Development workflows

As you progress from development to production, deploying changes to Materialize requires different workflows. This page provides an overview of best practices and deployment patterns across the different stages you will progress through using dbt as your deployment tool.


When you’re prototyping your use case and fine-tuning the underlying data model, your priority is iteration speed. dbt has many features that can help speed up development, like node selection and model preview. Before you start, we recommend getting familiar with how these features work with the dbt-materialize adapter to make the most of your development time.

Node selection

By default, the dbt-materialize adapter drops and recreates all models on each dbt run invocation. This can have unintended consequences, in particular if you’re managing sources and sinks as models in your dbt project. dbt allows you to selectively run specific models and exclude specific materialization types from each run using node selection.

Exclude sources and sinks

NOTE: As you move towards productionizing your data model, we recommend managing sources and sinks using Terraform instead.

You can manually exclude specific materialization types using the exclude flag in your dbt run invocations. To exclude sources and sinks, use:

dbt run --exclude config.materialized:source config.materialized:sink
YAML selectors

Instead of manually specifying node selection on each run, you can create a YAML selector that makes this the default behavior when running dbt:

# YAML selectors should be defined in a top-level file named selectors.yml
  - name: exclude_sources_and_sinks
    description: >
      Exclude models that use source or sink materializations in the command
    default: true
        # The fqn method combined with the "*" operator selects all nodes in the
        # dbt graph
        - method: fqn
          value: "*"
        - exclude:
            - 'config.materialized:source'
            - 'config.materialized:sink'

Because default: true is specified, dbt will use the selector’s criteria whenever you run an unqualified command (e.g. dbt build, dbt run). You can still override this default by adding selection criteria to commands, or adjust the value of default depending on the target environment. To learn more about using the default and exclude properties with YAML selectors, check the dbt documentation.

Run a subset of models

You can run individual models, or groups of models, using the select flag in your dbt run invocations:

dbt run --select "my_dbt_project_name"   # runs all models in your project
dbt run --select "my_dbt_model"          # runs a specific model
dbt run --select "my_model+"             # select my_model and all downstream dependencies
dbt run --select ""     # runs all models in a specific directory
dbt run --select "my_package.some_model" # runs a specific model in a specific package
dbt run --select "tag:nightly"           # runs models with the "nightly" tag
dbt run --select "path/to/models"        # runs models contained in path/to/models
dbt run --select "path/to/my_model.sql"  # runs a specific model by its path

For a full rundown of selection logic options, check the dbt documentation.

Preview model results

NOTE: The dbt show command uses a LIMIT clause under the hood, which has known performance limitations in Materialize.

To debug and preview the results of your models without materializing the results, you can use the dbt show command:

dbt show --select "model_name.sql"

23:02:20  Running with dbt=1.7.7
23:02:20  Registered adapter: materialize=1.7.3
23:02:20  Found 3 models, 1 test, 4 seeds, 1 source, 0 exposures, 0 metrics, 430 macros, 0 groups, 0 semantic models
23:02:23  Previewing node 'model_name':
| col                  |
| -------------------- |
| value1               |
| value2               |
| value3               |
| value4               |
| value5               |

By default, the dbt show command will return the first 5 rows from the query result (i.e. LIMIT 5). You can adjust the number of rows returned using the --limit n flag.

It’s important to note that previewing results compiles the model and runs the compiled SQL against Materialize; it doesn’t query the already-materialized database relation (see dbt-core #7391).


Once your dbt project is ready to move out of development, or as soon as you start managing multiple users and deployment environments, we recommend checking the code in to version control and setting up an automated workflow to control the deployment of changes.

Depending on the environment context and your deployment requirements, there are two patterns that help optimize and automate your dbt runs: blue/green deployments and slim deployments.

Blue/green deployments

The dbt-materialize adapter ships with helper macros to automate blue/green deployments. We recommend using the blue/green pattern any time you need to deploy changes to the definition of objects in Materialize in production environments and can’t tolerate downtime.

For development environments with no downtime considerations, you might prefer to use the slim deployment pattern instead for quicker iteration and reduced CI costs.

Configuration and initialization

In a blue/green deployment, you first deploy your code changes to a deployment environment (“green”) that is a clone of your production environment (“blue”), in order to validate the changes without causing unavailability. These environments are later swapped transparently.

  1. In dbt_project.yml, use the deployment variable to specify the cluster(s) and schema(s) that contain the changes you want to deploy.

            # To specify multiple clusters, use [<cluster1_name>, <cluster2_name>].
            - <cluster_name>
            # to specify multiple schemas, use [<schema1_name>, <schema2_name>].
            - <schema_name>
  2. Use the run-operation command to invoke the deploy_init macro:

    dbt run-operation deploy_init

    This macro spins up a new cluster named <cluster_name>_dbt_deploy and a new schema named <schema_name>_dbt_deploy using the same configuration as the current environment to swap with (including privileges).

