Procesanso datos con Hadoop: MapReduce y YARN

Charla dada bajo el marco de las Jornadas Técnicas Uex-CIEMAT: "Procesando grandes volúmenes de datos con Hadoop"

**NOTA** Algunas transparencias aparecen cortadas en el visor. La versión descargable funciona correctamente.

Transcript

1. Jornadas Técnicas Uex-CIEMAT // 10-12 Febrero 2015 Procesando grandes volúmenes de datos con HADOOP César Suárez Ortega cesar.suarez@externos.ciemat.es Procesando datos con Hadoop: MapReduce y YARN
2. Software Developer / Researcher César Suárez Ortega tharandur csuarez
4. Software Engineer / Researcher César Suárez Ortega
5. Software Engineer / Researcher César Suárez Ortega
6. Índice 1.  2.  3.  z  z  4.  Exprimiendo al máximo MapReduce 5. 
7. Introducción a MapReduce
8. Modelo de programación para el procesamiento de datos
10. ¿Qué es MapReduce? z  z  MUCHOS z  z 
11. by MapReduce: Simplified Data Processing on Large Clusters. Dean J. and Ghemawat S. (2004)!
12. Aprendiendo con ejemplos
13. National Climatic Data Center z  z  z  z  z 
14. 1958 -0021 1959 +0065 1960 +0054
15. Soluciones 1.  Script z  2.  Paralelización por año z  3.  Paralelización por partes iguales z  4.  MapReduce z 
16. MapReduce 101 z  z  <Key, Value> z  Hay que definir una función para cada etapa. MAP REDUCE
17. Map MAP <0, 005733213…> //line1 <160, 006844324…> //line2
18. Map MAP <K1, V1>
19. Map MAP <K1, V1> list(K2, V2)
20. 1958 -0021 1959 +0065 1960 +0054
21. Map MAP <0, 005733213…> //line1 <160, 006844324…> //line2 <1958, -21> <1959, +65>
22. Shuffle <1958, -21> <1959, +65> <1958, -34> <1959, +28> <1958, [-21, -34]> <1959, [+65, +28]>
23. Shuffle list(K2, V2) list(K2, list(V2))
24. Reduce REDUCE list(K2, list(V2)) list(K3, V3)
25. Reduce REDUCE <1958, [-21, -34]> <1959, [+65, +28]> <1958, -21> <1959, +65>
26. Map & Reduce REDUCE list(K2, list(V2)) list(K3, V3)) MAP <K1, V1> list(K2, V2)
27. Map&Reduce z  z  à z  à Job z  z  z 
28. split 0 split 1 split 2 map map map reduce output HDFS replication
29. split 0 split 1 split 2 map map map reduce part0 HDFS replication reduce part1 HDFS replication
30. Práctica 1 Mandando nuestro primer trabajo
31. MapReduce API z  z  1.  Blabla! z  2.  BlablaMapper! z  3.  BlablaReducer! z 
32. FlightsByCarrier z  z  ! ! https://github.com/csuarez/seminario-mapreduce!   flights-by-carrier/  
33. Year,Month,DayofMonth,DayOfWeek,DepTime,CRSDepTime,ArrTime,CRSArrTime,UniqueCarrier,FlightNum,Tai lNum,ActualElapsedTime,CRSElapsedTime,AirTime,ArrDelay,DepDelay,Origin,Dest,Distance,TaxiIn,TaxiO ut,Cancelled,CancellationCode,Diverted,CarrierDelay,WeatherDelay,NASDelay,SecurityDelay,LateAircr aftDelay! 