Production-ready WebSocket IOT System - Software Design Advice

I'm developing a system to control a range of IOT devices. Each set of devices is grouped into a "system" which monitors/controls a real-world process. For example system A may be managing process A and have:

• 3 cameras
• 1 accelerometer
• 1 magnetometer
• 5 thermocouples

The webserver maintains socket connections to each device. Users are able to connect (via a UI - again with websockets) to the webserver and receive updates about systems which they are subscribed to.

When a user wants to begin process A, they should press a 'start' button on the interface. This will start up the cameras, accelerometer, magnetometer, and thermocouples. These will begin sending data to the server. It also triggers the server to set recording mode to true for each device, which means the server will write output to a database. My question:

Should I send a single 'start' request from javascript code in my UI to the server, and allow the server to start each device individually (how do I then handle an error, for example if a single sensor isn't working - what about if two sensors don't work?). Or do I send individual requests from the UI to the server for each device, i.e. start camera 1, start camera 2, start accelerometer, startRecording camera1, etc. and handle each success/error state individually?

My preference throughout the system so far has been the latter approach - one request, one response; with an HTTP error code. However programming becomes more complex when there are many devices to control, for example - System B has 12 thermocouples.

Some components of the system are not vital - e.g. if 1 camera fails we can continue, however, if the accelerometer fails the whole system cannot run and so human monitoring is required. If the server started the devices individually from a single 'start' message, should I return an array of errors, or should the server know which components are vital and return a single error if a vital component fails. And in a failure state, should the server then handle stopping each sensor and returning back to the original state - and what if that then fails? I foresee this code becoming quite complex with this approach.

I've been going back and forth over the best way to approach this for months, but I can't find much advice online around building complex, production-ready IOT systems for the real world. If anybody has any advice or could point me towards any papers/books/etc. I would really appreciate it.

Thanks in advance,

Tom