This tutorial will help you get started using Apache Spark GraphX with Scala on the MapR
Sandbox.
GraphX is the Apache Spark component for graph-parallel computations, built upon a branch of
mathematics called graph theory. It is a distributed graph processing framework that sits on top of Spark
core.
Overview of some graph concepts
A graph is a mathematical structure used to model relations between objects. A graph is made up of
vertices and edges that connect them. The vertices are the objects and the edges are the relationships
between them.
A directed graph is a graph where the edges have a direction associated with them. An example of a
directed graph is a Twitter follower. User Bob can follow user Carol without implying that user Carol
follows user Bob.
A regular graph is a graph where each vertex has the same number of edges. An example of a regular
graphs is Facebook friends. If Bob is a friend of Carol, then Carol is also a friend of Bob.
Graphx Property Graph
GraphX extends the Spark RDD with a Resilient Distributed Property Graph.
Build a Standalone Spark Application
The property graph is a directed multigraph which can have multiple edges in parallel. Every edge and
vertex has user defined properties associated with it. The parallel edges allow multiple relationships
between the same vertices.
In this activity, you will use GraphX to analyze flight data.
PROPRIETARY AND CONFIDENTIAL INFORMATION
2015 MapR Technologies, Inc. All Rights Reserved.
2
Build a Standalone Spark Application
Scenario
As a starting simple example, we will analyze 3 flights, for each flight we have the following information:
Originating Airport Destination Airport Distance
SFO ORD 1800 miles
ORD DFW 800 miles
DFW SFO 1400 miles
In this scenario, we are going to represent the airports as vertices and routes as edges. For our graph we
will have three vertices, each representing an airport. The Vertices each have an Id and the airport code
as a property:
Vertex Table for Airports
ID Property
1 SFO
2 ORD
3 DFW
The Edges have the Source Id , the Destination Id and the distance as a property.
Edges Table for Routes
SrcId DestId Property
1 2 1800
2 3 800
3 1 1400
PROPRIETARY AND CONFIDENTIAL INFORMATION
2015 MapR Technologies, Inc. All Rights Reserved.
3
Build a Standalone Spark Application
Software
This tutorial will run on the MapR Sandbox, which includes Spark.
You can download the code and data to run these examples from here:
o https://github.com/caroljmcdonald/sparkgraphxexample
The examples in this post can be run in the spark-shell, after launching with the spark-shell
command.
You can also run the code as a standalone application as described in the tutorial on Getting Started
with Spark on MapR Sandbox.
Launch the Spark Interactive Shell
Log into the MapR Sandbox, as explained in Getting Started with Spark on MapR Sandbox, using userid
user01, password mapr. Start the spark shell with:
$ spark-shell
Define Vertices
First we will import the GraphX packages.
(In the code boxes, comments are in Green and output is in Blue)
import org.apache.spark._
import org.apache.spark.rdd.RDD
// import classes required for using GraphX
import org.apache.spark.graphx._
We define airports as vertices. Vertices have an Id and can have properties or attributes associated with
them. Each vertex consists of :
Vertex id Id (Long)
Vertex Property name (String)
PROPRIETARY AND CONFIDENTIAL INFORMATION
2015 MapR Technologies, Inc. All Rights Reserved.
4
Build a Standalone Spark Application
Vertex Table for Airports
ID Property(V)
1 SFO
We define an RDD with the above properties that is then used for the Vertexes .
// create vertices RDD with ID and Name
val vertices=Array((1L, ("SFO")),(2L, ("ORD")),(3L,("DFW")))
val vRDD= sc.parallelize(vertices)
vRDD.take(1)
// Array((1,SFO))
// Defining a default vertex called nowhere
val nowhere = "nowhere"
Define Edges
Edges are the routes between airports. An edge must have a source, a destination, and can have
properties. In our example, an edge consists of :
Edge origin id src (Long)
Edge destination id dest (Long)
Edge Property distance distance (Long)
Edges Table for Routes
srcid destid Property(E)
1 12 1800
We define an RDD with the above properties that is then used for the Edges . The edge RDD has the
form (src id, dest id, distance ).
