In the last days, 1012-b has seen a very high number of read requests. It would be good to look into the source of this imbalance. A good way to do so would be to take a sample of reads on 1012-b using nodetool toppartitions.
The source of those requests seems to be internal (production or labs), as traffic increased on both 1012-b and the general internal request metric at the same time: