SlideShare ist ein Scribd-Unternehmen logo
1 von 51
Downloaden Sie, um offline zu lesen
Scala Data
Pipelines @
Spotify
Neville Li
@sinisa_lyh
Who am I?
‣ SpotifyNYCsince2011
‣ FormerlyYahoo!Search
‣ Musicrecommendations
‣ Datainfrastructure
‣ Scalasince2013
Spotify in numbers
• Started in 2006, 58 markets
• 75M+ active users, 20M+ paying
• 30M+ songs, 20K new per day
• 1.5 billion playlists
• 1 TB logs per day
• 1200+ node Hadoop cluster
• 10K+ Hadoop jobs per day
Music recommendation @ Spotify
• Discover Weekly
• Radio
• RelatedArtists
• Discover Page
Recommendation systems
A little teaser
PGroupedTable<K,V>::combineValues(CombineFn<K,V> combineFn,
CombineFn<K,V> reduceFn)
Crunch: CombineFns are used to represent the associative operations…
Grouped[K, +V]::reduce[U >: V](fn: (U, U) U)
Scalding: reduce with fn which must be associative and commutative…
PairRDDFunctions[K, V]::reduceByKey(fn: (V, V) => V)
Spark: Merge the values for each key using an associative reduce function…
Monoid!
enables map side reduce
Actually it’s a semigroup
One more teaser
Linear equation inAlternate Least Square (ALS) Matrix factorization
xu = (YTY + YT(Cu − I)Y)−1YTCup(u)
vectors.map { case (id, v) => (id, v * v) }.map(_._2).reduce(_ + _) // YtY
ratings.keyBy(fixedKey).join(outerProducts) // YtCuIY
.map { case (_, (r, op)) =>
(solveKey(r), op * (r.rating * alpha))
}.reduceByKey(_ + _)
ratings.keyBy(fixedKey).join(vectors) // YtCupu
.map { case (_, (r, v)) =>
val (Cui, pui) = (r.rating * alpha + 1, if (Cui > 0.0) 1.0 else 0.0)
(solveKey(r), v * (Cui * pui))
}.reduceByKey(_ + _)
http://www.slideshare.net/MrChrisJohnson/scala-data-pipelines-for-music-recommendations
Success story
• Mid 2013: 100+ Python Luigi M/R jobs, few tests
• 10+ new hires since, most fresh grads
• Few with Java experience, none with Scala
• Now: 300+ Scalding jobs, 400+ tests
• More ad-hoc jobs untracked
• Spark also taking off
First 10 months
……
Activity over time
Guess how many jobs
written by yours truly?
Performance vs. Agility
https://nicholassterling.wordpress.com/2012/11/16/scala-performance/
Let’sdiveinto
something
technical
To join or not to join?
val streams: TypedPipe[(String, String)] = _ // (track, user)
val tgp: TypedPipe[(String, String)] = _ // (track, genre)
streams
.join(tgp)
.values // (user, genre)
.group
.mapValueStream(vs => Iterator(vs.toSet)) // reducer-only
Hash join
val streams: TypedPipe[(String, String)] = _ // (track, user)
val tgp: TypedPipe[(String, String)] = _ // (track, genre)
streams
.hashJoin(tgp.forceToDisk) // tgp replicated to all mappers
.values // (user, genre)
.group
.mapValueStream(vs => Iterator(vs.toSet)) // reducer-only
CoGroup
val streams: TypedPipe[(String, String)] = _ // (track, user)
val tgp: TypedPipe[(String, String)] = _ // (track, genre)
streams
.cogroup(tgp) { case (_, users, genres) =>
users.map((_, genres.toSet))
} // (track, (user, genres))
.values // (user, genres)

.group
.reduce(_ ++ _) // map-side reduce!
CoGroup
val streams: TypedPipe[(String, String)] = _ // (track, user)
val tgp: TypedPipe[(String, String)] = _ // (track, genre)
streams
.cogroup(tgp) { case (_, users, genres) =>
users.map((_, genres.toSet))
} // (track, (user, genres))
.values // (user, genres)

.group
.sum // SetMonoid[Set[T]] from Algebird
* sum[U >:V](implicit sg: Semigroup[U])
Key-value file as distributed cache
val streams: TypedPipe[(String, String)] = _ // (gid, user)
val tgp: SparkeyManager = _ // tgp replicated to all mappers
streams
.map { case (track, user) =>
(user, tgp.get(track).split(",").toSet)
}
.group
.sum
https://github.com/spotify/sparkey
SparkeyManagerwraps DistributedCacheFile
Joins and CoGroups
• Require shuffle and reduce step
• Some ops force everything to reducers

e.g. mapGroup, mapValueStream
• CoGroup more flexible for complex logic
• Scalding flattens a.join(b).join(c)…

into MultiJoin(a, b, c, …)
Distributed cache
• Fasterwith off-heap binary files
• Building cache = more wiring
• Memory mapping may interfere withYARN
• E.g. 64GB nodes with 48GB for containers (no cgroup)
• 12 × 2GB containers each with 2GB JVM heap + mmap cache
• OOM and swap!
• Keep files small (< 1GB) or fallback to joins…
Analyze your jobs
• Concurrent Driven
• Visualize job execution
• Workflow optimization
• Bottlenecks
• Data skew
Notenough
math?
Recommending tracks
• User listened to Rammstein - Du Hast
• Recommend 10 similartracks
• 40 dimension feature vectors fortracks
• Compute cosine similarity between all pairs
• O(n) lookup per userwhere n ≈ 30m
• Trythat with 50m users * 10 seed tracks each
ANNOY - cheat by approximation
• Approximate Nearest Neighbor OhYeah
• Random projections and binarytree search
• Build index on single machine
• Load in mappers via distribute cache
• O(log n) lookup
https://github.com/spotify/annoy
https://github.com/spotify/annoy-java
ANN Benchmark
https://github.com/erikbern/ann-benchmarks
Filtering candidates
• Users don’t like seeing artist/album/tracks they already know
• But may forget what they listened long ago
• 50m * thousands of items each
• Over 5 years of streaming logs
• Need to update daily
• Need to purge old items per user
Options
• Aggregate all logs daily
• Aggregate last x days daily
• CSVof artist/album/track ids
• Bloom filters
Decayed value with cutoff
• Compute new user-item score daily
• Weighted on context, e.g. radio, search, playlist
• score’ = score + previous * 0.99
• half life = log0.99
0.5 = 69 days
• Cut off at top 2000
• Items that users might remember seeing recently
Bloom filters
• Probabilistic data structure
• Encoding set of items with m bits and k hash functions
• No false negative
• Tunable false positive probability
• Size proportional to capacity & FP probability
• Let’s build one per user-{artists,albums,tracks}
• Algebird BloomFilterMonoid: z = all zero bits, + = bitwise OR
Size versus max items & FP prob
• User-item distribution is uneven
• Assuming same setting for all users
• # items << capacity → wasting space
• # items > capacity → high FP rate
Scalable Bloom Filter
• Growing sequence of standard BFs
• Increasing capacity and tighter FP probability
• Most users have few BFs
• Power users have many
• Serialization and lookup overhead
Scalable Bloom Filter
• Growing sequence of standard BFs
• Increasing capacity and tighter FP probability
• Most users have few BFs
• Power users have many
• Serialization and lookup overhead
n=1k
item
Scalable Bloom Filter
• Growing sequence of standard BFs
• Increasing capacity and tighter FP probability
• Most users have few BFs
• Power users have many
• Serialization and lookup overhead
n=1k n=10k
item
full
Scalable Bloom Filter
• Growing sequence of standard BFs
• Increasing capacity and tighter FP probability
• Most users have few BFs
• Power users have many
• Serialization and lookup overhead
item
n=1k n=10k n=100k
fullfull
Scalable Bloom Filter
• Growing sequence of standard BFs
• Increasing capacity and tighter FP probability
• Most users have few BFs
• Power users have many
• Serialization and lookup overhead
n=1k n=10k n=100k n=1m
item
fullfullfull
Opportunistic Bloom Filter
• Building n BFs of increasing capacity in parallel
• Up to << N max possible items
• Keep smallest one with capacity > items inserted
• Expensive to build
• Cheap to store and lookup
Opportunistic Bloom Filter
• Building n BFs of increasing capacity in parallel
• Up to << N max possible items
• Keep smallest one with capacity > items inserted
• Expensive to build
• Cheap to store and lookup
n=1k
 