// create routes RDD with srcid, destid , distance
val edges = Array(Edge(1L,2L,1800),Edge(2L,3L,800),Edge(3L,1L,1400))
PROPRIETARY AND CONFIDENTIAL INFORMATION
2015 MapR Technologies, Inc. All Rights Reserved.
5
Build a Standalone Spark Application
val eRDD= sc.parallelize(edges)
eRDD.take(2)
// Array(Edge(1,2,1800), Edge(2,3,800))
Create Property Graph
To create a graph, you need to have a Vertex RDD, Edge RDD and a Default vertex.
Create a property graph called graph.
// define the graph
val graph = Graph(vRDD,eRDD, nowhere)
// graph vertices
graph.vertices.collect.foreach(println)
// (2,ORD)
// (1,SFO)
// (3,DFW)
// graph edges
graph.edges.collect.foreach(println)
// Edge(1,2,1800)
// Edge(2,3,800)
// Edge(3,1,1400)
1. How many airports are there?
// How many airports?
val numairports = graph.numVertices
// Long = 3
2. How many routes are there?
// How many routes?
val numroutes = graph.numEdges
// Long = 3
3. which routes > 1000 miles distance?
// routes > 1000 miles distance?
PROPRIETARY AND CONFIDENTIAL INFORMATION
2015 MapR Technologies, Inc. All Rights Reserved.
6
Build a Standalone Spark Application
graph.edges.filter { case Edge(src, dst, prop) => prop > 1000 }.collect.foreach(println)
// Edge(1,2,1800)
// Edge(3,1,1400)
4. The EdgeTriplet class extends the Edge class by adding the srcAttr and dstAttr members which
contain the source and destination properties respectively.
// triplets
graph.triplets.take(3).foreach(println)
((1,SFO),(2,ORD),1800)
((2,ORD),(3,DFW),800)
((3,DFW),(1,SFO),1400)
5. Sort and print out the longest distance routes
// print out longest routes
graph.triplets.sortBy(_.attr, ascending=false).map(triplet =>
"Distance " + triplet.attr.toString + " from " + triplet.srcAttr + " to " + triplet.dstAttr +
".").collect.foreach(println)
Distance 1800 from SFO to ORD.
Distance 1400 from DFW to SFO.
Distance 800 from ORD to DFW.
Analyze real Flight data with GraphX
Scenario
Our data is from http://www.transtats.bts.gov/DL_SelectFields.asp?Table_ID=236&DB_Short_Name=On-
Time. We are using flight information for January 2015. For each flight we have the following information:
Field Description Example Value
dOfM(String) Day of month 1
dOfW (String) Day of week 4
PROPRIETARY AND CONFIDENTIAL INFORMATION
2015 MapR Technologies, Inc. All Rights Reserved.
7
Build a Standalone Spark Application
carrier (String) Carrier code AA
tailNum (String) Unique identifier for the plane - tail N787AA
number
flnum(Int) Flight number 21
org_id(String) Origin airport ID 12478
origin(String) Origin Airport Code JFK
dest_id (String) Destination airport ID 12892
dest (String) Destination airport code LAX
crsdeptime(Double) scheduled departure time 900
deptime (Double) actual departure time 855
depdelaymins (Double) departure delay in minutes 0
crsarrtime (Double) scheduled arrival time 1230
arrtime (Double) actual arrival time 1237
arrdelaymins (Double) Arrival delay minutes 7
crselapsedtime Elapsed time 390
(Double)
dist (Int) Distance 2475
In this scenario, we are going to represent the airports as vertices and routes as edges. We are interested
in visualizing airports and routes and would like to see the number of airports that have departures or
arrivals.
You can download the code and data to run these examples from here:
https://github.com/caroljmcdonald/sparkgraphxexample
Log into the MapR Sandbox, as explained in Getting Started with Spark on MapR Sandbox, using userid
user01, password mapr. Copy the sample data files to your sandbox home directory /user/user01 using
scp. Start the spark shell with:
$ spark-shell
Define Vertices
First we will import the GraphX packages.
