[Bug] Add param to defaults from child configs

🐛 Bug

Description

I’m am migrating from 0.11 and not able to resolve this one issue.

Checklist

• [*] I checked on the latest version of Hydra
• [*] I created a minimal repro

To reproduce

** Minimal Code/Config snippet to reproduce ** config.yaml

defaults:
  - model: classification/cnn1d_dilation_rnn_sm
  - data: classification
  - optimizer: adam
  - lr_scheduler: reduce_lr_on_plateau
  - training: gpu

aml:
  experiment_name: 'test'
  experiment_folder: 'experiment_src'

  docker:
    base_image: 'mcr.microsoft.com/azureml/openmpi3.1.2-cuda10.2-cudnn8-ubuntu18.04'

Now in 0.11 for the aml part, I had some extra params that were present in gpu.yaml that got added. here’s gpu.yaml

aml:
    cluster_name: 'gpu-cluster-nc6'
    docker:
        shm_size: '50g'

** Stack trace/error message ** However, after migrating I only see the aml params mentioned in config.yaml instead of combined aml config from both gpu.yaml and config.yaml. Can anyone point to me what I might be missing here?

Expected Behavior

0.11 gives the final output that combines aml from both config.yaml and gpu.yaml

aml:
    experiment_name: 'test'
    experiment_folder: 'experiment_src'
    cluster_name: 'gpu-cluster-nc6'
    docker:
        base_image: 'mcr.microsoft.com/azureml/openmpi3.1.2-cuda10.2-cudnn8-ubuntu18.04'
        shm_size: '50g' 

System information

• Hydra Version : 1.1.0
• Python version : 3.7.9
• Virtual environment type and version : Conda 4.10.1
• Operating system : mac

[omry]

Ok. Not that it matters now, but in general if you skip major Hydra versions when you upgrade you will have a harder time upgrading.

[omry]

Please follow the upgrade instructions here. In particular, Upgrade to Hydra 1.0, address all warnings, and only then upgrade to Hydra 1.1 (and address all warnings).

If you still see issues, please create a minimal runnable example that reproducing them and add it to the task (a zip file for example).