10.3. Spark Core

geomesa-spark-core is used to work directly with RDDs of features from GeoMesa and other geospatial data stores.

10.3.1. Example

The following is a complete Scala example of creating an RDD via a geospatial query against a GeoMesa data store:

// DataStore params to a hypothetical GeoMesa Accumulo table
val dsParams = Map(
  "accumulo.instance.id"   -> "instance",
  "accumulo.zookeepers"    -> "zoo1,zoo2,zoo3",
  "accumulo.user"          -> "user",
  "accumulo.password"      -> "*****",
  "accumulo.catalog"       -> "geomesa_catalog",
  "geomesa.security.auths" -> "USER,ADMIN")

// set SparkContext
val conf = new SparkConf().setMaster("local[*]").setAppName("testSpark")
val sc = SparkContext.getOrCreate(conf)

// create RDD with a geospatial query using GeoMesa functions
val spatialRDDProvider = GeoMesaSpark(dsParams)
val filter = ECQL.toFilter("CONTAINS(POLYGON((0 0, 0 90, 90 90, 90 0, 0 0)), geom)")
val query = new Query("chicago", filter)
val resultRDD = spatialRDDProvider.rdd(new Configuration, sc, dsParams, query)

resultRDD.collect
// Array[org.opengis.feature.simple.SimpleFeature] = Array(
//    ScalaSimpleFeature:4, ScalaSimpleFeature:5, ScalaSimpleFeature:6,
//    ScalaSimpleFeature:7, ScalaSimpleFeature:9)

10.3.2. Configuration

geomesa-spark-core provides an API for accessing geospatial data in Spark, by defining an interface called SpatialRDDProvider. Different implementations of this interface connect to different input sources. These different providers are described in more detail in Usage below.

GeoMesa provides several JAR-with-dependencies to simplify setting up the Spark classpath. To use these libraries in Spark, the appropriate shaded JAR can be passed (for example) to the spark-submit command via the --jars option:

--jars file://path/to/geomesa-accumulo-spark-runtime_2.11-$VERSION.jar

or passed to Spark via the appropriate mechanism in notebook servers such as Jupyter (see Deploying GeoMesa Spark with Jupyter Notebook) or Zeppelin.

The shaded JAR should also provide the dependencies needed for the Converter RDD Provider and GeoTools RDD Provider, so these JARs may simply be added to --jars as well (though in the latter case additional JARs may be needed to implement the GeoTools data store accessed).

10.3.3. Simple Feature Serialization

To serialize RDDs of SimpleFeatures between nodes of a cluster, Spark must be configured with a Kryo serialization registrator provided in geomesa-spark-core.

Note

Configuring Kryo serialization is not needed when running Spark in local mode, as jobs will be executed within a single JVM.

Add these two entries to $SPARK_HOME/conf/spark-defaults.conf (or pass them as --conf arguments to spark-submit):

spark.serializer        org.apache.spark.serializer.KryoSerializer
spark.kryo.registrator  org.locationtech.geomesa.spark.GeoMesaSparkKryoRegistrator

Note

Alternatively, these may be set in the SparkConf object used to create the SparkContext:

conf.set("spark.serializer", "org.apache.spark.serializer.KryoSerializer")
conf.set("spark.kryo.registrator", classOf[GeoMesaSparkKryoRegistrator].getName)

When using Spark in a notebook server, this will require disabling the automatic creation of a SparkContext.

After setting the configuration options, RDDs created by the GeoMesa SpatialRDDProvider implementations will be properly registered with the serializer provider.

10.3.4. Usage

The main point of entry for the functionality provided by geomesa-spark-core is the GeoMesaSpark object:

val spatialRDDProvider = GeoMesaSpark(params)

GeoMesaSpark loads a SpatialRDDProvider implementation via SPI when the appropriate JAR is included on the classpath. The implementation returned by GeoMesaSpark is chosen based on the parameters passed as an argument, as shown in the Scala code below:

// parameters to pass to the SpatialRDDProvider implementation
val params = Map(
  "param1" -> "foo",
  "param2" -> "bar")
// GeoTools Query; may be used to filter results retrieved from the data store
val query = new Query("foo")
// val query = new Query("foo", ECQL.toFilter("name like 'A%'"))
// get the RDD, using the SparkContext configured as above
val rdd = GeoMesaSpark(params).rdd(new Configuration(), sc, params, query)

To save features, use the save() method:

GeoMesaSpark(params).save(rdd, params, "gdelt")

Note that some providers may be read-only.

See Spatial RDD Providers for details on specific provider implementations.