Question

我使用带有Scala的Spark-jdbc从MS SQL服务器读取数据，我想按指定的列对数据进行分区。我不想手动设置分区列的上限和下限。我可以在此字段中读取某种最大值和最小值并将其设置为上限/下限吗？另外，使用此查询，我想从数据库中读取所有数据。目前，查询机制如下：

def jdbcOptions() = Map[String,String](
    "driver" -> "db.driver",
    "url" -> "db.url",
    "user" -> "db.user",
    "password" -> "db.password",
    "customSchema" -> "db.custom_schema",
    "dbtable" -> "(select * from TestAllData where dayColumn > 'dayValue') as subq",
    "partitionColumn" -> "db.partitionColumn",
    "lowerBound" -> "1",
    "upperBound" -> "30",
    "numPartitions" -> "5"
}

    val dataDF = sparkSession
      .read
      .format("jdbc")
      .options(jdbcOptions())
      .load()

Answer 1

如果dayColumn是数字或日期字段，则可以使用下一个代码检索边界：

def jdbcBoundOptions() = Map[String,String]{
    "driver" -> "db.driver",
    "url" -> "db.url",
    "user" -> "db.user",
    "password" -> "db.password",
    "customSchema" -> "db.custom_schema",
    "dbtable" -> "(select max(db.partitionColumn), min(db.partitionColumn) from TestAllData where dayColumn > 'dayValue') as subq",
    "numPartitions" -> "1"
}

val boundRow = sparkSession
    .read
    .format("jdbc")
    .options(jdbcBoundOptions())
    .load()
    .first()

val maxDay = boundRow.getInt(0)
val mimDay = boundRow.getInt(1)

请注意，numPartitions必须为1，在这种情况下，我们不需要指定分区详细信息，如Spark documentation中所述。

最后，您可以将检索到的边界用于原始查询：

def jdbcOptions() = Map[String,String]{
    "driver" -> "db.driver",
    "url" -> "db.url",
    "user" -> "db.user",
    "password" -> "db.password",
    "customSchema" -> "db.custom_schema",
    "dbtable" -> "(select * from TestAllData where dayColumn > 'dayValue') as subq",
    "partitionColumn" -> "db.partitionColumn",
    "lowerBound" -> minDay.toString,
    "upperBound" -> maxDay.toString,
    "numPartitions" -> "5"
}

计算分区Spark JDBC的上下限

1 个答案: