exaflow

Prerequisites

1. Install python3.10

2. Install poetry It is important to install poetry in isolation, so follow the recommended installation method.

Setup

Environment Setup

1. Install dependencies

   poetry install
   

2. Activate virtual environment

   poetry shell
   

3. Optional To install tab completion for invoke run (replacing bash with your shell)

   source <(poetry run inv --print-completion-script bash)
   

4. Optional pre-commit is included in development dependencies. To install hooks

   pre-commit install
   

Local Deployment

1. Create a deployment configuration file .deployment.toml using the following template

   ip = "172.17.0.1"
   federation = "dementia"
   log_level = "DEBUG"
   framework_log_level ="INFO"
   
   flower_algorithm_folders = "./exaflow/algorithms/flower,./tests/algorithms/flower"
   exareme3_algorithm_folders = "./exaflow/algorithms/exareme3,./tests/algorithms/exareme3"
   
   worker_landscape_aggregator_update_interval = 3000
   worker_tasks_timeout = 20
   
   [flower]
   enabled = true
   execution_timeout = 60
   server_port = 8080
   
   [controller]
   port = 5000
   
   [privacy]
   minimum_row_count = 10
   protect_local_data = false
   
   [aggregation_server]
   enabled = true
   port = 50051
   max_grpc_connections = 10
   max_wait_for_aggregation_inputs = 10
   
   [smpc]
   enabled=false
   optional=false
   get_result_interval = 10
   get_result_max_retries = 100
   smpc_image="gpikra/coordinator:v7.0.7.4"
   db_image="mongo:5.0.8"
   queue_image="redis:alpine3.15"
   [smpc.dp]
   enabled = false
   # sensitivity = 1
   # privacy_budget = 0.1
   
   [[workers]]
   id = "globalworker"
   role = "GLOBALWORKER"
   grpc_port=5670
   
   [[workers]]
   id = "localworker1"
   role = "LOCALWORKER"
   grpc_port=5671
   smpc_client_port=9001
   
   [[workers]]
   id = "localworker2"
   role = "LOCALWORKER"
   grpc_port=5672
   smpc_client_port=9002
   
   

   and then run the following command to create the config files that the worker services will use

   inv create-configs
   

2. Install dependencies, start the containers and then the services with

   inv deploy
   

3. Attach to some service's stdout/stderr with

   inv attach --controller
   

   or

   inv attach --worker <WORKER-NAME>
   

4. Restart all the worker/controller services and keep the same containers with

   inv start-worker --all && inv start-controller --detached
   

Local Deployment (without single configuration file)

1. Create the worker configuration files inside the ./configs/workers/ directory following the ./exaflow/worker/config.toml template.

2. Create the controller config at ./configs/controller/controller.toml using ./exaflow/controller/config.toml as a template, and set the localworkers.config_file field to ./configs/controller/localworkers_config.json.

3. Create the localworkers config file at ./configs/controller/localworkers_config.json with the list of worker ip:grpc_port entries.

4. If aggregation server is enabled, create ./configs/aggregation_server/aggregation_server.toml using ./aggregation_server/config.toml as a template.

Start monitoring tools

1. Start Flower monitoring tool

   by choosing a specific worker to monitor

   inv start-flower --worker <WORKER-NAME>
   

   or start a separate flower instance for all of the workers with

   inv start-flower --all
   

   Then go to the respective address on your browser to start monitoring the workers.

2. Kill all flower instances at any point with

   inv kill-flower
   

Execute an algorithm

• Examples

  ./run_algorithm -a pca -y leftamygdala lefthippocampus -d ppmi0 -m dementia:0.1
  

  ./run_algorithm -a pearson_correlation -y leftamygdala lefthippocampus -d ppmi0 -m dementia:0.1 -p alpha 0.95
  

Acknowledgement

This project/research received funding from the European Union’s Horizon 2020 Framework Programme for Research and Innovation under the Framework Partnership Agreement No. 650003 (HBP FPA).