Only have GPU-node attached when used

[DaBeIDS]

Currently the model pipeline on open shift uses GPU and has to be turned on and off manually. It would be better to have a process which triggers a mechanism to get the pod when needed. Here is a possible solution to the current process we aligned on today:

1. Change Dockerfile (install kubectl)
   corporate_data_extraction/data_extractor/code/coordinator/Dockerfile

RUN apt-get install -y kubectl
cp ... (maybe manifest to be added)

2. Job on Open Shift (contains manifest)

-- has to be set-up to create the model-docker (model-pipeline-server-docker)
corporate_data_extraction/data_extractor/code/model_pipeline/Dockerfile

3. entry.sh

change sleep infinity --> return exit code something

4. Change code infer_on_pdf.py
   corporate_data_extraction/data_extractor/code/infer_on_pdf.py Line 172: Something like that (maybe with Python package)

cmd = 'kubectl ...'
try:
os.system(cmd)
except Exception as e:
msg = "Error during kubectl"
return Response(msg, status=500)