从spark执行器查询cassandra

Question

I have a streaming app off of kafka, and I was wondering if there was a way to do a range query from inside a map function?

I group the messages from kafka by time range and key, and then based on those time ranges and keys I want to pull data from cassandra into that dstream.

Something like:

lookups
  .map(lookup => ((lookup.key, lookup.startTime, lookup.endTime), lookup))
  .groupByKey()
  .transform(rdd => {
    val cassandraSQLContext = new CassandraSQLContext(rdd.context)
    rdd.map(lookupPair => {
      val tableName = //variable based on lookup
      val startTime = aggLookupPair._1._2
      val endTime = aggLookupPair._1._3

      cassandraSQLContext
        .cassandraSql(s"SELECT * FROM ${CASSANDRA_KEYSPACE}.${tableName} WHERE key=${...} AND start_time >= ${startTime} AND start_time < ${endTime};")
        .map(row => {
           //match to {
            case /*case 1*/ => new object1(row)
            case /*case 2*/ =>new object2(row)
          }
        })
        .collect()
    })
  })

This gives me a null pointer exception:

org.apache.spark.SparkException: Job aborted due to stage failure: Task 0 in stage 59.0 failed 1 times, most recent failure: Lost task 0.0 in stage 59.0 (TID 63, localhost): java.lang.NullPointerException
at org.apache.spark.sql.SQLContext.parseSql(SQLContext.scala:231)
at org.apache.spark.sql.cassandra.CassandraSQLContext.cassandraSql(CassandraSQLContext.scala:70)
at RollupFineGrainIngestionService$$anonfun$11$$anonfun$apply$2.apply(MyFile.scala:130)
at RollupFineGrainIngestionService$$anonfun$11$$anonfun$apply$2.apply(MyFile.scala:123)
at scala.collection.Iterator$$anon$11.next(Iterator.scala:370)
at org.apache.spark.storage.MemoryStore.unrollSafely(MemoryStore.scala:285)
at org.apache.spark.CacheManager.putInBlockManager(CacheManager.scala:171)
at org.apache.spark.CacheManager.getOrCompute(CacheManager.scala:78)
at org.apache.spark.rdd.RDD.iterator(RDD.scala:268)
at org.apache.spark.rdd.MapPartitionsRDD.compute(MapPartitionsRDD.scala:38)
at org.apache.spark.rdd.RDD.computeOrReadCheckpoint(RDD.scala:306)
at org.apache.spark.rdd.RDD.iterator(RDD.scala:270)
at org.apache.spark.rdd.MapPartitionsRDD.compute(MapPartitionsRDD.scala:38)
at org.apache.spark.rdd.RDD.computeOrReadCheckpoint(RDD.scala:306)
at org.apache.spark.rdd.RDD.iterator(RDD.scala:270)

I've also tried to ssc.cassandraTable(CASSANDRA_KEYSPACE, tableName).where("key = ?", ...)... but spark crashes when trying to access StreamingContext inside of a map.

If anyone has any suggestions, I would appreciate it. Thanks!

Answer 1

You may want to use joinWithCassandraTable if your query is based off of a partition key.

But if you need more flexibility

CassandraConnector(sc.getConf).withSessionDo( session => ...)

Will allow you to access the session pool on the executor and execute whatever you want without managing connections. The code is all serializable and can be placed within maps.