  3. Run the dbt project containing the code changes against the new deployment environment.

    dbt run --vars 'deploy: True'

    The deploy: True variable instructs the adapter to append _dbt_deploy to the original schema or cluster specified for each model scoped for deployment, which transparently handles running that subset of models against the deployment environment.


We strongly recommend validating the results of the deployed changes on the deployment environment to ensure it’s safe to cutover.

  1. After deploying the changes, the objects in the deployment cluster need to fully hydrate before you can safely cut over. Use the run-operation command to invoke the deploy_await macro, which periodically polls the cluster readiness status, and waits for all objects to have a lag of less than 1 second to return successfully.

    dbt run-operation deploy_await

    The default poll interval is 15 seconds, but can be configured to a different value by passing the poll_interval=<value> argument in the macro invocation.

  2. Once deploy_await returns successfully, you can manually run tests against the new deployment environment to validate the results.

Cutover and cleanup

WARNING! To avoid breakages in your production environment, we recommend carefully validating the results of the deployed changes in the deployment environment before cutting over.
  1. Once deploy_await returns successfully and you have validated the results of the deployed changes on the deployment environment, it is safe to push the changes to your production environment.

    Use the run-operation command to invoke the deploy_promote macro, which (atomically) swaps the environments.

    dbt run-operation deploy_promote
    NOTE: The deploy_promote operation might fail if objects are concurrently modified by a different session. If this occurs, re-run the operation.

    This macro ensures all deployment targets, including schemas and clusters, are deployed together as a single atomic operation. If any part of the deployment fails, the entire deployment is rolled back to guarantee consistency and prevent partial updates.

  2. Use the run run-operation command to invoke the deploy_cleanup macro, which (cascade) drops the _dbt_deploy-suffixed cluster(s) and schema(s):

    dbt run-operation deploy_cleanup
    NOTE: Any active SUBSCRIBE commands attached to the swapped cluster(s) will break. On retry, the client will automatically connect to the newly deployed cluster

Slim deployments

On each run, dbt generates artifacts with metadata about your dbt project, including the manifest file (manifest.json). This file contains a complete representation of the latest state of your project, and you can use it to avoid re-deploying resources that didn’t change since the last run.

We recommend using the slim deployment pattern when you want to reduce development idle time and CI costs in development environments. For production deployments, you should prefer the blue/green deployment pattern.

NOTE: Check this demo for a sample end-to-end workflow using GitHub and GitHub Actions.
  1. Fetch the production manifest.json file into the CI environment:

          - name: Download production manifest from s3
              AWS_ACCESS_KEY_ID: ${{ secrets.AWS_ACCESS_KEY_ID }}
              AWS_SECRET_ACCESS_KEY: ${{ secrets.AWS_SECRET_ACCESS_KEY }}
              AWS_SESSION_TOKEN: ${{ secrets.AWS_SESSION_TOKEN }}
              AWS_REGION: us-east-1
            run: |
              aws s3 cp s3://mz-test-dbt/manifest.json ./manifest.json
  2. Then, instruct dbt to run and test changed models and dependencies only:

          - name: Build dbt
              MZ_HOST: ${{ secrets.MZ_HOST }}
              MZ_USER: ${{ secrets.MZ_USER }}
              MZ_PASSWORD: ${{ secrets.MZ_PASSWORD }}
              CI_TAG: "${{ format('{0}_{1}', 'gh_ci', github.event.number ) }}"
            run: |
              source .venv/bin/activate
              dbt run-operation drop_environment
              dbt build --profiles-dir ./ --select state:modified+ --state ./ --target production

    In the example above, --select state:modified+ instructs dbt to run all models that were modified (state:modified) and their downstream dependencies (+). Depending on your deployment requirements, you might want to use a different combination of state selectors, or go a step further and use the --defer flag to reduce even more the number of models that need to be rebuilt. For a full rundown of the available state modifier and graph operator options, check the dbt documentation.

  3. Every time you deploy to production, upload the new manifest.json file to blob storage (e.g. s3):

          - name: upload new manifest to s3
              AWS_ACCESS_KEY_ID: ${{ secrets.AWS_ACCESS_KEY_ID }}
              AWS_SECRET_ACCESS_KEY: ${{ secrets.AWS_SECRET_ACCESS_KEY }}
              AWS_SESSION_TOKEN: ${{ secrets.AWS_SESSION_TOKEN }}
              AWS_REGION: us-east-1
            run: |
              aws s3 cp ./target/manifest.json s3://mz-test-dbt

Other deployment tools

As a tool primarily meant to manage your data model, the dbt-materialize adapter does not expose all Materialize objects types. If there is a clear separation between data modeling and infrastructure management ownership in your team, and you want to manage objects like clusters, connections, or secrets as code, we recommend using the Materialize Terraform provider as a complementary deployment tool.

Back to top ↑