Use dbt to manage Materialize

dbt has become the standard for data transformation (“the T in ELT”). It combines the accessibility of SQL with software engineering best practices, allowing you to not only build reliable data pipelines, but also document, test and version-control them.

In this guide, we’ll cover how to use dbt and Materialize to transform streaming data in real time — from model building to continuous testing.

Setup

Setting up a dbt project with Materialize is similar to setting it up with any other database that requires a non-native adapter. To get up and running, you need to:

1. Install the dbt-materialize plugin (optionally using a virtual environment):

   python3 -m venv dbt-venv         # create the virtual environment
   source dbt-venv/bin/activate     # activate the virtual environment
   pip install dbt-materialize      # install the adapter
   

   The installation will include dbt-core and the dbt-postgres dependency. To check that the plugin was successfully installed, run:

   dbt --version
   

   materialize should be listed under “Plugins”. If this is not the case, double-check that the virtual environment is activated!

2. To get started, make sure you have a Materialize account.

Create and configure a dbt project

A dbt project is a directory that contains all dbt needs to run and keep track of your transformations. At a minimum, it must have a project file (dbt_project.yml) and at least one model (.sql).

To create a new project, run:

dbt init <project_name>

This command will bootstrap a starter project with default configurations and create a profiles.yml file, if it doesn’t exist. To help you get started, the dbt init project includes sample models to run the Materialize quickstart.

Connect to Materialize

dbt manages all your connection configurations (or, profiles) in a file called profiles.yml. By default, this file is located under ~/.dbt/.

1. Locate the profiles.yml file in your machine:

   dbt debug --config-dir
   

   Note: If you started from an existing project but it’s your first time setting up dbt, it’s possible that this file doesn’t exist yet. You can manually create it in the suggested location.

2. Open profiles.yml and adapt it to connect to Materialize using the reference profile configuration.

   As an example, the following profile would allow you to connect to Materialize in two different environments: a developer environment (dev) and a production environment (prod).

   default:
     outputs:
   
       prod:
         type: materialize
         threads: 1
         host: <host>
         port: 6875
         user: <user@domain.com>
         pass: <password>
         database: materialize
         schema: public
         # optionally use the cluster connection
         # parameter to specify the default cluster
         # for the connection
         cluster: <prod_cluster>
         sslmode: require
       dev:
         type: materialize
         threads: 1
         host: <host>
         port: 6875
         user: <user@domain.com>
         pass: <password>
         database: <dev_database>
         schema: <dev_schema>
         cluster: <dev_cluster>
         sslmode: require
   
     target: dev
   

   The target parameter allows you to configure the target environment that dbt will use to run your models.

3. To test the connection to Materialize, run:

   dbt debug
   

   If the output reads All checks passed!, you’re good to go! The dbt documentation has some helpful pointers in case you run into errors.

Build and run dbt models

For dbt to know how to persist (or not) a transformation, the model needs to be associated with a materialization strategy. Because Materialize is optimized for real-time transformations of streaming data and the core of dbt is built around batch, the dbt-materialize adapter implements a few custom materialization types:

Create a materialization for each SQL statement you’re planning to deploy. Each individual materialization should be stored as a .sql file under the directory defined by model-paths in dbt_project.yml.

Sources

In Materialize, a source describes an external system you want to read data from, and provides details about how to decode and interpret that data. You can instruct dbt to create a source using the custom source materialization. Once a source has been defined, it can be referenced from another model using the dbt ref() or source() functions.

{{ config(materialized='source') }}

FROM KAFKA CONNECTION kafka_connection (TOPIC 'topic_a')
FORMAT AVRO USING CONFLUENT SCHEMA REGISTRY CONNECTION csr_connection

database.schema.kafka_topic_a

{{ config(materialized='source') }}

FROM POSTGRES CONNECTION pg_connection (PUBLICATION 'mz_source')
FOR ALL TABLES

sources:
  - name: pg
    schema: "{{ target.schema }}"
    tables:
      - name: table_a
      - name: table_b

-- depends_on: {{ ref('pg') }}
{{ config(materialized='view') }}

SELECT
    table_a.foo AS foo,
    table_b.bar AS bar
FROM {{ source('pg','table_a') }}
INNER JOIN
     {{ source('pg','table_b') }}
    ON table_a.id = table_b.foo_id

database.schema.pg
database.schema.table_a
database.schema.table_b

{{ config(materialized='source') }}

FROM MYSQL CONNECTION mysql_connection
FOR ALL TABLES;

sources:
  - name: mysql
    schema: "{{ target.schema }}"
    tables:
      - name: table_a
      - name: table_b

-- depends_on: {{ ref('mysql') }}
{{ config(materialized='view') }}

SELECT
    table_a.foo AS foo,
    table_b.bar AS bar
FROM {{ source('mysql','table_a') }}
INNER JOIN
     {{ source('mysql','table_b') }}
    ON table_a.id = table_b.foo_id

database.schema.mysql
database.schema.table_a
database.schema.table_b

Views and materialized views

In dbt, a model is a SELECT statement that encapsulates a data transformation you want to run on top of your database.

