可以使用hivemetastoreclient检索表的分区数据：

import org.apache.hadoop.hive.conf.HiveConf
import scala.collection.JavaConverters._
import org.apache.hadoop.hive.metastore.HiveMetaStoreClient

val hiveConf = new HiveConf(spark.sparkContext.hadoopConfiguration, classOf[HiveConf])
val cli = new HiveMetaStoreClient(hiveConf)

/* Get list of partition values prior to DF insert */
val existingPartitions = cli.listPartitions("<db_name>", "<tbl_name>", Short.MaxValue).asScala.map(_.getValues.asScala.mkString(","))
/* Insert DF contents to table */
df.write.insertInto("<db_name>.<tbl_name>")
/* Fetch list of partition values again, and diff with previous list */
val newPartitions = cli.listPartitions("<db_name>", "<tbl_name>", Short.MaxValue).asScala.map(_.getValues.asScala.mkString(","))
val deltaPartitions = newPartitions.diff(existingPartitions)

val epochTime = <epoch time before inserting the dataframe>
val partitionName = <Partition Column Name>
df.write.insertInto("<db_name>.<tbl_name>")
val catalogPartitions = spark.sharedState.externalCatalog.listPartitions("<db_name>", "<tbl_name>")
val partitionValues = catalogPartitions.filter(cp => ((cp.parameters.get("transient_lastDdlTime").isDefined && cp.parameters.getOrElse("transient_lastDdlTime", "0").toLong >= epochTime / 1000) || cp.lastAccessTime >= epochTime || cp.createTime >= epochTime) && cp.spec.contains(datePartition)).map(cp => cp.spec.getOrElse(datePartition, "")
      }).toList
processedPartitions = partitionValues.toList

在大多数情况下，lastaccesstime为0。createtime具有创建分隔符的时间。但是在parameters中，我发现了一个新的param.u lastdltime，它包含分区的更新时间戳。更安全的一点是，检查所有三个分区，以获得在给定的epoch时间之后创建或修改的分区。