Upgrade to Pro — share decks privately, control downloads, hide ads and more …

Summingbird: Streaming MapReduce at Twitter

Summingbird: Streaming MapReduce at Twitter

Summingbird is a platform for streaming map/reduce used at Twitter to build aggregations in real-time or on hadoop. When the programmer describes her job, that job can be run without change on Storm or Hadoop. Additionally, summingbird can manage merging realtime/online computations with offline batches so that small errors in real-time do not accumulate. Put another way, summingbird gives eventual consistency in a manner that is easy for the programmer to reason about.

Sam Ritchie

June 20, 2013
Tweet

More Decks by Sam Ritchie

Other Decks in Programming

Transcript

  1. def wordCount[P <: Platform[P]]( source: Producer[P, Tweet], store: P#Store[String, Long])

    = source .flatMap { tweet => tweet.getText.split("\\s+").map(_ -> 1L) } .sumByKey(store) Friday, June 21, 13
  2. public class WordCount { public static class Map extends Mapper<LongWritable,

    Text, Text, IntWritable> { private final static IntWritable one = new IntWritable(1); private Text word = new Text(); public void map(LongWritable key, Text value, Context context) throws IOException, InterruptedException { String line = value.toString(); StringTokenizer tokenizer = new StringTokenizer(line); while (tokenizer.hasMoreTokens()) { word.set(tokenizer.nextToken()); context.write(word, one); } } } public static class Reduce extends Reducer<Text, IntWritable, Text, IntWritable> { public void reduce(Text key, Iterable<IntWritable> values, Context context) throws IOException, InterruptedException { int sum = 0; for (IntWritable val : values) { sum += val.get(); } context.write(key, new IntWritable(sum)); } } public static void main(String[] args) throws Exception { Configuration conf = new Configuration(); Job job = new Job(conf, "wordcount"); job.setOutputKeyClass(Text.class); job.setOutputValueClass(IntWritable.class); job.setMapperClass(Map.class); job.setReducerClass(Reduce.class); job.setInputFormatClass(TextInputFormat.class); job.setOutputFormatClass(TextOutputFormat.class); FileInputFormat.addInputPath(job, new Path(args[0])); FileOutputFormat.setOutputPath(job, new Path(args[1])); job.waitForCompletion(true); } } Friday, June 21, 13
  3. public class WordCountTopology { public static class SplitSentence extends ShellBolt

    implements IRichBolt { public SplitSentence() { super("python", "splitsentence.py"); } @Override public void declareOutputFields(OutputFieldsDeclarer declarer) { declarer.declare(new Fields("word")); } @Override public Map<String, Object> getComponentConfiguration() { return null; } } public static class WordCount extends BaseBasicBolt { Map<String, Integer> counts = new HashMap<String, Integer>(); @Override public void execute(Tuple tuple, BasicOutputCollector collector) { String word = tuple.getString(0); Integer count = counts.get(word); if(count==null) count = 0; count++; counts.put(word, count); collector.emit(new Values(word, count)); } @Override public void declareOutputFields(OutputFieldsDeclarer declarer) { declarer.declare(new Fields("word", "count")); } } public static void main(String[] args) throws Exception { TopologyBuilder builder = new TopologyBuilder(); builder.setSpout("spout", new RandomSentenceSpout(), 5); builder.setBolt("split", new SplitSentence(), 8) .shuffleGrouping("spout"); builder.setBolt("count", new WordCount(), 12) .fieldsGrouping("split", new Fields("word")); Config conf = new Config(); conf.setDebug(true); if(args!=null && args.length > 0) { conf.setNumWorkers(3); StormSubmitter.submitTopology(args[0], conf, builder.createTopology()); } else { conf.setMaxTaskParallelism(3); LocalCluster cluster = new LocalCluster(); cluster.submitTopology("word-count", conf, builder.createTopology()); Thread.sleep(10000); cluster.shutdown(); } } } Friday, June 21, 13
  4. def wordCount[P <: Platform[P]]( source: Producer[P, Tweet], store: P#Store[String, Long])

    = source .flatMap { tweet => tweet.getText.split("\\s+").map(_ -> 1L) } .sumByKey(store) // Running in Hadoop (via Scalding): Scalding.run { wordCount[Scalding]( Scalding.source[Tweet]("source_data_location"), Scalding.store[String, Long]("word_count_location") ) } // Running in Storm: Storm.run { wordCount[Storm](new TweetSpout(), new MemcacheStore[String, Long]) } Friday, June 21, 13
  5. def flatMap[T, U](fn: T => List[U]): List[U] def map[T, U](fn:

