMA.ICEBERG_TABLE_SNAPSHOTS(
  TABLE_NAME => 'analytics.lakehouse.web_events'
));

Standard Table vs Iceberg Table

Converting a Standard Table to an Iceberg Table

-- Convert via CTAS (CREATE TABLE AS SELECT)
CREATE OR REPLACE ICEBERG TABLE analytics.lakehouse.web_events_iceberg
  CATALOG = 'SNOWFLAKE'
  EXTERNAL_VOLUME = 'iceberg_s3_vol'
  BASE_LOCATION = 'analytics/web_events_iceberg/'
AS
  SELECT * FROM analytics.standard.web_events;

Best Practices

• Use a standard table if you do not need multi-engine access: for Snowflake-only workloads, standard tables deliver better DML performance and simpler management.
• Prefer the Snowflake catalog as the default: choose the Snowflake catalog when you need DML, and reach for an external catalog only when integrating with an existing Spark pipeline.
• REFRESH regularly when using an external catalog: changes made externally (e.g. by Spark) are not visible in Snowflake until REFRESH is executed.
• Use BASE_LOCATION to keep paths logically separated: assign a distinct BASE_LOCATION per table to keep cloud storage organized.
• Manage snapshots regularly: accumulating old snapshots bloats metadata, so define and enforce a retention policy.

Check Your Understanding

Frequently Asked Questions