PROPRIETARY AND CONFIDENTIAL INFORMATION
2015 MapR Technologies, Inc. All Rights Reserved.
8
Build a Standalone Spark Application
(In the code boxes, comments are in Green and output is in Blue)
import org.apache.spark._
import org.apache.spark.rdd.RDD
import org.apache.spark.util.IntParam
// import classes required for using GraphX
import org.apache.spark.graphx._
import org.apache.spark.graphx.util.GraphGenerators
Below we a Scala case classes to define the Flight schema corresponding to the csv data file.
// define the Flight Schema
case class Flight(dofM:String, dofW:String, carrier:String, tailnum:String, flnum:Int, org_id:Long,
origin:String, dest_id:Long, dest:String, crsdeptime:Double, deptime:Double,
depdelaymins:Double, crsarrtime:Double, arrtime:Double,
arrdelay:Double,crselapsedtime:Double,dist:Int)
The function below parses a line from the data file into the Flight class.
// function to parse input into Flight class
def parseFlight(str: String): Flight = {
val line = str.split(",")
Flight(line(0), line(1), line(2), line(3), line(4).toInt, line(5).toLong, line(6), line(7).toLong,
line(8), line(9).toDouble, line(10).toDouble, line(11).toDouble, line(12).toDouble,
line(13).toDouble, line(14).toDouble, line(15).toDouble, line(16).toInt)
}
Below we load the data from the csv file into a Resilient Distributed Dataset (RDD). RDDs can have
transformations and actions, the first() action returns the first element in the RDD.
// load the data into a RDD
val textRDD = sc.textFile("/user/user01/data/rita2014jan.csv")
// MapPartitionsRDD[1] at textFile
// parse the RDD of csv lines into an RDD of flight classes
val flightsRDD = textRDD.map(parseFlight).cache()
We define airports as vertices. Vertices can have properties or attributes associated with them. Each vertex
has the following property:
PROPRIETARY AND CONFIDENTIAL INFORMATION
2015 MapR Technologies, Inc. All Rights Reserved.
9
Build a Standalone Spark Application
Airport name (String)
Vertex Table for Airports
ID Property(V)
10397 ATL
We define an RDD with the above properties that is then used for the Vertexes .
// create airports RDD with ID and Name
val airports = flightsRDD.map(flight => (flight.org_id, flight.origin)).distinct
airports.take(1)
// Array((14057,PDX))
// Defining a default vertex called nowhere
val nowhere = "nowhere"
// Map airport ID to the 3-letter code to use for printlns
val airportMap = airports.map { case ((org_id), name) => (org_id -> name) }.collect.toList.toMap
// Map(13024 -> LMT, 10785 -> BTV,)
Define Edges
Edges are the routes between airports. An edge must have a source, a destination, and can have
properties. In our example, an edge consists of :
Edge origin id src (Long)
Edge destination id dest (Long)
Edge Property distance distance (Long)
Edges Table for Routes
srcid destid Property(E)
14869 14683 1087
PROPRIETARY AND CONFIDENTIAL INFORMATION
2015 MapR Technologies, Inc. All Rights Reserved.
10
Build a Standalone Spark Application
We define an RDD with the above properties that is then used for the Edges . The edge RDD has the
form (src id, dest id, distance ).
// create routes RDD with srcid, destid , distance
val routes = flightsRDD.map(flight => ((flight.org_id, flight.dest_id), flight.dist)).distinctdistinct
routes.take(2)
// Array(((14869,14683),1087), ((14683,14771),1482))
// create edges RDD with srcid, destid , distance
val edges = routes.map {
case ((org_id, dest_id), distance) =>Edge(org_id.toLong, dest_id.toLong, distance) }
edges.take(1)
//Array(Edge(10299,10926,160))
Create Property Graph
To create a graph, you need to have a Vertex RDD, Edge RDD and a Default vertex.
Create a property graph called graph.
