Delta Clone Deep Dive: Cloning Tables Safely and Fast with DEEP/SHALLOW

"I want to try new ETL logic against production data without polluting production." "Copying a huge table takes forever." Delta Clone solves both of these problems. DEEP CLONE produces a fully independent copy of the data, while SHALLOW CLONE shares the underlying data files with the source and creates the clone almost instantly.

This article walks through the differences between DEEP and SHALLOW CLONE, their syntax, how to choose between them by use case, how they interact with VACUUM, Time Travel, and metadata, and the points most likely to appear on the exam.

DEEP CLONE vs SHALLOW CLONE

CREATE TABLE ... CLONE Syntax

-- DEEP CLONE: データの完全コピー
CREATE TABLE dev.sandbox.orders_clone
DEEP CLONE prod.silver.orders;

-- SHALLOW CLONE: メタデータのみコピー（データは参照）
CREATE TABLE dev.sandbox.orders_shallow
SHALLOW CLONE prod.silver.orders;

-- 特定バージョンのクローン（Time Travel）
CREATE TABLE dev.sandbox.orders_v5
DEEP CLONE prod.silver.orders VERSION AS OF 5;

-- 特定タイムスタンプのクローン
CREATE TABLE dev.sandbox.orders_snapshot
SHALLOW CLONE prod.silver.orders TIMESTAMP AS OF '2026-03-25';

-- 既存テーブルの差分更新（増分DEEP CLONE）
CREATE OR REPLACE TABLE dev.sandbox.orders_clone
DEEP CLONE prod.silver.orders;

Re-running a DEEP CLONE with CREATE OR REPLACE only copies the delta since the previous clone. This is called an incremental clone, and it works well as a daily sync pattern.

Use case 1: Experimentation and dev environments

When a data scientist wants to try a new model against production data, SHALLOW CLONE is the best fit. Even a multi-terabyte table can be cloned in seconds, freeing you to experiment with joins and filters. Writes to the clone do not touch the source.

-- 実験用にSHALLOW CLONEを作成
CREATE TABLE dev.ml_experiments.user_features
SHALLOW CLONE prod.gold.user_features;

-- クローン上で自由に実験（ソースに影響なし）
ALTER TABLE dev.ml_experiments.user_features
ADD COLUMN age_bucket STRING;

UPDATE dev.ml_experiments.user_features
SET age_bucket = CASE
  WHEN age < 25 THEN 'young'
  WHEN age < 50 THEN 'middle'
  ELSE 'senior'
END;

Use case 2: ETL testing

When you test changes to ETL logic, create a full copy of the production table with DEEP CLONE and run the entire pipeline against it in your test environment. Because a DEEP CLONE is fully independent of the source, running VACUUM during the test or hitting a pipeline failure has no effect on production.

-- テスト用にDEEP CLONEを作成
CREATE TABLE staging.test.orders
DEEP CLONE prod.silver.orders;

CREATE TABLE staging.test.customers
DEEP CLONE prod.silver.customers;

-- テスト環境でETLパイプラインを実行
-- 結果が期待通りか検証後、本番に適用

Use case 3: Data archiving

When compliance requires long-term snapshots of your data, DEEP CLONE is the right tool. The archive holds a specific point in time and is unaffected by VACUUM or schema changes on the source table.

-- 月次のアーカイブスナップショット
CREATE TABLE archive.monthly.orders_202603
DEEP CLONE prod.silver.orders
  TIMESTAMP AS OF '2026-03-31 23:59:59';

Metadata replicated by a clone

• Schema (column names and data types)
• NOT NULL constraints
• CHECK constraints
• Table properties (e.g. delta.enableChangeDataFeed)
• Partition information
• Column and table comments

Not cloned: table change history, streaming checkpoints, and Unity Catalog grants (the clone inherits the default permissions of its destination).

Interaction with VACUUM

The biggest risk with SHALLOW CLONE is VACUUM on the source. Once VACUUM deletes data files older than the retention period, any query against a SHALLOW CLONE that still references those files will fail.

-- ソースのVACUUM保持期間を確認
DESCRIBE DETAIL prod.silver.orders;
-- delta.deletedFileRetentionDuration を確認

-- SHALLOW CLONEを安全に使う場合の推奨設定
ALTER TABLE prod.silver.orders
SET TBLPROPERTIES (
  'delta.deletedFileRetentionDuration' = 'interval 30 days'
);

What the exam tests

• The difference between DEEP CLONE and SHALLOW CLONE (whether data is copied, dependence on the source)
• That a SHALLOW CLONE is affected by VACUUM on the source table
• Time Travel options on clones (VERSION AS OF / TIMESTAMP AS OF)
• Incremental DEEP CLONE via CREATE OR REPLACE
• That metadata (constraints and properties) is cloned
• Picking the right clone type for the use case (experiments → SHALLOW, archives → DEEP)

Check your understanding

Frequently Asked Questions