cassandra 之 在spark-shell 中使用 spark cassandra connector 完整案例

1、cassandra 准备

启动cqlsh，

CQLSH_HOST=172.16.163.131 bin/cqlsh

cqlsh>CREATE KEYSPACE productlogs WITH REPLICATION = { 'class' : 'org.apache.cassandra.locator.SimpleStrategy', 'replication_factor': '2' } 

cqlsh>CREATE TABLE productlogs.logs (
    ids uuid,
    app_name text,
    app_version text,
    city text,
    client_time timestamp,
    country text,
    created_at timestamp,
    cs_count int,
    device_id text,
    id int,
    modle_name text,
    province text,
    remote_ip text,
    updated_at timestamp,
    PRIMARY KEY (ids)
)

2、spark cassandra conector jar包

新建空项目，使用sbt,引入connector,打包为spark-cassandra-connector-full.jar，在*.sbt文件中添加如下一行

libraryDependencies += "com.datastax.spark" % "spark-cassandra-connector_2.10" % "1.5.0"

这步的意义在于：官方的connector包没有将依赖打进去，所以，直接使用官方包的时候，需要自己将依赖找出来。不同版本依赖的包及版本也不相同，简单起见，直接打一个full包

3、启动spark-shell

/opt/db/spark-1.5.2-bin-hadoop2.6/bin/spark-shell --master spark://u1:7077  --jars ~/spark-cassandra-connector-full.jar

以下为sparkshell 命令

4、准备数据源：

//可能大多数文档都先stop掉当前sc，再重启一个，其实根本没必要，直接在原有sc上添加cassandra的参数就好
scala>sc.getConf.set("spark.cassandra.connection.host", "172.16.163.131")
//读取HDFS上的数据源
scala>val df = sc.textFile("/data/logs")
//引入需要的命令空间
scala>import org.apache.spark.sql._
scala>import org.apache.spark.sql.types._
scala>import com.datastax.spark.connector._
scala>import java.util.UUID
//定义shcmea
scala>val schema = StructType(
  StructField("ids", StringType, true) ::
    StructField("id", IntegerType, true) ::
    StructField("app_name", StringType, true) ::
    StructField("app_version", StringType, true) ::
    StructField("client_time", TimestampType, true) ::
    StructField("device_id", StringType, true) ::
    StructField("modle_name", StringType, true) ::
    StructField("cs_count", IntegerType, true) ::
    StructField("created_at", TimestampType, true) ::
    StructField("updated_at", TimestampType, true) ::
    StructField("remote_ip", StringType, true) ::
    StructField("country", StringType, true) ::
    StructField("province", StringType, true) ::
   StructField("city", StringType, true) :: Nil)
//指定数据源的schema
scala>val rowRDD = df.map(_.split("\t")).map(p => Row(UUID.randomUUID().toString(), p(0).toInt, p(1), p(2), java.sql.Timestamp.valueOf(p(3)), p(4), p(5), p(6).toInt, java.sql.Timestamp.valueOf(p(7)), java.sql.Timestamp.valueOf(p(8)), p(9), p(10), p(11), p(12)))
scala>val df= sqlContext.createDataFrame(rowRDD, schema)
scala>df.registerTempTable("logs")
//看下结果
scala>sqlContext.sql("select * from logs limit 1").show

如果你足够细心的话，你可能看到在类型为uuid的ids列，我用的是字符串UUID.randomUUID().toString()。为什么呢？其实在spark cassandra connector内部，会进行转换的。见附录1

5、将数据存入cassandra

scala>import org.apache.spark.sql.cassandra._
scala>df.write.format("org.apache.spark.sql.cassandra").options(Map("table" -> "logs", "keyspace" -> "productlogs")).save()

6、取出刚存的数据：

scala>import org.apache.spark.sql.cassandra._
scala>val cdf = sqlContext.read.
  format("org.apache.spark.sql.cassandra").
  options(Map("table" -> "logs", "keyspace" -> "productlogs")).
  load().registerTempTable("logs")
scala>sqlContext.sql("select * from logs_jsut_save limit 1").show

7、cassandra 与spark sql 数据类型对应关系

spark-cassandra-connector/spark-cassandra-connector/src/main/scala/org/apache/spark/sql/cassandra/DataTypeConverter.scala

  private[cassandra] val primitiveTypeMap = Map[connector.types.ColumnType[_], catalystTypes.DataType](
    connector.types.TextType       -> catalystTypes.StringType,
    connector.types.AsciiType      -> catalystTypes.StringType,
    connector.types.VarCharType    -> catalystTypes.StringType,

    connector.types.BooleanType    -> catalystTypes.BooleanType,

    connector.types.IntType        -> catalystTypes.IntegerType,
    connector.types.BigIntType     -> catalystTypes.LongType,
    connector.types.CounterType    -> catalystTypes.LongType,
    connector.types.FloatType      -> catalystTypes.FloatType,
    connector.types.DoubleType     -> catalystTypes.DoubleType,
    connector.types.SmallIntType   -> catalystTypes.ShortType,
    connector.types.TinyIntType    -> catalystTypes.ByteType,

    connector.types.VarIntType     -> catalystTypes.DecimalType(38, 0), // no native arbitrary-size integer type
    connector.types.DecimalType    -> catalystTypes.DecimalType(38, 18),

    connector.types.TimestampType  -> catalystTypes.TimestampType,
    connector.types.InetType       -> catalystTypes.StringType,
    connector.types.UUIDType       -> catalystTypes.StringType,
    connector.types.TimeUUIDType   -> catalystTypes.StringType,
    connector.types.BlobType       -> catalystTypes.BinaryType,
    connector.types.DateType       -> catalystTypes.DateType,
    connector.types.TimeType       -> catalystTypes.LongType
  )

备注：作者在spark-shell下，使用spark-cassandra-conector 主要使用了两个技巧

1、新建空项目，引入spark-cassandra-conector,将依赖包打进来

2、在spark-shell，直接获取conf，然后添加cassandra 连接参数，这样，就可以在默认的sparkcontext、sqlContext：HiveContext上使用，而不需要先sc.stop