1987,10,14,3,741,730,912,849,PS,1451,NA,91,79,NA,23,11,SAN,SFO,447,NA,NA,0,NA,0,NA,NA,NA,NA,NA! 1987,10,15,4,729,730,903,849,PS,1451,NA,94,79,NA,14,-1,SAN,SFO,447,NA,NA,0,NA,0,NA,NA,NA,NA,NA! 1987,10,17,6,741,730,918,849,PS,1451,NA,97,79,NA,29,11,SAN,SFO,447,NA,NA,0,NA,0,NA,NA,NA,NA,NA! 1987,10,18,7,729,730,847,849,PS,1451,NA,78,79,NA,-2,-1,SAN,SFO,447,NA,NA,0,NA,0,NA,NA,NA,NA,NA! 1987,10,19,1,749,730,922,849,PS,1451,NA,93,79,NA,33,19,SAN,SFO,447,NA,NA,0,NA,0,NA,NA,NA,NA,NA! 1987,10,21,3,728,730,848,849,PS,1451,NA,80,79,NA,-1,-2,SAN,SFO,447,NA,NA,0,NA,0,NA,NA,NA,NA,NA! 1987,10,22,4,728,730,852,849,PS,1451,NA,84,79,NA,3,-2,SAN,SFO,447,NA,NA,0,NA,0,NA,NA,NA,NA,NA! 1987,10,23,5,731,730,902,849,PS,1451,NA,91,79,NA,13,1,SAN,SFO,447,NA,NA,0,NA,0,NA,NA,NA,NA,NA! 1987,10,24,6,744,730,908,849,PS,1451,NA,84,79,NA,19,14,SAN,SFO,447,NA,NA,0,NA,0,NA,NA,NA,NA,NA! 1987,10,25,7,729,730,851,849,PS,1451,NA,82,79,NA,2,-1,SAN,SFO,447,NA,NA,0,NA,0,NA,NA,NA,NA,NA! 1987,10,26,1,735,730,904,849,PS,1451,NA,89,79,NA,15,5,SAN,SFO,447,NA,NA,0,NA,0,NA,NA,NA,NA,NA! 1987,10,28,3,741,725,919,855,PS,1451,NA,98,90,NA,24,16,SAN,SFO,447,NA,NA,0,NA,0,NA,NA,NA,NA,NA! 1987,10,29,4,742,725,906,855,PS,1451,NA,84,90,NA,11,17,SAN,SFO,447,NA,NA,0,NA,0,NA,NA,NA,NA,NA! 1987,10,31,6,726,725,848,855,PS,1451,NA,82,90,NA,-7,1,SAN,SFO,447,NA,NA,0,NA,0,NA,NA,NA,NA,NA! 1987,10,1,4,936,915,1035,1001,PS,1451,NA,59,46,NA,34,21,SFO,RNO,192,NA,NA,0,NA,0,NA,NA,NA,NA,NA! 1987,10,2,5,918,915,1017,1001,PS,1451,NA,59,46,NA,16,3,SFO,RNO,192,NA,NA,0,NA,0,NA,NA,NA,NA,NA! 1987,10,3,6,928,915,1037,1001,PS,1451,NA,69,46,NA,36,13,SFO,RNO,192,NA,NA,0,NA,0,NA,NA,NA,NA,NA! 1987,10,4,7,914,915,1003,1001,PS,1451,NA,49,46,NA,2,-1,SFO,RNO,192,NA,NA,0,NA,0,NA,NA,NA,NA,NA! 1987,10,5,1,1042,915,1129,1001,PS,1451,NA,47,46,NA,88,87,SFO,RNO,192,NA,NA,0,NA,0,NA,NA,NA,NA,NA! 1987,10,6,2,934,915,1024,1001,PS,1451,NA,50,46,NA,23,19,SFO,RNO,192,NA,NA,0,NA,0,NA,NA,NA,NA,NA! 1987,10,7,3,946,915,1037,1001,PS,1451,NA,51,46,NA,36,31,SFO,RNO,192,NA,NA,0,NA,0,NA,NA,NA,NA,NA! 1987,10,8,4,932,915,1033,1001,PS,1451,NA,61,46,NA,32,17,SFO,RNO,192,NA,NA,0,NA,0,NA,NA,NA,NA,NA! 1987,10,9,5,947,915,1036,1001,PS,1451,NA,49,46,NA,35,32,SFO,RNO,192,NA,NA,0,NA,0,NA,NA,NA,NA,NA! 1987,10,10,6,915,915,1022,1001,PS,1451,NA,67,46,NA,21,0,SFO,RNO,192,NA,NA,0,NA,0,NA,NA,NA,NA,NA! 1987,10,11,7,916,915,1006,1001,PS,1451,NA,50,46,NA,5,1,SFO,RNO,192,NA,NA,0,NA,0,NA,NA,NA,NA,NA! 1987,10,12,1,944,915,1027,1001,PS,1451,NA,43,46,NA,26,29,SFO,RNO,192,NA,NA,0,NA,0,NA,NA,NA,NA,NA! 1987,10,13,2,941,915,1036,1001,PS,1451,NA,55,46,NA,35,26,SFO,RNO,192,NA,NA,0,NA,0,NA,NA,NA,NA,NA! 1987,10,14,3,930,915,1029,1001,PS,1451,NA,59,46,NA,28,15,SFO,RNO,192,NA,NA,0,NA,0,NA,NA,NA,NA,NA! 1987,10,15,4,920,915,1023,1001,PS,1451,NA,63,46,NA,22,5,SFO,RNO,192,NA,NA,0,NA,0,NA,NA,NA,NA,NA! 1987,10,17,6,1009,915,1104,1001,PS,1451,NA,55,46,NA,63,54,SFO,RNO,192,NA,NA,0,NA,0,NA,NA,NA,NA,NA! 1987,10,18,7,915,915,1008,1001,PS,1451,NA,53,46,NA,7,0,SFO,RNO,192,NA,NA,0,NA,0,NA,NA,NA,NA,NA! 1987,10,19,1,940,915,1032,1001,PS,1451,NA,52,46,NA,31,25,SFO,RNO,192,NA,NA,0,NA,0,NA,NA,NA,NA,NA! 1987,10,21,3,913,915,1003,1001,PS,1451,NA,50,46,NA,2,-2,SFO,RNO,192,NA,NA,0,NA,0,NA,NA,NA,NA,NA!
