Materializations
Overview
Materializations are strategies for persisting dbt models in a warehouse. There are five types of materializations built into dbt. They are:
- table
- view
- incremental
- ephemeral
- materialized view
You can also configure custom materializations in dbt. Custom materializations are a powerful way to extend dbt's functionality to meet your specific needs.
Configuring materializations
By default, dbt models are materialized as "views". Models can be configured with a different materialization by supplying the materialized
configuration parameter as shown below.
# The following dbt_project.yml configures a project that looks like this:
# .
# └── models
# ├── csvs
# │ ├── employees.sql
# │ └── goals.sql
# └── events
# ├── stg_event_log.sql
# └── stg_event_sessions.sql
name: my_project
version: 1.0.0
config-version: 2
models:
my_project:
events:
# materialize all models in models/events as tables
+materialized: table
csvs:
# this is redundant, and does not need to be set
+materialized: view
Alternatively, materializations can be configured directly inside of the model sql files. This can be useful if you are also setting [Performance Optimization] configs for specific models (for example, Redshift specific configurations or BigQuery specific configurations).
{{ config(materialized='table', sort='timestamp', dist='user_id') }}
select *
from ...
Materializations
View
When using the view
materialization, your model is rebuilt as a view on each run, via a create view as
statement.
- Pros: No additional data is stored, views on top of source data will always have the latest records in them.
- Cons: Views that perform a significant transformation, or are stacked on top of other views, are slow to query.
- Advice:
- Generally start with views for your models, and only change to another materialization when you notice performance problems.
- Views are best suited for models that do not do significant transformation, e.g. renaming, or recasting columns.
Table
When using the table
materialization, your model is rebuilt as a table on each run, via a create table as
statement.
- Pros: Tables are fast to query
- Cons:
- Tables can take a long time to rebuild, especially for complex transformations
- New records in underlying source data are not automatically added to the table
- Advice:
- Use the table materialization for any models being queried by BI tools, to give your end user a faster experience
- Also use the table materialization for any slower transformations that are used by many downstream models
Incremental
incremental
models allow dbt to insert or update records into a table since the last time that model was run.
- Pros: You can significantly reduce the build time by just transforming new records
- Cons: Incremental models require extra configuration and are an advanced usage of dbt. Read more about using incremental models here.
- Advice:
- Incremental models are best for event-style data
- Use incremental models when your
dbt run
s are becoming too slow (i.e. don't start with incremental models)
Ephemeral
ephemeral
models are not directly built into the database. Instead, dbt will interpolate the code from an ephemeral model into its dependent models using a common table expression (CTE). You can control the identifier for this CTE using a model alias, but dbt will always prefix the model identifier with __dbt__cte__
.
- Pros:
- You can still write reusable logic
- Ephemeral models can help keep your data warehouse clean by reducing clutter (also consider splitting your models across multiple schemas by using custom schemas).
- Cons:
- You cannot select directly from this model.
- Operations (for example, macros called using
dbt run-operation
cannotref()
ephemeral nodes) - Overuse of ephemeral materialization can also make queries harder to debug.
- Ephemeral materialization doesn't support model contracts.
- Advice: Use the ephemeral materialization for:
- very light-weight transformations that are early on in your DAG
- are only used in one or two downstream models, and
- do not need to be queried directly
Materialized View
The materialized view
materialization allows the creation and maintenance of materialized views in the target database.
Materialized views are a combination of a view and a table, and serve use cases similar to incremental models.
- Pros:
- Materialized views combine the query performance of a table with the data freshness of a view
- Materialized views operate much like incremental materializations, however they are usually able to be refreshed without manual interference on a regular cadence (depending on the database), forgoing the regular dbt batch refresh required with incremental materializations
dbt run
on materialized views corresponds to a code deployment, just like views
- Cons:
- Due to the fact that materialized views are more complex database objects, database platforms tend to have fewer configuration options available; see your database platform's docs for more details
- Materialized views may not be supported by every database platform
- Advice:
- Consider materialized views for use cases where incremental models are sufficient, but you would like the data platform to manage the incremental logic and refresh.
Configuration Change Monitoring
This materialization makes use of the on_configuration_change
config, which aligns with the incremental nature of the namesake database object. This setting tells dbt to attempt to
make configuration changes directly to the object when possible, as opposed to completely recreating
the object to implement the updated configuration. Using dbt-postgres
as an example, indexes can
be dropped and created on the materialized view without the need to recreate the materialized view itself.
Scheduled Refreshes
In the context of a dbt run
command, materialized views should be thought of as similar to views.
For example, a dbt run
command is only needed if there is the potential for a change in configuration or sql;
it's effectively a deploy action.
By contrast, a dbt run
command is needed for a table in the same scenarios AND when the data in the table needs to be updated.
This also holds true for incremental and snapshot models, whose underlying relations are tables.
In the table cases, the scheduling mechanism is either dbt Cloud or your local scheduler;
there is no built-in functionality to automatically refresh the data behind a table.
However, most platforms (Postgres excluded) provide functionality to configure automatically refreshing a materialized view.
Hence, materialized views work similarly to incremental models with the benefit of not needing to run dbt to refresh the data.
This assumes, of course, that auto refresh is turned on and configured in the model.
dbt-snowflake
does not support materialized views, it uses Dynamic Tables instead. For details, refer to Snowflake specific configurations.
Python materializations
Python models support two materializations:
table
incremental
Incremental Python models support all the same incremental strategies as their SQL counterparts. The specific strategies supported depend on your adapter.
Python models can't be materialized as view
or ephemeral
. Python isn't supported for non-model resource types (like tests and snapshots).
For incremental models, like SQL models, you will need to filter incoming tables to only new rows of data:
- Snowpark
- PySpark
import snowflake.snowpark.functions as F
def model(dbt, session):
dbt.config(materialized = "incremental")
df = dbt.ref("upstream_table")
if dbt.is_incremental:
# only new rows compared to max in current table
max_from_this = f"select max(updated_at) from {dbt.this}"
df = df.filter(df.updated_at >= session.sql(max_from_this).collect()[0][0])
# or only rows from the past 3 days
df = df.filter(df.updated_at >= F.dateadd("day", F.lit(-3), F.current_timestamp()))
...
return df
import pyspark.sql.functions as F
def model(dbt, session):
dbt.config(materialized = "incremental")
df = dbt.ref("upstream_table")
if dbt.is_incremental:
# only new rows compared to max in current table
max_from_this = f"select max(updated_at) from {dbt.this}"
df = df.filter(df.updated_at >= session.sql(max_from_this).collect()[0][0])
# or only rows from the past 3 days
df = df.filter(df.updated_at >= F.date_add(F.current_timestamp(), F.lit(-3)))
...
return df
Note: Incremental models are supported on BigQuery/Dataproc for the merge
incremental strategy. The insert_overwrite
strategy is not yet supported.