When you use dbt with Materialize, your models stay up-to-date without manual or configured refreshes. This allows you to efficiently transform streaming data using the same thought process you’d use for batch transformations on top of any other database.

Views

dbt models are materialized as views by default, so to create a view in Materialize you can simply provide the SQL statement in the model (and skip the materialized configuration parameter).

Filename: models/view_a.sql

SELECT
    col_a, ...
FROM {{ ref('kafka_topic_a') }}

The model above would be compiled to database.schema.view_a. One thing to note here is that the model depends on the Kafka source defined above. To express this dependency and track the lineage of your project, you can use the dbt ref() function.

Materialized views

This is where Materialize goes beyond dbt’s incremental models (and traditional databases), with materialized views that continuously update as the underlying data changes:

Filename: models/materialized_view_a.sql

{{ config(materialized='materialized_view') }}

SELECT
    col_a, ...
FROM {{ ref('view_a') }}

The model above would be compiled to database.schema.materialized_view_a. Here, the model depends on the view defined above, and is referenced as such via the dbt ref() function.

Sinks

In Materialize, a sink describes an external system you want to write data to, and provides details about how to encode that data. You can instruct dbt to create a sink using the custom sink materialization.

{{ config(materialized='sink') }}

FROM {{ ref('materialized_view_a') }}
INTO KAFKA CONNECTION kafka_connection (TOPIC 'topic_c')
FORMAT AVRO USING CONFLUENT SCHEMA REGISTRY CONNECTION csr_connection
ENVELOPE DEBEZIUM

database.schema.kafka_topic_c

Configuration: clusters, databases and indexes

Clusters

Use the cluster option to specify the cluster in which a materialized view, index, source, or sink model is created. If unspecified, the default cluster for the connection is used.

{{ config(materialized='materialized_view', cluster='cluster_a') }}

Databases

Use the database option to specify the database in which a source, view, materialized view or sink is created. If unspecified, the default database for the connection is used.

{{ config(materialized='materialized_view', database='database_a') }}

Indexes

Use the indexes option to define a list of indexes on source, view, table or materialized view materializations. Each index option can have the following components:

Creating a multi-column index

{{ config(materialized='view',
          indexes=[{'columns': ['col_a','col_b'], 'cluster': 'cluster_a'}]) }}

Creating a default index

{{ config(materialized='view',
    indexes=[{'default': True}]) }}

Configuration: model contracts and constraints

Model contracts

Minimum requirements: dbt-materialize v1.6.0+

You can enforce model contracts for view, materialized_view and table materializations to guarantee that there are no surprise breakages to your pipelines when the shape of the data changes.

    - name: model_with_contract
    config:
      contract:
        enforced: true
    columns:
      - name: col_with_constraints
        data_type: string
      - name: col_without_constraints
        data_type: int

Setting the contract configuration to enforced: true requires you to specify a name and data_type for every column in your models. If there is a mismatch between the defined contract and the model you’re trying to run, dbt will fail during compilation! Optionally, you can also configure column-level constraints.

Constraints

Minimum requirements: dbt-materialize v1.6.1+

Materialize supports enforcing column-level not_null constraints for materialized_view materializations. No other constraint or materialization types are supported.

    - name: model_with_constraints
    config:
      contract:
        enforced: true
    columns:
      - name: col_with_constraints
        data_type: string
        constraints:
          - type: not_null
      - name: col_without_constraints
        data_type: int

A not_null constraint will be compiled to an ASSERT NOT NULL option for the specified columns of the materialize view.

CREATE MATERIALIZED VIEW model_with_constraints
WITH (
        ASSERT NOT NULL col_with_constraints
     )
AS
SELECT NULL AS col_with_constraints,
       2 AS col_without_constraints;

Build and run dbt

1. Run the dbt models:

   dbt run
   

   This command generates executable SQL code from any model files under the specified directory and runs it in the target environment. You can find the compiled statements under /target/run and target/compiled in the dbt project folder.

