Single workstation
The simplest deployment runs the GUI, services and experiment runtime on one PC. Start the project launcher; it opens the Operations Console and starts configured services. Closing the last window stops all services cleanly.
Split control and analysis
For larger experiments, run data acquisition and hardware control on the experiment PC, and analysis/GUI on a second machine:
- configure the data and property Hubs to listen on a reachable TCP endpoint;
- connect the analysis clients to that endpoint;
- keep the experiment network behind a firewall or VPN.
The transport layer recovers from malformed frames and slow clients; a slow analysis machine cannot block acquisition.
Remote access
The web console can be reached over SSH tunnelling or a reverse proxy with authentication. Do not expose control endpoints directly to an untrusted network. See the web console README for token and endpoint guidance.
Production checklist
- Pin the Python environment and driver versions.
- Back up
configuration/and the property database. - Monitor disk headroom for HDF5 acquisition.
- Verify device leases and service health after a reboot.
- Run the release precheck and a simulated experiment before a campaign.
- Keep the desktop and web console versions matched.
Updates
Upgrade the framework in a staging environment first. Check the changelog for incompatible changes, regenerate the project if the template changed, and verify that project-local drivers still import. scripts/check.sh contains the same gates used by CI and is a useful post-upgrade smoke test.