Edge Cluster Frontend

The Edge Cluster Frontend is the entrypoint for offloading functions from the Device Client in each Edge Cluster. It acts as a load balancer, to handle the redirection to the correct Serverless Runtime running within the particular Edge Cluster where it is located.

Architecture

The Edge Cluster Frontend has 2 components

• A REST API: It handles the execution requests based on functions, parameters and requirements
• A Broker: It queues the execution requests based on the requirements and queues the results of these executions.

When a request is issued

• The token is verified based on the cognit frontend public key
• The requested function and requirements are then read from OpenNebula
• An execution request for the Flavour specified in the requirements is placed on a queue dedicated to the Flavour. Each execution request is tagged with an ID.
• A group of Serverless Runtime machines backing that Flavour listen for requests on that queue. Each machine only handles 1 execution at a time.
• The execution result is then sent to the Results Exchange with the request ID used as a routing key.
• The REST API connection that queued the execution request is listening for that result on the exchange as well.
• The execution result (either successful or not) is then returned to the Device Client

Install

The application was developed with python 3.10. Check the dependencies.

Automated

You can use the Edge Cluster playbook to deploy the application with ansible. This playbook can also deploy it with ssl termination using nginx. An example inventory file is provided.

Manual

It is recommended to create a virtual environment.

Install virtualenv

pip install virtualenv

Create virtualenv

cd /path/to/cognit-frontend-repo
python -m venv venv
source venv/bin/activate
pip install -r requirements.txt

The Application needs to reach

• the OpenNebula Cloud Edge manager endpoints
  • XMLRPC
  • oneflow
• the cognit frontend
• a rabbitmq broker

Configure this at /etc/cognit-edge_cluster_frontend.conf.

Run the application

./src/main.py

It should result in uvicorn starting the web server and logging requests

(venv)  ◰³ venv  ~/P/C/edgecluster-frontend   v1 *  ./src/main.py                                                                                                        53s
INFO:     Started server process [27235]
INFO:     Waiting for application startup.
INFO:     Application startup complete.
INFO:     Uvicorn running on http://0.0.0.0:1339 (Press CTRL+C to quit)
INFO:     127.0.0.1:56536 - "GET /docs HTTP/1.1" 200 OK
INFO:     127.0.0.1:56536 - "GET /openapi.json HTTP/1.1" 200 OK
32
INFO:     127.0.0.1:56548 - "POST /v1/functions/32/execute?app_req_id=28&mode=sync HTTP/1.1" 200 OK

Unload the virtual env after stopping the application

deactivate

Use

The API documentation is available where the api is running, by default at http://localhost:1338/docs.

The App client must previously get an auth token issued by the cognit frontend and send it on the requests header.

Troubleshooting

Set the log_level: info to debug in the configuration. This should help with spotting problems easier.

Issues with cognit frontend

Make sure the cognit frontend is reachable. Otherwise the application will not start

Cannot load public key for biscuit token authentication from http://localhost:1338/v1/public_key
HTTPConnectionPool(host='localhost', port=1338): Max retries exceeded with url: /v1/public_key (Caused by NewConnectionError('<urllib3.connection.HTTPConnection object at 0x10fae9600>: Failed to establish a new connection: [Errno 61] Connection refused'))⏎

The cognit frontend shares a public key that the edge cluster frontend uses to authorize the execution requests. This means, only auth tokens issued by that cognit frontend are trusted by this edge cluster.

Issues with permissions

When requesting an execution, the request should pass an application requirement ID and a function ID. These are the IDs of documents stored in the OpenNebula database. The user requesting the execution needs to be able to read said documents.