Materializations (view, table, incremental, ephemeral) in dbt

Introduction

Materializations decide how dbt saves your data models. They help you choose if data is saved as a table, a view, or handled differently for faster or flexible use.

When you want to save a model as a permanent table for fast queries.

When you want a lightweight view that always shows fresh data.

When you want to update only new or changed data to save time.

When you want to use a model only inside other models without saving it.

Syntax

dbt

materialized: [view|table|incremental|ephemeral]

Specify materialization in the model's config block or in the dbt_project.yml file.

Each materialization affects performance and storage differently.

Examples

This saves the model as a physical table in the database.

dbt

config(materialized='table')

This creates a view that runs the query fresh each time.

dbt

config(materialized='view')

This updates only new or changed rows, saving time on large datasets.

dbt

config(materialized='incremental')

This does not create a table or view but inlines the SQL in dependent models.

dbt

config(materialized='ephemeral')

Sample Program

This example shows how to set each materialization type in dbt models. The incremental model updates only new data. The ephemeral model is used inside another model without creating a table or view.

dbt

/* Example dbt model using different materializations */

-- models/my_table_model.sql
{{ config(materialized='table') }}

select id, name from source_data

-- models/my_view_model.sql
{{ config(materialized='view') }}

select id, name from source_data

-- models/my_incremental_model.sql
{{ config(materialized='incremental', unique_key='id') }}

select id, name from source_data

{% if is_incremental() %}
  where updated_at > (select max(updated_at) from {{ this }})
{% endif %}

-- models/my_ephemeral_model.sql
{{ config(materialized='ephemeral') }}

select id, name from source_data where active = true

-- models/using_ephemeral.sql
select * from {{ ref('my_ephemeral_model') }}

Important Notes

Incremental models need a unique key to identify new or changed rows.

Ephemeral models are useful for reusable SQL snippets without extra storage.

Views always show the latest data but can be slower than tables for big data.

Summary

Materializations control how dbt saves or uses your data models.

Choose 'table' for permanent storage, 'view' for fresh queries, 'incremental' for efficient updates, and 'ephemeral' for inline SQL.

Each type helps balance speed, storage, and freshness based on your needs.

Practice

1. Which dbt materialization creates a permanent table in the database that stores data physically?

easy

A. table

B. view

C. incremental

D. ephemeral

Solution

1. Step 1: Understand the purpose of 'table' materialization

   The 'table' materialization creates a physical table in the database that stores data permanently.

2. Step 2: Compare with other materializations

   'view' creates a virtual table, 'incremental' updates existing tables efficiently, and 'ephemeral' runs inline SQL without creating tables.

3. Final Answer:

   table -> Option A

4. Quick Check:

   Permanent storage = table [OK]

Hint: Permanent data storage means 'table' materialization [OK]

Common Mistakes:

• Confusing 'view' with 'table' as both represent data
• Thinking 'incremental' creates a full new table every time
• Assuming 'ephemeral' creates physical tables

2. Which of the following is the correct syntax to specify an incremental materialization in a dbt model's config block?

easy

A. config(materialization = 'incremental')

B. config(materialized = 'incremental')

C. materialized('incremental')

D. set materialized = incremental

Solution

1. Step 1: Recall dbt config syntax for materialization

   dbt uses config() with the keyword 'materialized' to set materialization type.

2. Step 2: Identify the correct keyword and format

   The correct syntax is config(materialized = 'incremental'). Other options use wrong keywords or syntax.

3. Final Answer:

   config(materialized = 'incremental') -> Option B

4. Quick Check:

   Correct keyword is 'materialized' inside config() [OK]

Hint: Use config(materialized = 'type') syntax for materializations [OK]

Common Mistakes:

• Using 'materialization' instead of 'materialized'
• Trying to call materialized as a function
• Using SQL-like SET syntax instead of config()

3. Given this dbt model config and SQL snippet:

-- model.sql
{{ config(materialized='incremental', unique_key='id') }}
select id, value from source_table
{% if is_incremental() %}
  where updated_at > (select max(updated_at) from {{ this }})
{% endif %}

What happens when you run this model multiple times?

medium

A. The model rebuilds the entire table every time

B. The model creates a view that always shows fresh data

C. The model appends only new or updated rows based on 'updated_at'

D. The model runs inline SQL without creating a table

Solution

1. Step 1: Understand incremental materialization with unique_key

   The model uses incremental materialization with a unique key 'id' to update data efficiently.

2. Step 2: Analyze the is_incremental() condition

   When running incrementally, it filters rows where 'updated_at' is newer than the max in the existing table, appending only new or updated rows.

3. Final Answer:

   The model appends only new or updated rows based on 'updated_at' -> Option C

4. Quick Check:

   Incremental + filter = append updates [OK]

Hint: Incremental with is_incremental() filters new data only [OK]

Common Mistakes:

• Thinking incremental rebuilds full table every run
• Confusing view materialization with incremental
• Ignoring the is_incremental() condition

4. You wrote this dbt model:

{{ config(materialized='ephemeral') }}
select * from source_table

But when you run dbt, you get an error saying the model is not found. What is the likely cause?

medium

A. Ephemeral models do not create tables or views, so they cannot be run directly

B. The config syntax for ephemeral is incorrect

C. Ephemeral models require a unique_key to run

D. You must specify incremental materialization for ephemeral models

Solution

1. Step 1: Recall what ephemeral materialization does

   Ephemeral models do not create tables or views; their SQL is inlined into dependent models.

2. Step 2: Understand why the error occurs

   Since ephemeral models don't create database objects, running them directly causes a 'model not found' error.

3. Final Answer:

   Ephemeral models do not create tables or views, so they cannot be run directly -> Option A

4. Quick Check:

   Ephemeral = inline SQL, no table/view created [OK]

Hint: Ephemeral models can't be run alone; they inline SQL [OK]

Common Mistakes:

• Trying to run ephemeral models directly
• Assuming ephemeral needs unique_key
• Confusing ephemeral with incremental

5. You want to build a dbt model that:
- Stores data permanently
- Updates only new rows efficiently
- Avoids rebuilding the entire dataset each run

Which materialization should you choose and why?

hard

A. Use 'table' materialization because it stores data permanently and rebuilds fully each run

B. Use 'ephemeral' materialization because it runs inline SQL without storage

C. Use 'view' materialization because it always shows fresh data without storage

D. Use 'incremental' materialization because it stores data permanently and updates only new rows

Solution

1. Step 1: Identify permanent storage requirement

   Both 'table' and 'incremental' materializations store data permanently.

2. Step 2: Consider update efficiency

   'Table' rebuilds fully each run, while 'incremental' updates only new or changed rows efficiently.

3. Step 3: Match requirements

   Since you want to avoid full rebuilds and update only new rows, 'incremental' fits best.

4. Final Answer:

   Use 'incremental' materialization because it stores data permanently and updates only new rows -> Option D

5. Quick Check:

   Permanent + efficient updates = incremental [OK]

Hint: Incremental = permanent storage + efficient updates [OK]

Common Mistakes:

• Choosing 'table' and expecting incremental updates
• Picking 'view' which does not store data permanently
• Confusing ephemeral with storage options