LP Multicast Tree Example

The examples/lp folder contains the multicast optimization logic used to compute directed multicast trees from a controller topology. The quickest way to run it end to end is the toy harness in examples/lp/toy, which starts one source node and two receivers.

At runtime, the source node waits for receiver handshakes, creates a multicast group, optionally probes link rates through the controller database, solves the LP with mFlow, converts the chosen tree to DAG edges, installs those edges with set_group_routes(), and sends a lossless payload to both receivers.

Run

From the repository root:

cd examples/lp/toy
docker compose up --build

If you want to inspect only the solver output against the toy controller config (without launching containers), run this from the repository root:

uv venv --python 3.13
source .venv/bin/activate
uv pip install numpy cvxopt "psycopg[binary]"

python -m examples.lp.main \
  --controller-config examples/lp/toy/controller-config.toml \
  --src 1 \
  --dests 2,3

Verification

Follow source and receiver logs:

cd examples/lp/toy
docker compose logs -f node1 node2 node3

A successful run should include:

• source: group created, conversion solver=mflow, and send done ok=True
• each receiver: joined group ... and READY, then recv ok=True with the expected payload

You can also confirm group membership state in Postgres:

cd examples/lp/toy
docker compose exec postgres psql -U pgusr -d nextmini -c "SELECT group_id, src_node_id FROM groups;"
docker compose exec postgres psql -U pgusr -d nextmini -c "SELECT group_id, member_node_id FROM group_members ORDER BY member_node_id;"

Cleanup

cd examples/lp/toy
docker compose down --volumes