Databricks Asset Bundles complex variable for cluster configuration substitute

[jonyvp]

Using this page of the DAB docs, I tried to substitute cluster configuration by a variable. That way, I want to predefine different job cluster configurations. Doing exactly what is used in the docs yields this error:


Error: failed to load [...]/specific_job.yml: error unmarshaling JSON: json: cannot unmarshal object into Go struct field Variable.variables.default of type string

[filipniziol]

Hi @jonyvp ,
could you share your .yml?

[jonyvp]

I do an include of the yml in my databricks.yml. The yml pasted below:

variables:
  my_cluster:
    description: "My cluster definition"
    type: complex
    default:
      spark_version: "13.2.x-scala2.11"
      node_type_id: "Standard_DS3_v2"
      num_workers: 2
      spark_conf:
        spark.speculation: true
        spark.databricks.delta.retentionDurationCheck.enabled: false

resources:
  jobs:
    my_job:
      name: "my_job"
      max_concurrent_runs: 1
      job_clusters:
        - job_cluster_key: my_cluster_key
          new_cluster: ${var.my_cluster}
      tasks:
        - task_key: my_job
          spark_python_task:
            python_file: /Workspace/Shared/.bundle/[...]/specific_task.py
          job_cluster_key: my_cluster_key
          libraries:
            - whl: ../dist/*.whl

[filipniziol]

Hi @jonyvp ,

This won't work. The ${var.my_cluster} can substitute just one individual value.
With variable you can substitute specific fields, but you cannot substitue the entire cluster configuration object at once. Thus the error.

Check the article on how to override cluster settings:
Override cluster settings in Databricks Asset Bundles | Databricks on AWS

[jonyvp]

Hi filipniziol, according to this docs, it should be possible?

https://docs.databricks.com/en/dev-tools/bundles/variables.html#define-a-complex-variable

[filipniziol]

@jonyvp , wow, that's something new to me! Thanks for sharing!
The error though is about inability to convert your object to string, so it expects a string.
Are you using the newest databricks cli? Maybe that's a relatively new feature and you need to use the up-to-date cli

[jonyvp · Accepted solution]

@filipniziol that's it! Upgrading the CLI ( 0.209.1 -> 0.228.0 ) did the trick.

[filipniziol]

Amazing! Great!