80%
n=10k
 
8%
n=100k
 
0.8%
n=1m
 
0.08%
item
Opportunistic Bloom Filter
• Building n BFs of increasing capacity in parallel
• Up to  N max possible items
• Keep smallest one with capacity  items inserted
• Expensive to build
• Cheap to store and lookup
n=1k
 
100%
n=10k
 
70%
n=100k
 
7%
n=1m
 
0.7%
item
full
Opportunistic Bloom Filter
• Building n BFs of increasing capacity in parallel
• Up to  N max possible items
• Keep smallest one with capacity  items inserted
• Expensive to build
• Cheap to store and lookup
n=1k
 
100%
n=10k
 
100%
n=100k
 
60%
n=1m

Weitere ähnliche Inhalte

Was ist angesagt?

Approximate nearest neighbor methods and vector models – NYC ML meetup
Approximate nearest neighbor methods and vector models – NYC ML meetupApproximate nearest neighbor methods and vector models – NYC ML meetup
Approximate nearest neighbor methods and vector models – NYC ML meetupErik Bernhardsson
 
CF Models for Music Recommendations At Spotify
CF Models for Music Recommendations At SpotifyCF Models for Music Recommendations At Spotify
CF Models for Music Recommendations At SpotifyVidhya Murali
 
(BDT318) How Netflix Handles Up To 8 Million Events Per Second
(BDT318) How Netflix Handles Up To 8 Million Events Per Second(BDT318) How Netflix Handles Up To 8 Million Events Per Second
(BDT318) How Netflix Handles Up To 8 Million Events Per SecondAmazon Web Services
 
Near Real-Time Netflix Recommendations Using Apache Spark Streaming with Nit...
 Near Real-Time Netflix Recommendations Using Apache Spark Streaming with Nit... Near Real-Time Netflix Recommendations Using Apache Spark Streaming with Nit...
Near Real-Time Netflix Recommendations Using Apache Spark Streaming with Nit...Databricks
 
Music recommendations @ MLConf 2014
Music recommendations @ MLConf 2014Music recommendations @ MLConf 2014
Music recommendations @ MLConf 2014Erik Bernhardsson
 
Recommending and Searching (Research @ Spotify)
Recommending and Searching (Research @ Spotify)Recommending and Searching (Research @ Spotify)
Recommending and Searching (Research @ Spotify)Mounia Lalmas-Roelleke
 
Recsys 2014 Tutorial - The Recommender Problem Revisited
Recsys 2014 Tutorial - The Recommender Problem RevisitedRecsys 2014 Tutorial - The Recommender Problem Revisited
Recsys 2014 Tutorial - The Recommender Problem RevisitedXavier Amatriain
 
Recommender Systems (Machine Learning Summer School 2014 @ CMU)
Recommender Systems (Machine Learning Summer School 2014 @ CMU)Recommender Systems (Machine Learning Summer School 2014 @ CMU)
Recommender Systems (Machine Learning Summer School 2014 @ CMU)Xavier Amatriain
 
Netflix viewing data architecture evolution - QCon 2014
Netflix viewing data architecture evolution - QCon 2014Netflix viewing data architecture evolution - QCon 2014
Netflix viewing data architecture evolution - QCon 2014Philip Fisher-Ogden
 
RocksDB compaction
RocksDB compactionRocksDB compaction
RocksDB compactionMIJIN AN
 
Apache Iceberg - A Table Format for Hige Analytic Datasets
Apache Iceberg - A Table Format for Hige Analytic DatasetsApache Iceberg - A Table Format for Hige Analytic Datasets
Apache Iceberg - A Table Format for Hige Analytic DatasetsAlluxio, Inc.
 
