Spark RDD 加入 Cassandra Table

Question

我加入了SparkRDD和Cassandra表（查找），但我不能理解一些东西。

1. 将火花从Cassandra表中拉取range_start和range_end之间的所有记录，然后在火花内存中与RDD连接，或者将所有值从RDD下推到Cassandra并在那里执行连接
2. 限制（1）将应用在哪里？（Cassandra或Spark）
3. 无论应用什么限制（1或1000），Spark总是从Cassandra中提取相同数量的记录吗？

以下代码:

//creating dataframe with fields required for join with cassandra table
//and converting same to rdd
val df_for_join = src_df.select(src_df("col1"),src_df("col2"))
val rdd_for_join = df_for_join.rdd

val result_rdd = rdd_for_join
.joinWithCassandraTable("my_keyspace", "my_table"
,selectedColumns = SomeColumns("col1","col2","col3","col4")
,SomeColumns("col1", "col2")
).where("created_at >''range_start'' and created_at<= range_end")
.clusteringOrder(Ascending).limit(1)

Cassandra表格详细信息-

PRIMARY KEY ((col1, col2), created_at) WITH CLUSTERING ORDER BY (created_at ASC)

Answer 1

joinWithCassandra 表从传递的 RDD 中提取分区/主键值，并将它们转换为针对 Cassandra 中分区的单个请求。然后，最重要的是，SCC 可能会应用额外的过滤，例如，您在哪里条件。如果我没记错的话，但我可能是错的，限制不会完全推到 Cassandra - 它仍然可能为每个分区获取限制行。

您始终可以通过执行 result_rdd.toDebugString 来检查连接发生的位置。对于我的代码：

val df_for_join = Seq((2, 5),(5, 2)).toDF("col1", "col2")
val rdd_for_join = df_for_join.rdd

val result_rdd = rdd_for_join
.joinWithCassandraTable("test", "jt"
,selectedColumns = SomeColumns("col1","col2", "v")
,SomeColumns("col1", "col2")
).where("created_at >'2020-03-13T00:00:00Z' and created_at<= '2020-03-14T00:00:00Z'")
.limit(1)

它给出了以下内容：

scala> result_rdd.toDebugString
res7: String =
(2) CassandraJoinRDD[14] at RDD at CassandraRDD.scala:19 []
 |  MapPartitionsRDD[2] at rdd at <console>:45 []
 |  MapPartitionsRDD[1] at rdd at <console>:45 []
 |  ParallelCollectionRDD[0] at rdd at <console>:45 []

而如果您进行“正常”联接，您将获得以下内容：

scala> val rdd1 = sc.parallelize(Seq((2, 5),(5, 2)))
rdd1: org.apache.spark.rdd.RDD[(Int, Int)] = ParallelCollectionRDD[21] at parallelize at <console>:44
scala> val ct = sc.cassandraTable[(Int, Int)]("test", "jt").select("col1", "col2")
ct: com.datastax.spark.connector.rdd.CassandraTableScanRDD[(Int, Int)] = CassandraTableScanRDD[31] at RDD at CassandraRDD.scala:19

scala> rdd1.join(ct)
res15: org.apache.spark.rdd.RDD[(Int, (Int, Int))] = MapPartitionsRDD[34] at join at <console>:49
scala> rdd1.join(ct).toDebugString
res16: String =
(6) MapPartitionsRDD[37] at join at <console>:49 []
 |  MapPartitionsRDD[36] at join at <console>:49 []
 |  CoGroupedRDD[35] at join at <console>:49 []
 +-(3) ParallelCollectionRDD[21] at parallelize at <console>:44 []
 +-(6) CassandraTableScanRDD[31] at RDD at CassandraRDD.scala:19 []

更多信息可在SCC文件的相应章节中找到。