From that entry it looks like data values are being included within the metric as labels. These need splitting out into their own separate metrics (e.g. NNI_RTT_Min and Packet_Loss). As it stands you are creating new time series for basically every scrape, which results in the lines you are seeing in your graph. Also as it is a cardinality issue it will cause problems with the data storage side of things too.
It is important that labels are never used for data values, and
just contain fixed descriptive entries (such as names, etc.).
If you correct that things should look better.
-- Stuart Clark