Tech Talk: RocksDB Slides by Dhruba Borthakur & Haobo Xu of Facebook
Tech Talk: RocksDB Slides by Dhruba Borthakur & Haobo Xu of FacebookTech Talk: RocksDB Slides by Dhruba Borthakur & Haobo Xu of Facebook
Tech Talk: RocksDB Slides by Dhruba Borthakur & Haobo Xu of FacebookThe Hive
 
ML+Hadoop at NYC Predictive Analytics
ML+Hadoop at NYC Predictive AnalyticsML+Hadoop at NYC Predictive Analytics
ML+Hadoop at NYC Predictive AnalyticsErik Bernhardsson
 
Interactive Recommender Systems with Netflix and Spotify
Interactive Recommender Systems with Netflix and SpotifyInteractive Recommender Systems with Netflix and Spotify
Interactive Recommender Systems with Netflix and SpotifyChris Johnson
 
Zipline—Airbnb’s Declarative Feature Engineering Framework
Zipline—Airbnb’s Declarative Feature Engineering FrameworkZipline—Airbnb’s Declarative Feature Engineering Framework
Zipline—Airbnb’s Declarative Feature Engineering FrameworkDatabricks
 
The Evolution of Hadoop at Spotify - Through Failures and Pain
The Evolution of Hadoop at Spotify - Through Failures and PainThe Evolution of Hadoop at Spotify - Through Failures and Pain
The Evolution of Hadoop at Spotify - Through Failures and PainRafał Wojdyła
 
Storm at Spotify
Storm at SpotifyStorm at Spotify
Storm at SpotifyNeville Li
 
Deep Learning for Recommender Systems
Deep Learning for Recommender SystemsDeep Learning for Recommender Systems
Deep Learning for Recommender SystemsJustin Basilico
 
Batch Processing at Scale with Flink & Iceberg
Batch Processing at Scale with Flink & IcebergBatch Processing at Scale with Flink & Iceberg
Batch Processing at Scale with Flink & IcebergFlink Forward
 

Was ist angesagt? (20)

Approximate nearest neighbor methods and vector models – NYC ML meetup
Approximate nearest neighbor methods and vector models – NYC ML meetupApproximate nearest neighbor methods and vector models – NYC ML meetup
Approximate nearest neighbor methods and vector models – NYC ML meetup
 
CF Models for Music Recommendations At Spotify
CF Models for Music Recommendations At SpotifyCF Models for Music Recommendations At Spotify
CF Models for Music Recommendations At Spotify
 
(BDT318) How Netflix Handles Up To 8 Million Events Per Second
(BDT318) How Netflix Handles Up To 8 Million Events Per Second(BDT318) How Netflix Handles Up To 8 Million Events Per Second
(BDT318) How Netflix Handles Up To 8 Million Events Per Second
 
Near Real-Time Netflix Recommendations Using Apache Spark Streaming with Nit...
 Near Real-Time Netflix Recommendations Using Apache Spark Streaming with Nit... Near Real-Time Netflix Recommendations Using Apache Spark Streaming with Nit...
Near Real-Time Netflix Recommendations Using Apache Spark Streaming with Nit...
 
Music recommendations @ MLConf 2014
Music recommendations @ MLConf 2014Music recommendations @ MLConf 2014
Music recommendations @ MLConf 2014
 
Recommending and searching @ Spotify
Recommending and searching @ SpotifyRecommending and searching @ Spotify
Recommending and searching @ Spotify
 
Recommending and Searching (Research @ Spotify)
Recommending and Searching (Research @ Spotify)Recommending and Searching (Research @ Spotify)
Recommending and Searching (Research @ Spotify)
 
Recsys 2014 Tutorial - The Recommender Problem Revisited
Recsys 2014 Tutorial - The Recommender Problem RevisitedRecsys 2014 Tutorial - The Recommender Problem Revisited
Recsys 2014 Tutorial - The Recommender Problem Revisited
 
Recommender Systems (Machine Learning Summer School 2014 @ CMU)
Recommender Systems (Machine Learning Summer School 2014 @ CMU)Recommender Systems (Machine Learning Summer School 2014 @ CMU)
Recommender Systems (Machine Learning Summer School 2014 @ CMU)
 
Netflix viewing data architecture evolution - QCon 2014
Netflix viewing data architecture evolution - QCon 2014Netflix viewing data architecture evolution - QCon 2014
Netflix viewing data architecture evolution - QCon 2014
 
RocksDB compaction
RocksDB compactionRocksDB compaction
RocksDB compaction
 
Apache Iceberg - A Table Format for Hige Analytic Datasets
Apache Iceberg - A Table Format for Hige Analytic DatasetsApache Iceberg - A Table Format for Hige Analytic Datasets
Apache Iceberg - A Table Format for Hige Analytic Datasets
 
