README.md

FL Hub (Experimental)

1. Create your FL workspaces and start all FL systems

1.1 Prepare workspaces

cd ./workspaces
for SYSTEM in "t1" "t2a" "t2b"; do
  nvflare provision -p ./${SYSTEM}_project.yml
  cp -r ./workspace/${SYSTEM}_project/prod_00 ./${SYSTEM}_workspace
done
cd ..

1.2 Adjust hub configs

Modify hub clients:

cp -r ./config/site_a/* ./workspaces/t1_workspace/t1_client_a/local/.
cp -r ./config/site_b/* ./workspaces/t1_workspace/t1_client_b/local/.

Modify t2 server configs:

for SYSTEM in "t2a" "t2b"; do
    T2_SERVER_LOCAL=./workspaces/${SYSTEM}_workspace/localhost/local
    mv ${T2_SERVER_LOCAL}/resources.json.default ${T2_SERVER_LOCAL}/resources.json
    sed -i "s|/tmp/nvflare/snapshot-storage|/tmp/nvflare/snapshot-storage_${SYSTEM}|g" ${T2_SERVER_LOCAL}/resources.json
    sed -i "s|/tmp/nvflare/jobs-storage|/tmp/nvflare/hub/jobs/${SYSTEM}|g" ${T2_SERVER_LOCAL}/resources.json
done

Enable only one job at a time:

# move resources file for t1 server
SERVER_LOCAL=./workspaces/t1_workspace/localhost/local
mv ${SERVER_LOCAL}/resources.json.default ${SERVER_LOCAL}/resources.json
# set max_jobs value
for SYSTEM in "t1" "t2a" "t2b"; do
    SERVER_LOCAL=./workspaces/${SYSTEM}_workspace/localhost/local
    sed -i 's|"max_jobs": 4|"max_jobs": 1|g' ${SERVER_LOCAL}/resources.json
done

1.3 Start FL systems

T1 system with 2 clients (a & b)

./start_t1.sh

T2a system with 1 client

./start_t2a.sh

T2b system with 2 clients

./start_t2b.sh

2. Submit job

Open admin for hub. Provide admin username: [email protected]

./workspaces/t1_workspace/[email protected]/startup/fl_admin.sh

Submit job in console. Replace [HUB_EXAMPLE] with your local path of this folder

For a simple example, run

submit_job [HUB_EXAMPLE]/jobs/numpy-cross-val

3. (Optional) Clean-up

Shutdown all FL systems

./shutdown_systems.sh

Note, you can check if nvflare processes are still running with ps -asx | grep flare.

Delete workspaces & temp folders

rm -r workspaces/workspace
rm -r workspaces/*_workspace

rm -r /tmp/nvflare