I’ve had the best luck treating the measurement as a namespace instead of an exact identifier, so I would recommend using a general measurement name that applies to all the data with the same shape. The metadata will make good tags and you can look up the exact device using these.
We could use the regex processor to do some processing on the path field, but it isn’t powerful enough to split the data into multiple tags. Is all the data coming in through the tail plugin with one file per device? Is there any flexibility in the how the input data looks?
Additional plugins would be more expensive, mostly in terms of memory use. Each plugin type has different performance characteristics but usually each one is fairly cheap, still I doubt you will want to define thousands if its avoidable.
