Two-node weather collaboration
Node A publishes synthetic observations; node B publishes station climate baselines. Union coordinates streaming daily statistics and anomaly calculation. The client connects only to Union, using its public dataset namespace.
Prepare
Build/stage dacp-server and dacp-union-server with sbt test stage in each module. Build the Python operator runtime as described in Python operators. From the repository root:
bash scripts/certs.sh
.venv/bin/python examples/union/generate_data.py
.venv/bin/python examples/union/prepare.py
The preparation script creates the stable Union identity, node-issued service credentials bound to the Union certificate, and a user token. Secrets are written to protected files, not printed. Account creation issues both DACP and S3 credentials, saved in protected demo-user-account.json. Re-running preparation with the same account fails; retain the generated state and use the admin tools to manage credentials.
Start
Run these commands in separate terminals with the indicated working directories:
# dacp-server
target/universal/stage/bin/dacp-server ../examples/union/node-a.yaml
# dacp-server
target/universal/stage/bin/dacp-server ../examples/union/node-b.yaml
# dacp-union-server
target/universal/stage/bin/dacp-union-server serve ../.union-demo/x.yaml
# Repository root
PYTHONPATH=dacp-client .venv/bin/python examples/union/client.py
The client reads the online catalog, resolves public dataset names and passes the Item URIs as workflow variables. Source inputs are empty; processing inputs reference only upstream output streams. Operators must implement structure inference. Final outputs stream through Union; intermediate outputs stream between nodes. There is no explicit materialization policy in this example.
Open the internal read-only dashboard at http://localhost:19880/. Use Ctrl-C or SIGTERM to stop services. State and execution history persist; running jobs do not resume after restart. See security, workflow variables, policy and replay, and Union configuration.