Skip to content

Orchestrate dbt with Treasure Workflow

By KuoHuei (Ansel) LinNovember 01, 2023
post

In this post we are going to focus on dbt Core’s components integration with Treasure Data’s solutions.

What is dbt?

dbt (data build tool) is an open-source analytics engineering tool primarily focused on data transformation within data engineering systems. It allows users to define, execute, and manage data transformations in a structured and repeatable manner. By leveraging SQL-based transformations and a version-controlled workflow, dbt-Core facilitates the organization and maintenance of complex data transformation processes, enabling more efficient analytics and reporting pipelines. It's widely used to prepare data for analytics, modeling, and visualization, making it a crucial tool in modern data pipelines.

Apart from dbt Cloud, which is a SaaS for Data Ops, dbt Core is an open source command line tool and library that enables data teams to transform data using analytics engineering best practices.

More references:

dbt Core, as a CLI tool, with a bunch of SQL files, configuration files for defining the transformation, requires some external components:

  • Databases / Data Warehouses, where the data transformation are executed and stored.

    • Examples include Snowflake, BigQuery, Redshift, and others, allowing you to directly query and transform data stored in these platforms.
    • Here, we are going to use Treasure Data (TD)’s Presto Query Engine as demonstration.

  • Orchestration Tools, where the dbt CLI is executed and scheduled.

    • Some common orchestration tools are Apache Airflow, Prefect, or Dagster.

    • Here, we are going to use Treasure Data (TD)’s Treasure Workflow as demonstration.

Besides, there are still some optional but common external components, like:

  • Version Control Systems, where the configuration and definition of transformation can be versioned, collaborated, and shared as package importing.
  • Data Catalogs and Metadata Management, where the data assets and transformations can be observed, searched, traversed.

Minimum architecture for dbt-core & external components <center>Minimum architecture for dbt-core & external components

Prerequisites

Software Resuirements

(The specified versions above are for reference and verified in following examples.)

Leverage Treasure Data Presto Query Engine for Data Warehousing

Setup & init

  • Create a connection profile and put into dbt_profiles/profiles.yml.
    • Filling your TD API Key into user.
    • Decide a default TD database name into schema.
dbt_project:
  target: td
  outputs:
    td:
      type: trino
      method: none
      user: XXXXXX # TD_API_KEY
      database: td-presto # TD presto schema name, not changable
      host: api-presto.treasuredata.com # TD presto endpoint
      port: 443
      schema: dbt_example_db # TD database name
      threads: 2
      http_scheme: https
      prepared_statements_enabled: false
  • Create the TD database if it has not been created yet. In this example we are calling the database dbt_example_db.
$ td db:create dbt_example_db
  • Initiate a dbt project called dbt_project.
$ dbt init dbt_project --profile dbt_profiles/profiles.yml
  • The folder structure should look like the following:
├── dbt_profiles
│   └── profiles.yml
└── dbt_project
    ├── README.md
    ├── analyses
    ├── dbt_project.yml
    ├── macros
    ├── models
    │   └── example
    │       ├── my_first_dbt_model.sql
    │       ├── my_second_dbt_model.sql
    │       └── schema.yml
    ├── seeds
    ├── snapshots
    └── tests

Tuning Tips

Quoting

Add the following configuration to dbt_project/dbt_project.yml, since TD’s schema name td-presto needs be quoted, otherwise, the syntax error will be reported.

quoting:
  database: true

Disable View & Materialized view

The Treasure Data Presto Query Engine does not support view or materialized view. If you try to use unsupported views the following error displays.