// define the graph
val graph = Graph(airports, edges, nowhere)
// graph vertices
graph.vertices.take(2)
Array((10208,AGS), (10268,ALO))
// graph edges
graph.edges.take(2)
Array(Edge(10135,10397,692), Edge(10135,13930,654))
6. How many airports are there?
// How many airports?
val numairports = graph.numVertices
// Long = 301
7. How many routes are there?
PROPRIETARY AND CONFIDENTIAL INFORMATION
2015 MapR Technologies, Inc. All Rights Reserved.
11
Build a Standalone Spark Application
// How many airports?
val numroutes = graph.numEdges
// Long = 4090
8. which routes > 1000 miles distance?
// routes > 1000 miles distance?
graph.edges.filter { case ( Edge(org_id, dest_id,distance))=> distance > 1000}.take(3)
// Array(Edge(10140,10397,1269), Edge(10140,10821,1670), Edge(10140,12264,1628))
9. The EdgeTriplet class extends the Edge class by adding the srcAttr and dstAttr members which
contain the source and destination properties respectively.
// triplets
graph.triplets.take(3).foreach(println)
((10135,ABE),(10397,ATL),692)
((10135,ABE),(13930,ORD),654)
((10140,ABQ),(10397,ATL),1269)
10. Sort and print out the longest distance routes
// Sort and print out the longest distance routes
graph.triplets.sortBy(_.attr, ascending=false).map(triplet =>
"Distance " + triplet.attr.toString + " from " + triplet.srcAttr + " to " + triplet.dstAttr +
".").take(10).foreach(println)
Distance 4983 from JFK to HNL.
Distance 4983 from HNL to JFK.
Distance 4963 from EWR to HNL.
Distance 4963 from HNL to EWR.
Distance 4817 from HNL to IAD.
Distance 4817 from IAD to HNL.
Distance 4502 from ATL to HNL.
Distance 4502 from HNL to ATL.
Distance 4243 from HNL to ORD.
Distance 4243 from ORD to HNL.
PROPRIETARY AND CONFIDENTIAL INFORMATION
2015 MapR Technologies, Inc. All Rights Reserved.
12
Build a Standalone Spark Application
11. compute the highest degree vertex
// Define a reduce operation to compute the highest degree vertex
def max(a: (VertexId, Int), b: (VertexId, Int)): (VertexId, Int) = {
if (a._2 > b._2) a else b
}
val maxInDegree: (VertexId, Int) = graph.inDegrees.reduce(max)
maxInDegree: (org.apache.spark.graphx.VertexId, Int) = (10397,152)
val maxOutDegree: (VertexId, Int) = graph.outDegrees.reduce(max)
maxOutDegree: (org.apache.spark.graphx.VertexId, Int) = (10397,153)
val maxDegrees: (VertexId, Int) = graph.degrees.reduce(max)
val maxDegrees: (VertexId, Int) = graph.degrees.reduce(max)
airportMap(10397)
res70: String = ATL
12. which airport has the most incoming flights?
// get top 3
val maxIncoming = graph.inDegrees.collect.sortWith(_._2 > _._2).map(x => (airportMap(x._1),
x._2)).take(3)
maxIncoming.foreach(println)
(ATL,152)
(ORD,145)
(DFW,143)
// which airport has the most outgoing flights?
val maxout= graph.outDegrees.join(airports).sortBy(_._2._1, ascending=false).take(3)
maxout.foreach(println)
(10397,(153,ATL))
(13930,(146,ORD))
(11298,(143,DFW))
PROPRIETARY AND CONFIDENTIAL INFORMATION
2015 MapR Technologies, Inc. All Rights Reserved.
13
Build a Standalone Spark Application
PageRank
Another GraphX operator is PageRank. which is based on the Google PageRank algorithm.
PageRank measures the importance of each vertex in a graph, by determining which vertexes have the
most edges with other vertexes. In our example we can use PageRank to determine which airports are
the most important, by measuring which airports have the most connections to other airports.
We have to specify the tolerance, which is the measure of convergence.