Tech Talk: RocksDB Slides by Dhruba Borthakur & Haobo Xu of Facebook
Tech Talk: RocksDB Slides by Dhruba Borthakur & Haobo Xu of FacebookTech Talk: RocksDB Slides by Dhruba Borthakur & Haobo Xu of Facebook
Tech Talk: RocksDB Slides by Dhruba Borthakur & Haobo Xu of Facebook
 
ML+Hadoop at NYC Predictive Analytics
ML+Hadoop at NYC Predictive AnalyticsML+Hadoop at NYC Predictive Analytics
ML+Hadoop at NYC Predictive Analytics
 
Interactive Recommender Systems with Netflix and Spotify
Interactive Recommender Systems with Netflix and SpotifyInteractive Recommender Systems with Netflix and Spotify
Interactive Recommender Systems with Netflix and Spotify
 
Zipline—Airbnb’s Declarative Feature Engineering Framework
Zipline—Airbnb’s Declarative Feature Engineering FrameworkZipline—Airbnb’s Declarative Feature Engineering Framework
Zipline—Airbnb’s Declarative Feature Engineering Framework
 
The Evolution of Hadoop at Spotify - Through Failures and Pain
The Evolution of Hadoop at Spotify - Through Failures and PainThe Evolution of Hadoop at Spotify - Through Failures and Pain
The Evolution of Hadoop at Spotify - Through Failures and Pain
 
Storm at Spotify
Storm at SpotifyStorm at Spotify
Storm at Spotify
 
Deep Learning for Recommender Systems
Deep Learning for Recommender SystemsDeep Learning for Recommender Systems
Deep Learning for Recommender Systems
 
Batch Processing at Scale with Flink & Iceberg
Batch Processing at Scale with Flink & IcebergBatch Processing at Scale with Flink & Iceberg
Batch Processing at Scale with Flink & Iceberg
 

Andere mochten auch

Playlist Recommendations @ Spotify
Playlist Recommendations @ SpotifyPlaylist Recommendations @ Spotify
Playlist Recommendations @ SpotifyNikhil Tibrewal
 
Mugo one pager
Mugo one pagerMugo one pager
Mugo one pagerori segal
 
Jackdaw research music survey report
Jackdaw research music survey reportJackdaw research music survey report
Jackdaw research music survey reportJan Dawson
 
How We Listen to Music - SXSW 2015
How We Listen to Music - SXSW 2015How We Listen to Music - SXSW 2015
How We Listen to Music - SXSW 2015Paul Lamere
 
Music Personalization At Spotify
Music Personalization At SpotifyMusic Personalization At Spotify
Music Personalization At SpotifyVidhya Murali
 

Andere mochten auch (7)

Playlist Recommendations @ Spotify
Playlist Recommendations @ SpotifyPlaylist Recommendations @ Spotify
Playlist Recommendations @ Spotify
 
Music survey results (2)
Music survey results (2)Music survey results (2)
Music survey results (2)
 
Music & interaction
Music & interactionMusic & interaction
Music & interaction
 
Mugo one pager
Mugo one pagerMugo one pager
Mugo one pager
 
Jackdaw research music survey report
Jackdaw research music survey reportJackdaw research music survey report
Jackdaw research music survey report
 
How We Listen to Music - SXSW 2015
How We Listen to Music - SXSW 2015How We Listen to Music - SXSW 2015
How We Listen to Music - SXSW 2015
 
Music Personalization At Spotify
Music Personalization At SpotifyMusic Personalization At Spotify
Music Personalization At Spotify
 

Ähnlich wie Scala Data Pipelines @ Spotify

London devops logging
London devops loggingLondon devops logging
London devops loggingTomas Doran
 
CPANTS: Kwalitative website and its tools
CPANTS: Kwalitative website and its toolsCPANTS: Kwalitative website and its tools
CPANTS: Kwalitative website and its toolscharsbar
 
Introduction to Apache Beam & No Shard Left Behind: APIs for Massive Parallel...
Introduction to Apache Beam & No Shard Left Behind: APIs for Massive Parallel...Introduction to Apache Beam & No Shard Left Behind: APIs for Massive Parallel...
Introduction to Apache Beam & No Shard Left Behind: APIs for Massive Parallel...Dan Halperin
 
