Materializations (view, table, incremental, ephemeral) in dbt - Step-by-Step Execution

Concept Flow - Materializations (view, table, incremental, ephemeral)

Start dbt model run

↓

Check materialization type

↓

View

↓

Create/View

↓

Model data ready for next step

dbt runs models by choosing a materialization type, then creates or updates data accordingly.

Execution Sample

dbt

model.sql:
-- materialization: incremental
select * from source_table
where updated_at > (select max(updated_at) from {{ this }})

This incremental model adds only new or updated rows since last run.

Execution Table

Step	Materialization Type	Action	Data Output	Notes
1	View	Create or replace SQL view	Virtual table with fresh data	No data stored, always fresh
2	Table	Create or replace physical table	Full data snapshot stored	Data stored on disk, refreshed fully
3	Incremental	Append or update rows in table	Only new/changed rows added	Efficient for large datasets
4	Ephemeral	No table created, used in SQL compilation	Temporary CTEs in queries	No physical table, fast for small logic
5	End	Model materialized	Data ready for downstream use	Process complete

💡 All materializations complete, data ready for next steps

Variable Tracker

Variable	Start	After Step 1	After Step 2	After Step 3	After Step 4	Final
materialization_type	None	View	Table	Incremental	Ephemeral	Set per model
data_storage	None	Virtual (view)	Physical table	Updated table	None (CTE)	Depends on materialization
data_freshness	Unknown	Always fresh	Snapshot at run	Partial update	Depends on query	Ready for use

Key Moments - 3 Insights

Why does an incremental model not recreate the whole table every run?

Because incremental materialization appends or updates only new or changed rows, as shown in execution_table step 3, making it efficient for large data.

What happens to data in an ephemeral materialization?

No physical table is created; ephemeral models are compiled as CTEs inside other queries, so data exists only during query execution (execution_table step 4).

How is a view different from a table in dbt materializations?

A view is a virtual table that runs the query fresh each time (step 1), while a table stores data physically and is refreshed fully on each run (step 2).

Visual Quiz - 3 Questions

Look at the execution_table, what action does the incremental materialization perform at step 3?

A Create or replace SQL view

B Append or update rows in table

C Create or replace physical table

D Use temporary CTEs in queries

💡 Hint

Check execution_table row with Step 3 under 'Action' column

According to variable_tracker, what is the data freshness state after Step 2 (Table materialization)?

A Always fresh

B Partial update

C Snapshot at run

D Depends on query

💡 Hint

Look at 'data_freshness' row, column 'After Step 2'

If you want no physical table created and only temporary logic in queries, which materialization should you choose?

A Ephemeral

B Incremental

C View

D Table

💡 Hint

Refer to execution_table step 4 and variable_tracker 'data_storage' after Step 4

Concept Snapshot

dbt materializations control how models store data:
- view: virtual table, always fresh
- table: full physical table, replaced each run
- incremental: update/append rows, efficient for big data
- ephemeral: no table, used as CTEs in queries
Choose based on data size and freshness needs.

Full Transcript

In dbt, materializations decide how your data models are stored and updated. Views create virtual tables that run fresh queries every time. Tables store full snapshots of data and replace them fully on each run. Incremental materializations add or update only new or changed rows, saving time on large datasets. Ephemeral models don't create tables but compile as temporary query parts (CTEs) inside other models. This flow helps dbt efficiently manage data transformations depending on your needs.