topic Re: Passing parameters in Databricks workflows in Machine Learning

Passing parameters in Databricks workflows

argl1995dbks — Tue, 23 Jul 2024 04:18:42 GMT

Hi Databricks, we have created several Databricks workflows and the `json-definition.json` for the same is stored inside version control i.e. GitHub. There are several parameters which are referred from params.json inside this job definition but the issue is the params.json is hardcoded as of current design.

We are trying to figure out a way to pass config parameters to the Controller workflow without using hotfix branch / Databricks UI with some approvals/governance in place. Is there a more better way of doing this so that we don't have to dig inside the params.json to update/modify the existing keys/values.

I am adding the basic tree structure for my MLOps repo:

├── README.md

├── databricks_notebook_jobs

│ ├── CONTROLLER_JOB_CLUSTER

│ │ ├── YOUR_FIRST_NOTEBOOK_NAME.py

│ │ ├── job-definition.json

│ │ └── spark_env_vars

│ │ ├── dev.json

│ │ ├── expl.json

│ │ ├── prod.json

│ │ └── qa.json

│ ├── all_cluster_init.sh

│ ├── default_job_config.json

│ ├── example_module.py

│ └── release.json

├── devops

│ └── params.json

├── requirements.txt

│ └── utils

│ ├── Config.py

│ ├── Config_old.py

│ └── setenv.py

├── test_requirements.txt

├── tests

│ └── test_cleaning_utils.py

└── version.py

Please suggest.

Re: Passing parameters in Databricks workflows

jacovangelder — Tue, 23 Jul 2024 06:42:36 GMT

Have you considered using Databricks Asset Bundles? Very easy to parameterize!

Re: Passing parameters in Databricks workflows

argl1995dbks — Thu, 25 Jul 2024 11:00:46 GMT

Hi @Retired_mod please allow me sometime, I will get back to you.

Re: Passing parameters in Databricks workflows

argl1995dbks — Thu, 25 Jul 2024 11:33:39 GMT

Hi @jacovangelder , I have created a `variables.yaml` file i.e.

variables:

job_cluster_key:

description: Databricks job cluster

default: job_cluster

and while I am trying to refer this variable in my `job.yaml` file like this:

tasks:

- task_key: Sum_Task

job_cluster_key: ${var.job_cluster_key}

spark_python_task:

python_file: ../src/sum.py

and when I tried to validate it using `databricks bundle validate`, I got this error:

**Error: reference does not exist: ${variables.job_cluster_key}**

However, when I added variables in the same job.yaml file it is validated successfully.

Any reasons, why is that?

Re: Passing parameters in Databricks workflows

argl1995dbks — Thu, 25 Jul 2024 11:56:52 GMT

Hi @jacovangelder , I got the solution when you're referring the variables from the same file it should be referred as ${var.varname} and when you're referring from a different file use ${vars.varname}. Though I suspect why is that. Is there any specific reason?

Also I got an error at the time of deploying a bundle while referring the variables from another file, i.e. **A managed resource "vars" "job_cluster_key" has not been declared in the root
module.**