Skip to content

Latest commit

 

History

History
103 lines (79 loc) · 3.24 KB

README.md

File metadata and controls

103 lines (79 loc) · 3.24 KB
Raw

PAPI

Process Wrapper API. The purpose of this API is to provide a generic submit/status REST/JWT interface, to submit slurm jobs on a remote machine.

The scheduling of slurm jobs is done via ssh. For this the API uses the paramiko python module.

A "job" is seperated into 4 different phases. In each Phase a command on a remote SSH machine is executed.

The command is expected to return an sbatch output with the slurm id of the job and the machine, running the job.

Use Docker to run API

    docker build -t papi .
    docker -p 5000:5000 run papi

Demo of the API with docker

This demo shows the basic functionality of the API without requiring actually submitting real jobs on a SLURM cluster. The following steps will start the API within a container, submit a demo job via the API, store job information within the database and retrieve job status information from the API. The status message for the finished job shows a successful job with results that can be downloaded.

  • Start the Docker container as mentioned above
  • Use curl to submit a demo job:
$ curl --header "Content-Type: application/json" --request POST --data '{"seeding_date": "sub15days", "irrigation": "true", "nutrition_factor": "0.25", "phenology_factor": "1", "debug": "1"}' http://localhost:5000/submit/

{
  "message": 1, 
  "status": "created"
}
  • Use curl to get the job status:
$ curl --header "Content-Type: application/json" --request GET http://localhost:5000/status/1

{
  "message": {
    "message": "{'message': ''}", 
    "result_path": "/dss/dssfs01/pn56go/pn56go-dss-0000/process/UC5/results/test", 
    "result_webdav": "http://lobcder.process-project.eu:32432/lrzcluster/results/test"
  }, 
  "status": "finished"
}

Start API Webserver

python main.py

Deploy to production

To deploy the API to production, use the WSGI-files for apache2 or nginx and configure the appropriate paths within:

activate_this = '<PATH-TO-PYTHON-ENV>/bin/activate_this.py'
sys.path.append('<PATH-TO-SRC-DIR>')

Possible Request Methods

http://localhost:5000/submit/<id>
http://localhost:5000/status/<id>

Test the API

python test.py --id <id> --method [submit|status] [--payload <json>]

API Configration

Implemented in config.py

config['DEBUG'] = True  			#API Debug Flag
config['PORT'] = '5000' 			#API HTTP Port
config['SSH_HOSTNAME'] = 'localhost' 		#SSH Hostname
config['SSH_USERNAME'] = 'user' 		#SSH Username
config['SSH_PASSWORD'] = 'pass' 		#SSH Password
config['STATUS_COMMAND'] = "echo" 		#SSH CMD executed on Status request
config['SUBMIT_COMMAND'] = "echo" 		#SSH CMD executed on submit request
config['SUBMIT_FIELDS_REQUIRED'] = ['test'] 	#Mandatory Payload fields for submit requests

Use API for different Use Cases / Applications

This API is written for UC5 of PROCESS. It is however easily adoptable for different use cases, that need a proxy API as an intermediary between job submission systems.

To adapt this API for different scenarios, first configure the API in config.py. Second, review the application logic of submission phases within papi/handler/submit.py and papi/handler/status.py. Finally configure cluster/wrapper.sh for the actual job submission on the cluster and copy it to a login-node on your cluster.