TrinoUserError: Table 'system.metadata.materialized_views' does not exist

To prevent this error, try the following tips:

models:
  dbt_project:
    # Config indicated by + and applies to all files under models/example/
    example:
      +materialized: table
      # +materialized: view ## Don't use view in any cases
  • Adapter patch to bypass fetching view related catalog.
    • Create a macro file dbt_project/macros/td_adapter_patch/adapters.sql.
-- Bypass `materialized_views` to be compatible with TD Presto
-- Related to https://github.com/starburstdata/dbt-trino/issues/298
{% macro trino__list_relations_without_caching(relation) %}
  {% call statement('list_relations_without_caching', fetch_result=True) -%}
    select
      table_catalog as database,
      table_name as name,
      table_schema as schema,
      'table' as table_type
    from {{ relation.information_schema() }}.tables
    where table_schema = '{{ relation.schema | lower }}'
  {% endcall %}
  {{ return(load_result('list_relations_without_caching').table) }}
{% endmacro %}

First run

Once all the configurations are complete, your first dbt run should look like this:

$ dbt run --profiles-dir dbt_profiles/ --project-dir dbt_project

  19:34:57  Running with dbt=1.6.6
  19:34:57  Registered adapter: trino=1.6.2
  19:34:57  Unable to do partial parsing because a project config has changed
  19:34:57  Found 2 models, 4 tests, 0 sources, 0 exposures, 0 metrics, 372 macros, 0 groups, 0 semantic models
  19:34:57
  19:35:01  Concurrency: 2 threads (target='td')
  19:35:01
  19:35:01  1 of 2 START sql table model dbt_example_db.my_first_dbt_model ........ [RUN]
  19:35:06  1 of 2 OK created sql table model dbt_example_db.my_first_dbt_model ... [SUCCESS in 4.81s]
  19:35:06  2 of 2 START sql table model dbt_example_db.my_second_dbt_model ....... [RUN]
  19:35:09  2 of 2 OK created sql table model dbt_example_db.my_second_dbt_model .. [SUCCESS in 3.80s]
  19:35:09
  19:35:09  Finished running 2 table models in 0 hours 0 minutes and 11.92 seconds (11.92s).
  19:35:09
  19:35:09  Completed successfully
  19:35:09
  19:35:09  Done. PASS=2 WARN=0 ERROR=0 SKIP=0 TOTAL=2

The two executed models should show up in the Treasure Data workbench as tables.

Leverage Treasure Workflow for Orchestration

The Treasure Workflow's are an enhanced version of the digdag workflow engine. In order to leverage TD Workflow as the orchestration tool we will need to create a digdag project.

Ordinary digdag Project

  • Create a folder for workflow project called td_wf_project

  • Move both dbt_profiles and dbt_project into td_wf_project folder.

  • Prepare Python Wrapper for invoking dbtRunner in td_wf_project/scripts/dbt_wrapper.py so that digdag engine can use `py>`` operator to invoke dbt CLI library.

    from dbt.cli.main import dbtRunner

def invoke(command_list):
    for cmd in command_list:
        dbtRunner().invoke([
            *(cmd.split()),
            '--project-dir', './dbt_project',
            '--profiles-dir', './dbt_profiles',
        ])

  • Prepare workflow definition in td_wf_project/dbt_exec_workflow.dig.

    
+dbt_invoke:
  py>: scripts.dbt_wrapper.invoke
  _export:
    command_list:
      - run

  • Now the folder structure should look like:

    td_wf_project
├── dbt_exec_workflow.dig
├── dbt_packages
├── dbt_profiles
│   └── profiles.yml
├── dbt_project
│   ├── README.md
│   ├── analyses
│   ├── dbt_project.yml
│   ├── macros
│   │   └── td_adapter_patch
│   │       └── adapters.sql
│   ├── models
│   │   └── example
│   │       ├── my_first_dbt_model.sql
│   │       ├── my_second_dbt_model.sql
│   │       └── schema.yml
│   ├── seeds
│   ├── snapshots
│   └── tests
└── scripts
    └── dbt_wrapper.py

  • Run the workflow (digdag run dbt_exec_workflow.dig) to see results.

    $ digdag run dbt_exec_workflow.dig

...
22:38:09  Finished running 2 table models in 0 hours 0 minutes and 12.71 seconds (12.71s).
22:38:09
22:38:09  Completed successfully
22:38:09
22:38:09  Done. PASS=2 WARN=0 ERROR=0 SKIP=0 TOTAL=2
Success. Task state is saved at .../dbt_in_td_example/td_wf_project/.digdag/status/20231016T000000+0000 directory.
  * Use --session <daily | hourly | "yyyy-MM-dd[ HH:mm:ss]"> to not reuse the last session time.
  * Use --rerun, --start +NAME, or --goal +NAME argument to rerun skipped tasks.

