为何Spark使用predicates读取JDBC的速度远慢于分区参数方式？

我用Spark的spark.read.jdbc接口读取PostgreSQL表，测试了两种读取方式：

方式1：使用upperBound、lowerBound、numPartitions参数 #

df = spark.read.format("jdbc")\
    .option("url",f"jdbc:postgresql://{host}:{port}/{dbname}")\
    .option("dbtable",dbtable)\
    .option("user",user)\
    .option("password",password)\
    .option("partitionColumn","person_id")\
    .option("numPartitions",2)\
    .option("lowerBound",261461)\
    .option("upperBound",29283487)\
    .load()
df.count()

方式2：使用predicates参数 #

df = spark.read.jdbc(
    url = f"jdbc:postgresql://{host}:{port}/{dbname}",
    table = dbtable,
    predicates = ["(person_id < 14772474 OR person_id is null)", "(person_id >= 14772474)"],
    properties = {"user": f"{user}", "password": f"{password}"}
)
df.count()

我原本预期两种方式执行时间相近，但实际方式1仅耗时10秒，方式2却耗时40秒。查看PostgreSQL日志，两种方式执行的SQL语句完全一致：

LOG:  execute <unnamed>: SELECT 1 FROM xxxxx.person  WHERE "person_id" < 14772474 or "person_id" is null   
LOG:  execute <unnamed>: SELECT 1 FROM xxxxx.person  WHERE "person_id" >= 14772474   

内容的提问来源于stack exchange，提问作者willshen