app.config block does not generate app.yaml during bundle deploy

[ncf5031]

The bundle resources documentation states:

config — App configuration commands and environment variables. When specified, this configuration is written to an app.yaml file in the source code path during deployment. This allows you to define app configuration directly in the bundle YAML instead of maintaining a separate app.yaml file.

However, when using the config block on an app resource, databricks bundle deploy does not generate an app.yaml file — neither locally in the source_code_path nor in the workspace. The app starts with no environment variables set.

Reproduction
CLI version: v0.292.0

Cloud: Azure Databricks

Bundle target mode: development

databricks.yml

bundle:
 name: my-app
 databricks_cli_version: ">=0.283.0"

variables:
 catalog_name:
 description: Unity Catalog name
 warehouse_id:
 description: SQL Warehouse ID

sync:
 include:
 - backend/static/**
 - backend/app.yaml

resources:
 apps:
 my_app:
 name: my-app
 source_code_path: ./backend
 config:
 command: ["uvicorn", "app:app"]
 env:
 - name: MY_CATALOG
 value: ${var.catalog_name}
 - name: MY_WAREHOUSE
 value: ${var.warehouse_id}
 - name: MY_JOB_ID
 value: ${resources.jobs.my_job.id}

 resources:
 - name: my-job
 job:
 id: ${resources.jobs.my_job.id}
 permission: CAN_MANAGE_RUN
 - name: sql-warehouse
 sql_warehouse:
 id: ${var.warehouse_id}
 permission: CAN_USE

 jobs:
 my_job:
 name: My Job
 tasks:
 - task_key: run
 spark_python_task:
 python_file: backend/jobs/run.py
 source: WORKSPACE
 environment_key: default
 environments:
 - environment_key: default
 spec:
 environment_version: "2"
 dependencies:
 - databricks-sdk==0.78.0

targets:
 dev:
 mode: development
 default: true
 workspace:
 host: https://adb-xxxxx.azuredatabricks.net/
 variables:
 catalog_name:
 default: my_catalog
 warehouse_id:
 default: abc123

No backend/app.yaml file exists, the expectation is that the config block generates it.

Steps

databricks bundle validate -t dev # passes with no errors
databricks bundle deploy -t dev # completes successfully

Expected behavior
After deploy:

1. backend/app.yaml is generated locally with resolved values
2. The file is uploaded to the workspace at the app's source code path
3. The app starts with the configured environment variables

Actual behavior

• No app.yaml is generated locally
• No app.yaml exists in the workspace (workspace export returns "Path doesn't exist")
• The app starts with no environment variables set
• databricks apps get my-app shows no config or effective_config field
• bundle validate and bundle deploy produce no errors or warnings about the config block
• Additional context
• The config block is silently ignored, no error, no warning
• bundle validate accepts the config block without complaint

Additional context

• The config block is silently ignored, no error, no warning
• bundle validate accepts the config block without complaint
• This feature was introduced in CLI 0.283.0 per the resources documentation
• Workaround: maintain a separate app.yaml file manually with valueFrom references

[andrewnester]

The documentation unfortunately is not correct, it's not supposed to generate app.yaml file but instead uses these field from Deploy API https://docs.databricks.com/api/workspace/apps/deploy#env_vars

The reason why you don't see your env variable is that you have to start / run the app after the deploy by running databricks bundle run <my_app> https://docs.databricks.com/aws/en/dev-tools/bundles/apps-tutorial#deploy-the-updated-bundle

Separately, we're working on a feature which will start the app automatically on "bundle deploy" (no need to run "bundle run") #4672

[ffinfo]

I'm facing the same issue. This issue only happens when the compute is not active. If it is active it does the deployment correctly.
As workaround I now do this in my cicd:

databricks bundle validate -t dev # passes with no errors
databricks bundle deploy -t dev # completes successfully
# If the first step is not succes it did bring the compute to 'active'. After that I do the same, second time I brings the app live
# If the compute is already active only the first will run (if successful)
databricks bundle run -t dev <my_app> || databricks bundle run -t dev <my_app>

This is stable for now at our side.

Still it should be nice if #4672 will be implemented so I can remove this workaround.

[andrewnester]

@ffinfo #4672 is merged and will be released in the upcoming CLI release

[andrewnester]

You can now use the following configuration so that apps is automatically started and deployed on the bundle deploy call.
Please note this option is only supported for a new direct deployment mode which will be prefered way to deploy moving forward

resources:
 apps:
 myapp:
 name: my_app
 description: my_app_description
 source_code_path: ./app
 lifecycle:
 started: true