    T => U): List[U] def filter[T](fn: T => Boolean): List[T] The “Map” of MapReduce Friday, June 21, 13
  6. def wordCount[P <: Platform[P]]( source: Producer[P, Tweet], store: P#Store[String, Long])

    = source .flatMap { tweet => tweet.getText.split("\\s+").map(_ -> 1L) } .sumByKey(store) ”Bears up in my room” (”just” -> 1) (“setting” -> 1) (“up” -> 1) (my” -> 1) (“twttr” -> 1) “Just setting up my twttr” (”Bears” -> 1) (“up” -> 1) (“in” -> 1) (my” -> 1) (“room” -> 1) (“Bears” -> (1)) (”just” -> (1)) (“setting” -> (1)) (“up” -> (1, 1)) (“in” -> (1)) (“my” -> (1, 1)) (“room” -> (1)) (“twttr” -> (1)) P#Store[String, Long] Producer[P, Tweet] (“Bears” -> 1) (”just” -> 1) (“setting” -> 1) (“up” -> 2) (“in” -> 1) (“my” -> 2) (“room” -> 1) (“twttr” -> 1) Friday, June 21, 13
  7. Producer Store Producer[T] T => List[(K, V)] Store[K, V] Client

    flatMap flatMap flatMap flatMap Producer Producer Reduce Reduce (K, List[V]) => (K, V) Friday, June 21, 13
  8. Client Producer Store flatMap flatMap flatMap flatMap Producer Producer Reduce

    Reduce Producer Store flatMap flatMap flatMap flatMap Producer Producer Reduce Reduce Storm Hadoop Friday, June 21, 13
  9. def wordCount[P <: Platform[P]]( source: Producer[P, Tweet], store: P#Store[(Long, String),

    Long]) = source .flatMap { tweet => tweet.getText.split("\\s+") .map { word => ((tweet.getHour, word) -> 1L) } }.sumByKey(store) Key: (Long, String) Value: Long Friday, June 21, 13
  10. 0 2 4 5 7 1 2 3 4 5

    6 7 8 9 10 11 12 13 14 15 16 “ape” by Hour (Hadoop) Friday, June 21, 13
  11. 0 2 4 5 7 1 2 3 4 5

    6 7 8 9 10 11 12 13 14 15 16 “ape” by Hour (Hadoop) Friday, June 21, 13
  12. 0 2 4 5 7 1 2 3 4 5

    6 7 8 9 10 11 12 13 14 15 16 “ape” by Hour (Storm) Friday, June 21, 13
  13. 0 1 3 4 5 1 2 3 4 5

    6 7 8 9 10 11 12 13 14 15 16 “ape” by Hour (Faulty Storm) Friday, June 21, 13
  14. 0 2 4 5 7 1 2 3 4 5

    6 7 8 9 10 11 12 13 14 15 16 “ape” by Hour (Hadoop) “ape” by Hour (Storm) “ape” by Hour (Merged) Friday, June 21, 13
  15. 0 2 4 5 7 1 2 3 4 5

    6 7 8 9 10 11 12 13 14 15 16 “ape” by Hour (Merged) Friday, June 21, 13
  16. Client Producer Store flatMap flatMap flatMap flatMap Producer Producer Reduce

    Reduce Producer Store flatMap flatMap flatMap flatMap Producer Producer Reduce Reduce Storm Hadoop Friday, June 21, 13
  17. a + b + c + d (a + b)

    + (c + d) (a + b + c) + d a + (b + c + d) Friday, June 21, 13
  18. /** * 3 + 3 ==> 6 */ class AddingMonoid[T

    <: Numeric] extends Monoid[T] { override val zero = 0 override def plus(l: T, r: T) = l + r } Friday, June 21, 13
  19. /** * Set(1,2,3) + Set(2,3,4) * ==> Set(1,2,3,4) */ class

    SetMonoid[T] extends Monoid[Set[T]] { override val zero = Set.empty[T] override def plus(l: Set[T], r: Set[T]) = l.concatenate(r) } Friday, June 21, 13
  20. /** * The value type’s monoid is used recursively: *

    * Map("a" -> 1, "b" -> 1) + Map("a" -> 2, "c" -> 1) * ==> Map("a" -> 3, "b" -> 1, "c" -> 1) */ class MapMonoid[K, V](vMonoid: Monoid[V]) extends Monoid[Map[K, V]] { override val zero = Map.empty[K, V] override def plus(x: Map[K, V], y: Map[K, V]) = l.foldLeft(r) { (oldMap, (k, leftV)) => oldMap + (r.get(k) match { case Some(rightV) => vMonoid.plus(leftV, rightV) case None => leftV }) } } Friday, June 21, 13
  21. /** * 3 “+” 3 ==> 9 // wtf? */

    class MultiplyingMonoid[T <: Numeric] extends Monoid[T] { override val zero = 1 override def plus(l: T, r: T) = l * r } Friday, June 21, 13
  22. Client Producer Store flatMap flatMap flatMap flatMap Producer Producer Reduce