34. public class FlightsByCarrier {! public static void main (String[] args) throws Exception {! ! Job job = new Job();! job.setJarByClass(FlightsByCarrier.class); ! ! job.setJobName("FlightsByCarrier”);! ! ! TextInputFormat.addInputPath(job, new Path(args[0])); TextOutputFormat.setOutputPath(job, new Path(args[1]));! job.setInputFormatClass(TextInputFormat.class);! job.setOutputKeyClass(Text.class);! job.setOutputValueClass(IntWritable.class);! ! ! job.setMapperClass(FlightsByCarrierMapper.class);! job.setReducerClass(FlightsByCarrierReducer.class); ! job.setOutputFormatClass(TextOutputFormat.class); ! ! job.addFileToClassPath(new Path("/user/root/opencsv-2.3.jar")); ! job.waitForCompletion(true);! }! }!
35. //Mapper<KeyIn, ValueIn, KeyOut, ValueOut>! public class FlightsByCarrierMapper extends Mapper<LongWritable, Text, Text, IntWritable> {! ! @Override! protected void map(LongWritable key, Text value, Context context) throws IOException, InterruptedException {! ! if (key.get() > 0) { //Ignora la primera linea! String[] lines = new CSVParser().parseLine(value.toString());! ! context.write(new Text(lines[8]), new IntWritable(1));! }! }! }!
36. public class FlightsByCarrierReducer extends Reducer<Text, IntWritable, Text, IntWritable> {! ! @Override! protected void reduce(Text token, Iterable<IntWritable> counts, Context context) throws IOException, InterruptedException {! ! int sum = 0;! ! for (IntWritable count : counts) {! sum += count.get();! }! ! context.write(token, new IntWritable(sum));! }! }! !
37. ! ! $ git clone https://github.com/csuarez/seminario-mapreduce.git! ! [...]! ! $ tar xvzf 1987.tar.gz! ! $ hdfs dfs –copyFromLocal lib/opencsv-2.3.jar /user/root! ! $ hdfs dfs –copyFromLocal 1987.csv /user/root! ! $ sh build.sh! ! $ hadoop jar FlightsByCarrier.jar FlightsByCarrier /user/root/ 1987.csv /user/root/output/flightsCount! ! $ hdfs dfs -cat /user/root/output/flightsCount/part-r-00000! ! ! Ejecución
38. FlightsByCarrier output INFO mapreduce.JobSubmitter: number of splits:2! ! ...! ! INFO mapreduce.Job: map 0% reduce 0%! INFO mapreduce.Job: map 22% reduce 0%! INFO mapreduce.Job: map 41% reduce 0%! INFO mapreduce.Job: map 83% reduce 0%! INFO mapreduce.Job: map 100% reduce 0%! INFO mapreduce.Job: map 100% reduce 100%! ! ...! ! Job Counters! Launched map tasks=2! Launched reduce tasks=1! Rack-local map tasks=2! Total time spent by all maps in occupied slots (ms)=42442! Total time spent by all reduces in occupied slots (ms)=13465! !
