spark sql is faster for the same query the second time it runs, how to make the query time stable?

Question

I\'m using pyspark to query from a collection of parquet files stored on hdfs. However, it seems that the query response time is faster the second time it runs. Below are th