Adjusting threadpool queue size to reduce number rof rejected operations

How many indices and shards do you have in the cluster? How much data?