39. MapReduce Cómo funciona
40. MapReduce v1
41. MapReduce Program Job Client Job Tracker Task Tracker HDFS Child M/R task TASKTRACKER NODE JVM 1. run 2. get id 3. copy 4. submit job 5. init. job 6. get splits 7. heartbeats 9. launch 10. run 8. get resources JVM CLIENT NODE JOB TRACKER NODE
42. Actores 1.  2.  3.  4.  MapReduce Program Job Client Job Tracker Task Tracker HDFS Child M/R task TASKTRACKER NODE JVM 1. run 2. get id 3. copy 4. submit job 5. init. job 6. get splits 7. heartbeats 9. launch 10. run 8. get resources JVM CLIENT NODE JOB TRACKER NODE
43. Ejecución MapReduce z  z  z  z  z  z  z 
44. Paso 1 Job Submission z  z  z  z  MapReduce Program Job Client Job Tracker Task Tracker HDFS Child M/R task TASKTRACKER NODE JVM 1. run 2. get id 3. copy 4. submit job 5. init. job 6. get splits 7. heartbeats 9. launch 10. run 8. get resources JVM CLIENT NODE JOB TRACKER NODE
45. Paso 2 Job Initialization z  z  z  z  z  MapReduce Program Job Client Job Tracker Task Tracker HDFS Child M/R task TASKTRACKER NODE JVM 1. run 2. get id 3. copy 4. submit job 5. init. job 6. get splits 7. heartbeats 9. launch 10. run 8. get resources JVM CLIENT NODE JOB TRACKER NODE
46. Paso 3 Task Assignment z  z  z  z  MapReduce Program Job Client Job Tracker Task Tracker HDFS Child M/R task TASKTRACKER NODE JVM 1. run 2. get id 3. copy 4. submit job 5. init. job 6. get splits 7. heartbeats 9. launch 10. run 8. get resources JVM CLIENT NODE JOB TRACKER NODE
47. Paso 4 Task Execution z  z  z  MapReduce Program Job Client Job Tracker Task Tracker HDFS Child M/R task TASKTRACKER NODE JVM 1. run 2. get id 3. copy 4. submit job 5. init. job 6. get splits 7. heartbeats 9. launch 10. run 8. get resources JVM CLIENT NODE JOB TRACKER NODE
48. Paso 5 Progress & Status z  z  z  z  MapReduce Program Job Client Job Tracker Task Tracker HDFS Child M/R task TASKTRACKER NODE JVM 1. run 2. get id 3. copy 4. submit job 5. init. job 6. get splits 7. heartbeats 9. launch 10. run 8. get resources JVM CLIENT NODE JOB TRACKER NODE
49. Paso 6 Job Completion z  z  MapReduce Program Job Client Job Tracker Task Tracker HDFS Child M/R task TASKTRACKER NODE JVM 1. run 2. get id 3. copy 4. submit job 5. init. job 6. get splits 7. heartbeats 9. launch 10. run 8. get resources JVM CLIENT NODE JOB TRACKER NODE
50. MapReduce v2 + YARN
51. ¿Por qué YARN? z  z  z  z  z 
52. MapReduce Progran Job Client Resource Manager Node Manager HDFS YARN Child M/R task JVM 1. run JVM CLIENT NODE RESOURCE MANAGER NODE Node Manager Application Master 2. get ID 3. copy job resources 4. submit application 5a. start container 5b. launch 6. init. job 7. get splits 8. allocate resources 9a. start container 9b. launch 10. get data NODE MANAGER NODE NODE MANAGER NODE
53. YARN: Actores 1.  z  2.  z  3.  z  4.  z  5. 
54. vs. MapReduce v1: Actores 1.  2.  3.  4.  MapReduce Progran Job Client Job Tracker Task Tracker HDFS Child M/R task TASKRACKER NODE JVM 1. run 2. get id 3. copy 4. submit job 5. init. job 6. get splits 7. heartbeats 9. launch 10. run 8. get resources JVM CLIENT NODE JOB TRACKER NODE 1.  YARN Resource Manager 2.  YARN Node Manager 3.  Application Master
56. MapReduce v1 API Processing FW Resource Manager Storage MR MapReduce v1   HDFS   PIG   HIVE   HBASE  
57. YARN API Processing FW Resource Manager Storage MR YARN   HDFS   PIG   STORM   MR v2 TEZ   MPI  
58. MapReduce Progran Job Client Resource Manager Node Manager HDFS YARN Child M/R task JVM 1. run JVM CLIENT NODE RESOURCE MANAGER NODE Node Manager Application Master 2. get ID 3. copy job resources 4. submit application 5a. start container 5b. launch 6. init. job 7. get splits 8. allocate resources 9a. start container 9b. launch 10. get data NODE MANAGER NODE NODE MANAGER NODE
59. Paso 1 Job Submission z  z  z  z  z  z 
60. MapReduce Progran Job Client Resource Manager Node Manager HDFS YARN Child M/R task JVM 1. run JVM CLIENT NODE RESOURCE MANAGER NODE Node Manager Application Master 2. get ID 3. copy job resources 4. submit application 5a. start container 5b. launch 6. init. job 7. get splits 8. allocate resources 9a. start container 9b. launch 10. get data NODE MANAGER NODE NODE MANAGER NODE
61. Paso 2 Job Initialization z  z  z  z  z  z  z 
62. Paso 2** Uber Mode z  z  z  z  z 
63. MapReduce Progran Job Client Resource Manager Node Manager HDFS YARN Child M/R task JVM 1. run JVM CLIENT NODE RESOURCE MANAGER NODE Node Manager Application Master 2. get ID 3. copy job resources 4. submit application 5a. start container 5b. launch 6. init. job 7. get splits 8. allocate resources 9a. start container 9b. launch 10. get data NODE MANAGER NODE NODE MANAGER NODE
64. Paso 3 Task Assignment z  z  z  z  z  z 
65. MapReduce Progran Job Client Resource Manager Node Manager HDFS YARN Child M/R task JVM 1. run JVM CLIENT NODE RESOURCE MANAGER NODE Node Manager Application Master 2. get ID 3. copy job resources 4. submit application 5a. start container 5b. launch 6. init. job 7. get splits 8. allocate resources 9a. start container 9b. launch 10. get data NODE MANAGER NODE NODE MANAGER NODE
66. Paso 4 Task Execution z  z  z  z  z 
67. MapReduce Progran Job Client Resource Manager Node Manager HDFS YARN Child M/R task JVM 1. run JVM CLIENT NODE RESOURCE MANAGER NODE Node Manager Application Master 2. get ID 3. copy job resources 4. submit application 5a. start container 5b. launch 6. init. job 7. get splits 8. allocate resources 9a. start container 9b. launch 10. get data NODE MANAGER NODE NODE MANAGER NODE
68. Paso 5 Progress & Status z  z  z  z  z 
69. MapReduce Progran Job Client Resource Manager Node Manager HDFS YARN Child M/R task JVM 1. run JVM CLIENT NODE RESOURCE MANAGER NODE Node Manager Application Master 2. get ID 3. copy job resources 4. submit application 5a. start container 5b. launch 6. init. job 7. get splits 8. allocate resources 9a. start container 9b. launch 10. get data NODE MANAGER NODE NODE MANAGER NODE
70. Exprimiendo MapReduce
71. Apache Pig
72. Apache Pig z  z  z  z  z 
73. records = LOAD '1987.csv' USING PigStorage(',') AS! (Year,Month,DayofMonth,DayOfWeek,DepTime,CRSDepTime,ArrTime,CRSArrTime,UniqueCarrie r,FlightNum,TailNum,ActualElapsedTime,CRSElapsedTime,AirTime,ArrDelay,DepDelay,Orig in,Dest,Distance:int,TaxiIn,TaxiOut,Cancelled,CancellationCode,Diverted,CarrierDela y,WeatherDelay,NASDelay,DecurityDelay,LateAircraftDelay);! ! milage_recs = GROUP records ALL;! ! tot_miles = FOREACH milage_recs GENERATE SUM(records.Distance);! ! STORE tot_miles INTO '/user/root/totalmiles';! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! https://github.com/csuarez/seminario-mapreduce! $ cd pig-total-miles/! $ pig totalmiles.pig!