2. Using the SQL Shell, or your preferred SQL client connected to Materialize, double-check that all objects have been created:

   SHOW SOURCES [FROM database.schema];
          name
   -------------------
    mysql_table_a
    mysql_table_b
    postgres_table_a
    postgres_table_b
    kafka_topic_a
   
   SHOW VIEWS;
          name
   -------------------
    view_a
   
    SHOW MATERIALIZED VIEWS;
          name
   -------------------
    materialized_view_a
   

That’s it! From here on, Materialize makes sure that your models are incrementally updated as new data streams in, and that you get fresh and correct results with millisecond latency whenever you query your views.

Test and document a dbt project

Configure continuous testing

Using dbt in a streaming context means that you’re able to run data quality and integrity tests non-stop, and monitor failures as soon as they happen. This is useful for unit testing during the development of your dbt models, and later in production to trigger real-time alerts downstream.

1. To configure your project for continuous testing, add a tests property to dbt_project.yml with the store_failures configuration:

   tests:
     dbt_project.name:
       models:
         +store_failures: true
         +schema: 'etl_failure'
   

   This will instruct dbt to create a materialized view for each configured test that can keep track of failures over time. By default, test views are created in a schema suffixed with dbt_test__audit. To specify a custom suffix, use the schema config.

   Note: As an alternative, you can specify the --store-failures flag when running dbt test.

2. Add tests to your models using the tests property in the model configuration .yml files:

   models:
     - name: materialized_view_a
       description: 'materialized view a description'
       columns:
         - name: col_a
           description: 'column a description'
           tests:
             - not_null
             - unique
   

   The type of test and the columns being tested are used as a base for naming the test materialized views. For example, the configuration above would create views named not_null_col_a and unique_col_a.

3. Run the tests:

   dbt test
   

   When configured to store_failures, this command will create a materialized view for each test using the respective SELECT statements, instead of doing a one-off check for failures as part of its execution.

   This guarantees that your tests keep running in the background as views that are automatically updated as soon as an assertion fails.

4. Using the SQL Shell, or your preferred SQL client connected to Materialize, that the schema storing the tests has been created, as well as the test materialized views:

   SHOW SCHEMAS;
          name
   -------------------
    public
    public_etl_failure
   
   SHOW MATERIALIZED VIEWS FROM public_etl_failure;;
          name
   -------------------
    not_null_col_a
    unique_col_a
   

With continuous testing in place, you can then build alerts off of the test materialized views using any common PostgreSQL-compatible client library and SUBSCRIBE(see the Python cheatsheet for a reference implementation).

Generate documentation

dbt can automatically generate documentation for your project as a shareable website. This brings data governance to your streaming pipelines, speeding up life-saving processes like data discovery (where to find what data) and lineage (the path data takes from source (s) to sink(s), as well as the transformations that happen along the way).

If you’ve already created .yml files with helpful properties about your project resources (like model and column descriptions, or tests), you are all set.

1. To generate documentation for your project, run:

   dbt docs generate
   

   dbt will grab any additional project information and Materialize catalog metadata, then compile it into .json files (manifest.json and catalog.json, respectively) that can be used to feed the documentation website. You can find the compiled files under /target, in the dbt project folder.

2. Launch the documentation website. By default, this command starts a web server on port 8000:

   dbt docs serve #--port <port>
   

3. In a browser, navigate to localhost:8000. There, you can find an overview of your dbt project, browse existing models and metadata, and in general keep track of what’s going on.

   If you click View Lineage Graph in the lower right corner, you can even inspect the lineage of your streaming pipelines!

   

Persist documentation

Minimum requirements: dbt-materialize v1.6.1+

To persist model- and column-level descriptions as comments in Materialize, use the persist_docs configuration.

1. To enable docs persistence, add a models property to dbt_project.yml with the persist-docs configuration:

   models:
     +persist_docs:
       relation: true
       columns: true
   

   As an alternative, you can configure persist-docs in the config block of your models:

   {{ config(
       materialized=materialized_view,
       persist_docs={"relation": true, "columns": true}
   ) }}
   

2. Once persist-docs is configured, any description defined in your .yml files is persisted to Materialize in the mz_internal.mz_comments system catalog table on every dbt run:

     SELECT * FROM mz_internal.mz_comments;
   
       id  |    object_type    | object_sub_id |              comment
     ------+-------------------+---------------+----------------------------------
      u622 | materialize-view  |               | materialized view a description
      u626 | materialized-view |             1 | column a description
      u626 | materialized-view |             2 | column b description