Intelligent Search
Intelligent SearchIntelligent Search
Intelligent SearchTed Dunning
 
Vertica architecture
Vertica architectureVertica architecture
Vertica architectureZvika Gutkin
 
Introduction to Vertica (Architecture & More)
Introduction to Vertica (Architecture & More)Introduction to Vertica (Architecture & More)
Introduction to Vertica (Architecture & More)LivePerson
 
Stream processing from single node to a cluster
Stream processing from single node to a clusterStream processing from single node to a cluster
Stream processing from single node to a clusterGal Marder
 
Tuning Your Engine
Tuning Your EngineTuning Your Engine
Tuning Your Enginejoelbradbury
 
Mendeley’s Research Catalogue: building it, opening it up and making it even ...
Mendeley’s Research Catalogue: building it, opening it up and making it even ...Mendeley’s Research Catalogue: building it, opening it up and making it even ...
Mendeley’s Research Catalogue: building it, opening it up and making it even ...Kris Jack
 
Scala in practice - 3 years later
Scala in practice - 3 years laterScala in practice - 3 years later
Scala in practice - 3 years laterpatforna
 
Scala in-practice-3-years by Patric Fornasier, Springr, presented at Pune Sca...
Scala in-practice-3-years by Patric Fornasier, Springr, presented at Pune Sca...Scala in-practice-3-years by Patric Fornasier, Springr, presented at Pune Sca...
Scala in-practice-3-years by Patric Fornasier, Springr, presented at Pune Sca...Thoughtworks
 
Scaling ingest pipelines with high performance computing principles - Rajiv K...
Scaling ingest pipelines with high performance computing principles - Rajiv K...Scaling ingest pipelines with high performance computing principles - Rajiv K...
Scaling ingest pipelines with high performance computing principles - Rajiv K...SignalFx
 
Analyze one year of radio station songs aired with Spark SQL, Spotify, and Da...
Analyze one year of radio station songs aired with Spark SQL, Spotify, and Da...Analyze one year of radio station songs aired with Spark SQL, Spotify, and Da...
Analyze one year of radio station songs aired with Spark SQL, Spotify, and Da...Paul Leclercq
 
Message:Passing - lpw 2012
Message:Passing - lpw 2012Message:Passing - lpw 2012
Message:Passing - lpw 2012Tomas Doran
 

Ähnlich wie Scala Data Pipelines @ Spotify (20)

London devops logging
London devops loggingLondon devops logging
London devops logging
 
Intelligent Search
Intelligent SearchIntelligent Search
Intelligent Search
 
CPANTS: Kwalitative website and its tools
CPANTS: Kwalitative website and its toolsCPANTS: Kwalitative website and its tools
CPANTS: Kwalitative website and its tools
 
Introduction to Apache Beam & No Shard Left Behind: APIs for Massive Parallel...
Introduction to Apache Beam & No Shard Left Behind: APIs for Massive Parallel...Introduction to Apache Beam & No Shard Left Behind: APIs for Massive Parallel...
Introduction to Apache Beam & No Shard Left Behind: APIs for Massive Parallel...
 
Let's Get to the Rapids
Let's Get to the RapidsLet's Get to the Rapids
Let's Get to the Rapids
 
Intelligent Search
Intelligent SearchIntelligent Search
Intelligent Search
 
Vertica architecture
Vertica architectureVertica architecture
Vertica architecture
 
Introduction to Vertica (Architecture & More)
Introduction to Vertica (Architecture & More)Introduction to Vertica (Architecture & More)
Introduction to Vertica (Architecture & More)
 
Stream processing from single node to a cluster
Stream processing from single node to a clusterStream processing from single node to a cluster
Stream processing from single node to a cluster
 
Akka streams
Akka streamsAkka streams
Akka streams
 
Tuning Your Engine
Tuning Your EngineTuning Your Engine
Tuning Your Engine
 
Mendeley’s Research Catalogue: building it, opening it up and making it even ...
Mendeley’s Research Catalogue: building it, opening it up and making it even ...Mendeley’s Research Catalogue: building it, opening it up and making it even ...
Mendeley’s Research Catalogue: building it, opening it up and making it even ...
 