13. What are the most important airports according to PageRank?
// use pageRank
val ranks = graph.pageRank(0.1).vertices
val impAirports = ranks.join(airports).sortBy(_._2._1, false).map(_._2._2)
impAirports.take(4)
//res6: Array[String] = Array(ATL, ORD, DFW, DEN)
Pregel
Many important graph algorithms are iterative algorithms since properties of vertices depend on
properties of their neighbors, which depend on properties of their neighbors. Pregel is an iterative graph
processing model, developed at Google, which uses a sequence of iterations of message passing
between vertices in a graph. GraphX Implements a Pregel-like bulk-synchronous message-passing API.
With the Pregel implementation in GraphX vertices can only send messages to neighboring vertices.
The Pregel operator is executed in a series of super steps. In each super step:
the vertices receive the sum of their inbound messages from the previous super step,
compute a new value for the vertex property,
send messages to the neighboring vertices in the next super step.
When there are no more messages remaining, the Pregel operator will end the iteration and the final
graph is returned.
PROPRIETARY AND CONFIDENTIAL INFORMATION
2015 MapR Technologies, Inc. All Rights Reserved.
14
Build a Standalone Spark Application
The code below computes the cheapest airfare using Pregel with the following formula to compute airfare.
50 + distance / 20
// starting vertex
val sourceId: VertexId = 13024
// a graph with edges containing airfare cost calculation
val gg = graph.mapEdges(e => 50.toDouble + e.attr.toDouble/20 )
// initialize graph, all vertices except source have distance infinity
val initialGraph = gg.mapVertices((id, _) => if (id == sourceId) 0.0 else Double.PositiveInfinity)
// call pregel on graph
val sssp = initialGraph.pregel(Double.PositiveInfinity)(
// Vertex Program
(id, dist, newDist) => math.min(dist, newDist),
triplet => {
// Send Message
PROPRIETARY AND CONFIDENTIAL INFORMATION
2015 MapR Technologies, Inc. All Rights Reserved.
15
Build a Standalone Spark Application
if (triplet.srcAttr + triplet.attr < triplet.dstAttr) {
Iterator((triplet.dstId, triplet.srcAttr + triplet.attr))
} else {
Iterator.empty
}
},
// Merge Message
(a,b) => math.min(a,b)
)
// routes , lowest flight cost
println(sssp.edges.take(4).mkString("\n"))
Edge(10135,10397,84.6)
Edge(10135,13930,82.7)
Edge(10140,10397,113.45)
Edge(10140,10821,133.5)
// routes with airport codes , lowest flight cost
ssp.edges.map{ case ( Edge(org_id, dest_id,price))=> ( (airportMap(org_id), airportMap(dest_id),
price)) }.takeOrdered(10)(Ordering.by(_._3))
Array((WRG,PSG,51.55), (PSG,WRG,51.55), (CEC,ACV,52.8), (ACV,CEC,52.8), (ORD,MKE,53.35),
(IMT,RHI,53.35), (MKE,ORD,53.35), (RHI,IMT,53.35), (STT,SJU,53.4), (SJU,STT,53.4))
// airports , lowest flight cost
println(sssp.vertices.take(4).mkString("\n"))
(10208,277.79)
(10268,260.7)
(14828,261.65)
(14698,125.25)
// airport codes , sorted lowest flight cost
sssp.vertices.collect.map(x => (airportMap(x._1), x._2)).sortWith(_._2 < _._2)
res21: Array[(String, Double)] = Array(PDX,62.05), (SFO,65.75), (EUG,117.35)
Want to learn more?
Free Spark Training at http://learn.mapr.com
http://spark.apache.org/docs/latest/graphx-programming-guide.html
PROPRIETARY AND CONFIDENTIAL INFORMATION
2015 MapR Technologies, Inc. All Rights Reserved.
16
Build a Standalone Spark Application
http://ampcamp.berkeley.edu/big-data-mini-course/graph-analytics-with-graphx.html
PROPRIETARY AND CONFIDENTIAL INFORMATION
2015 MapR Technologies, Inc. All Rights Reserved.
17