    Reduce Producer Store flatMap flatMap flatMap flatMap Producer Producer Reduce Reduce Storm Hadoop Friday, June 21, 13
  23. // Timeline Service (UserID, Hour), List[Tweet] // Impression Counts per

    URL [TweetID, CountMinSketch[URL, ImpCount]] Friday, June 21, 13
  24. Client Producer Store flatMap flatMap flatMap flatMap Producer Producer Reduce

    Reduce Producer Store flatMap flatMap flatMap flatMap Producer Producer Reduce Reduce Storm Hadoop Friday, June 21, 13
  25. trait Bijection[A, B] extends (A => B) { def apply(a:

    A): B def invert(b: B): A } trait Injection[A, B] extends (A => B) { def apply(a: A): B def invert(b: B): Option[A] } Friday, June 21, 13
  26. class LongToBytes extends Injection[Long, Array[Byte]] { val size = 8

    def apply(value: Long) = { val buf = ByteBuffer.allocate(size) buf.putLong(value) buf.array } override def invert(b: Array[Byte]) = try { Some(ByteBuffer.wrap(b).getLong) } catch { case _ => None } } Friday, June 21, 13
  27. val injection = Injection.connect[Long, String, Array[Byte], Base64String] scala> injection.apply(243L) res17:

    com.twitter.bijection.Base64String = Base64String(MjQz) scala> injection.invert(res17) res18: Option[Long] = Some(243) Friday, June 21, 13
  28. class BijectedMonoid[T, U](monoid: Monoid[T], bij: Bijection[T, U]) extends Monoid[U] {

    override def zero: U = monoid.zero.as[U] override def plus(l: U, r: U): U = monoid.plus(l.as[T], r.as[T]).as[U] } class MonoidBijection[T, U](monoid: Monoid[T], bij: Bijection[T, U]) extends Bijection[Monoid[T], Monoid[U]] { override def apply(mon: Monoid[T]) = new BijectedMonoid[T, U](mon, bij) override def invert(mon: Monoid[U]) = new BijectedMonoid[U, T](mon, bij.inverse) } Friday, June 21, 13
  29. val englishMonoid = intMonoid.as[Monoid[EnglishInt]] scala> englishMonoid.plus(1.as[EnglishInt], 2.as[EnglishInt]) res21: EnglishInt =

    EnglishInt("three") scala> englishMonoid.plus(4321.as[EnglishInt], 1234.as[EnglishInt]) res22: EnglishInt = EnglishInt("five thousand five hundred fifty five") Friday, June 21, 13
  30. Client Producer Store flatMap flatMap flatMap flatMap Producer Producer Reduce

    Reduce Producer Store flatMap flatMap flatMap flatMap Producer Producer Reduce Reduce Storm Hadoop Friday, June 21, 13
  31. Client Producer Store flatMap flatMap flatMap flatMap Producer Producer Reduce

    Reduce Producer Store flatMap flatMap flatMap flatMap Producer Producer Reduce Reduce Storm Hadoop Store Friday, June 21, 13
  32. trait ReadableStore[-K, +V] extends Closeable { def get(k: K): Future[Option[V]]

    = multiGet(Set(k)).apply(k) def multiGet[K1 <: K](ks: Set[K1]): Map[K1, Future[Option[V]]] = ks.map { k => (k, self.get(k)) }.toMap override def close { } } Friday, June 21, 13
  33. class ReadableStoreMonoid[K, V](monoid: Monoid[V]) extends Monoid[ReadableStore[K, V]] { def zero

    = ReadableStore.const(Monoid.zero[V]) def plus(l: ReadableStore[K, V], r: ReadableStore[K,V]) = new ReadableStore[K, V] { override def get(k: K) = Monoid.plus(l.get(k), r.get(k)) } } Friday, June 21, 13
  34. // Store giving us access to hadoop data val offlineStore:

    ReadableStore[(Int, String), Long] // Store containing storm data val onlineStore: ReadableStore[(Int, String), Long] // Boom! Both! val combinedStore: ReadableStore[(Int, String), Long] = Monoid.plus(offlineStore, onlineStore) // Wuh? val cachedStore: ReadableStore[(Int, String), Long] = combinedStore.withCache(Cache.empty) Friday, June 21, 13
  35. Client Producer Store flatMap flatMap flatMap flatMap Producer Producer Reduce

    Reduce Producer Store flatMap flatMap flatMap flatMap Producer Producer Reduce Reduce Storm Hadoop Store Friday, June 21, 13