74. Hadoop Streaming
75. ! ! ! ! ! ! ! ! ! ! ! ! ! ! $ hadoop jar $HADOOP_HOME/hadoop-streaming.jar \! -input myInputDirs \! -output myOutputDir \! -mapper myPythonScript.py \! -reducer /bin/wc \! -file myPythonScript.py!
78. Práctica 2 MapReduce Job
79. GreekGodCounter (Estándar) z  z  ! ! https://github.com/csuarez/seminario-mapreduce!   greek-god-counter-standard/  
80. ! ! ! ! ! private final static String[] gods = {! "Zeus",! "Hera",! "Poseidón",! "Dioniso",! "Apolo",! "Artemisa",! "Hermes",! "Atenea",! "Ares",! "Afrodita",! "Hefesto",! "Deméter”! };! ! ! ! ! ! !
81. //Initializing the initial structure! for (String god : gods) {! godMap.put(god, 0);! }! ! try {! //Reading input! br = new BufferedReader(new FileReader(args[0]));! String line = br.readLine();! while (line != null) {! StringTokenizer tokenizer = new StringTokenizer(line);! while (tokenizer.hasMoreTokens()) {! String token = tokenizer.nextToken();! if (godMap.containsKey(token)) {! godMap.put(token, godMap.get(token) + 1);! }! }! line = br.readLine();! }! ! //Writing output! Writer writer = new BufferedWriter(new FileWriter("gods.txt"));! for (Entry<String, Integer> entry : godMap.entrySet()) {! writer.write(entry.getKey() + " = " + entry.getValue());! writer.write(System.lineSeparator());! }! writer.close();! }!
82. ¡¡¡TAREA!!!
83. GreekGodCounter (MapReduce) z  z  ! ! https://github.com/csuarez/seminario-mapreduce!   greek-god-counter-mapreduce/  
84. import org.apache.hadoop.fs.Path;! import org.apache.hadoop.io.*;! import org.apache.hadoop.mapreduce.Job;! import org.apache.hadoop.mapreduce.lib.input.TextInputFormat;! import org.apache.hadoop.mapreduce.lib.output.TextOutputFormat;! import org.apache.hadoop.filecache.DistributedCache;! ! ! public class GreekGodCounterMapReduce {! public static void main (String[] args) throws Exception {! ! Job job = new Job();! job.setJarByClass(GreekGodCounterMapReduce.class);! job.setJobName("GreekGodCounterMapReduce");! TextInputFormat.addInputPath(job, new Path(args[0])); ! job.setInputFormatClass(TextInputFormat.class);! job.setMapperClass(GreekGodCounterMapReduceMapper.class);! job.setReducerClass(GreekGodCounterMapReduceReducer.class); ! TextOutputFormat.setOutputPath(job, new Path(args[1]));! job.setOutputFormatClass(TextOutputFormat.class);! job.setOutputKeyClass(Text.class);! job.setOutputValueClass(IntWritable.class);! job.waitForCompletion(true);! ! }! }!
85. import java.io.IOException;! import org.apache.hadoop.io.*;! import org.apache.hadoop.mapreduce.Mapper;! import java.util.HashMap;! import java.util.Map.Entry;! import java.util.StringTokenizer;! ! public class GreekGodCounterMapReduceMapper ! extends Mapper<LongWritable, Text, Text, IntWritable> {! private final static String[] gods = {! "Zeus",! "Hera",! "Poseidón",! "Dioniso",! "Apolo",! "Artemisa",! "Hermes",! "Atenea",! "Ares",! "Afrodita",! "Hefesto",! "Deméter"! };! ! @Override! protected void map(LongWritable key, Text value, Context context) throws IOException, InterruptedException {! ! }! }!
86. import java.io.IOException;! import org.apache.hadoop.io.*;! import org.apache.hadoop.mapreduce.Reducer;! ! public class GreekGodCounterMapReduceReducer ! extends Reducer<Text, IntWritable, Text, IntWritable> {! @Override! protected void reduce! (Text token, Iterable<IntWritable> counts, Context context) ! throws IOException, InterruptedException {! ! }! }!
87. Recursos
88. Recursos
89. ¡Gracias! ¿Alguna pregunta? cesar.suarez@externos.ciemat.es