The setup is very similar to an IoT architecture. Remote devices are PIs with CAN interfaces. They collect and forward measurements to a concentrator. This is a one way route by intent. The protocol used is AMQP. Multiple sinks are connected to the AMQP source. A monetdb database used for handling DB request from the node.js application server. Another sink is the range checker which runs independent from the DB. Another stream goes to a development system. Here an Apache Spark system is attached. There was already a function to calculate the correlation between sensors and outputs to detect unwanted hydraulic dependencies. This function moves out of being served of the in-memory monetdb into Apache Spark.
There are thoughts to move to Amazon AWS IoT framework and service for store & forward. Compelling is the device handling and the simplicity of the service.