Scala in practice - 3 years later
Scala in practice - 3 years laterScala in practice - 3 years later
Scala in practice - 3 years later
 
Scala in-practice-3-years by Patric Fornasier, Springr, presented at Pune Sca...
Scala in-practice-3-years by Patric Fornasier, Springr, presented at Pune Sca...Scala in-practice-3-years by Patric Fornasier, Springr, presented at Pune Sca...
Scala in-practice-3-years by Patric Fornasier, Springr, presented at Pune Sca...
 
Hive at Last.fm
Hive at Last.fmHive at Last.fm
Hive at Last.fm
 
Graphite
GraphiteGraphite
Graphite
 
Scaling ingest pipelines with high performance computing principles - Rajiv K...
Scaling ingest pipelines with high performance computing principles - Rajiv K...Scaling ingest pipelines with high performance computing principles - Rajiv K...
Scaling ingest pipelines with high performance computing principles - Rajiv K...
 
Apache HAWQ Architecture
Apache HAWQ ArchitectureApache HAWQ Architecture
Apache HAWQ Architecture
 
Analyze one year of radio station songs aired with Spark SQL, Spotify, and Da...
Analyze one year of radio station songs aired with Spark SQL, Spotify, and Da...Analyze one year of radio station songs aired with Spark SQL, Spotify, and Da...
Analyze one year of radio station songs aired with Spark SQL, Spotify, and Da...
 
Message:Passing - lpw 2012
Message:Passing - lpw 2012Message:Passing - lpw 2012
Message:Passing - lpw 2012
 

Mehr von Neville Li

Sorry - How Bieber broke Google Cloud at Spotify
Sorry - How Bieber broke Google Cloud at SpotifySorry - How Bieber broke Google Cloud at Spotify
Sorry - How Bieber broke Google Cloud at SpotifyNeville Li
 
Scio - Moving to Google Cloud, A Spotify Story
 Scio - Moving to Google Cloud, A Spotify Story Scio - Moving to Google Cloud, A Spotify Story
Scio - Moving to Google Cloud, A Spotify StoryNeville Li
 
Scio - A Scala API for Google Cloud Dataflow & Apache Beam
Scio - A Scala API for Google Cloud Dataflow & Apache BeamScio - A Scala API for Google Cloud Dataflow & Apache Beam
Scio - A Scala API for Google Cloud Dataflow & Apache BeamNeville Li
 
From stream to recommendation using apache beam with cloud pubsub and cloud d...
From stream to recommendation using apache beam with cloud pubsub and cloud d...From stream to recommendation using apache beam with cloud pubsub and cloud d...
From stream to recommendation using apache beam with cloud pubsub and cloud d...Neville Li
 
Why functional why scala
Why functional  why scala Why functional  why scala
Why functional why scala Neville Li
 

Mehr von Neville Li (6)

Sorry - How Bieber broke Google Cloud at Spotify
Sorry - How Bieber broke Google Cloud at SpotifySorry - How Bieber broke Google Cloud at Spotify
Sorry - How Bieber broke Google Cloud at Spotify
 
Scio - Moving to Google Cloud, A Spotify Story
 Scio - Moving to Google Cloud, A Spotify Story Scio - Moving to Google Cloud, A Spotify Story
Scio - Moving to Google Cloud, A Spotify Story
 
Scio - A Scala API for Google Cloud Dataflow & Apache Beam
Scio - A Scala API for Google Cloud Dataflow & Apache BeamScio - A Scala API for Google Cloud Dataflow & Apache Beam
Scio - A Scala API for Google Cloud Dataflow & Apache Beam
 
Scio
ScioScio
Scio
 
From stream to recommendation using apache beam with cloud pubsub and cloud d...
From stream to recommendation using apache beam with cloud pubsub and cloud d...From stream to recommendation using apache beam with cloud pubsub and cloud d...
From stream to recommendation using apache beam with cloud pubsub and cloud d...
 
Why functional why scala
Why functional  why scala Why functional  why scala
Why functional why scala
 

Kürzlich hochgeladen

Alluxio Monthly Webinar | Cloud-Native Model Training on Distributed Data
Alluxio Monthly Webinar | Cloud-Native Model Training on Distributed DataAlluxio Monthly Webinar | Cloud-Native Model Training on Distributed Data
Alluxio Monthly Webinar | Cloud-Native Model Training on Distributed DataAlluxio, