Probably the first and the last problem are easy to process and there are lots of metrics for that Have a look at the list at the end of the article (the list is from Microsoft website).
Input can be analyzed for network problems by OMS or by checking the IoT hub for the information flow.
Output is easy too as its goes to PowerBi or a Databrick and there its been managed by the other systems.
but what about the inner function of the stream? how long it takes to get the data, run ML on it and send to PowerBI ? Now you can know! new counter with the name Watermark Delay
what is it ?
Simple, the time stream got out – time stream got in = Watermark Delay
if you have problems with your inner functions, for example your ML model taking too long, you can fix it right away and most important, you can create an alert of the stream takes too long. just press the bell of the Configure alerts, set your threshold and your done!