Convert the DigDag project to Treasure Workflow project

The py> operator in Treasure Workflow is containerized as part of the Custom Scripts feature. To enable this in our project we need to add a few dependencies to our project.

  • Add runtime pip install in td_wf_project/scripts/dbt_wrapper.py.

    import os
import sys
os.system(f"{sys.executable} -m pip install dbt-core==1.6.6 dbt-trino==1.6.2")

from dbt.cli.main import dbtRunner

def invoke(command_list):
    for cmd in command_list:
        dbtRunner().invoke([
            *(cmd.split()),
            '--project-dir', './dbt_project',
            '--profiles-dir', './dbt_profiles',
        ])

  • Add docker option in td_wf_project/dbt_exec_workflow.dig.

    +dbt_invoke:
  py>: scripts.dbt_wrapper.invoke
  _export:
    command_list:
      - run
  docker:
    image: "digdag/digdag-python:3.10"

Protect Secrets

Keeping secrets safe is always important. To do this we are going to follow the How to use Secrets in Custom Scripts docs to protect TD API Key.

  • Adding an environment variable in td_wf_project/dbt_exec_workflow.dig.

    +dbt_invoke:
  py>: scripts.dbt_wrapper.invoke
  _export:
    command_list:
      - run
  _env:
    TD_API_KEY: ${secret:td.apikey}
  docker:
    image: "digdag/digdag-python:3.10"

  • Replace the TD API Key, previously set in user td_wf_project/dbt_profiles/profiles.yml with a referring to the environment variable.

          ...
      user: "{{ env_var('TD_API_KEY') }}"
      ...

Deploy to Treasure Workflow

To deploy the workflow to the Treasure Data platform you need to clean up and the push the project.

  • Clean up before encapsulating package.

    $ dbt clean --profiles-dir dbt_profiles/ --project-dir dbt_project

  • Push project to Treasure Workflow.

td workflow push td_wf_project

2023-10-16 16:31:56 -0700: Digdag v0.10.5 Creating .digdag/tmp/archive-3702033545268916293.tar.gz... Archiving dbt_profiles/profiles.yml Archiving dbt_exec_workflow.dig ... Workflows: dbt_exec_workflow.dig Uploaded: id: xxxxxx name: td_wf_project ...

Use td workflow workflows to show all workflows.


- Set secret for Treasure Workflow `td_wf_project` project.

```bash
$ td workflow secrets --project td_wf_project --set td.apikey=XXXXXX

2023-10-16 16:39:58 -0700: Digdag v0.10.5
Secret 'td.apikey' set

  • In Treasure Data Workbench’s Workflows page, search project name td_wf_project, then you should see td.apikey is set.

  • Trigger a “New Run”, the execution results will show in the workflow execution logs.

Misc Options

Some additional things you may want to consider doing with the project include:

  • Add more models or referring other sources in dbt_project.

  • Add scheduling to periodically trigger workflow and dbt run in td_wf_project/dbt_exec_workflow.dig, like:

    timezone: UTC

schedule:
  daily>: 07:00:00

Then simply re-push project by:

$ td workflow push td_wf_project

Next Time

From above steps, a minimum architecture integration is created and deployed. However, there are still many optional augments and tips can greatly empower data operation’s efficiency, maintainability, and observability. Following topics will be covered a future post.

  • Invoke with parameters
  • Version Control & Documentation
  • Observability - log & store results
  • Materialized - window refresh for batch processing pipelines
  • Package management
  • Schema auto evolution

Latest from our blog

Leveraging Query ReExecution for Smooth ...

An unsung hero in an enterprise-grade Hive deployment

READ MORE →

Orchestrate dbt with Treasure Workflow E...

Practices sharing of how Data Team in Treasure Data leverages dbt Core (command line tools) with the Treasure Data ecosystem.

READ MORE →

Upcoming Evolution of Treasure Data Quer...

READ MORE →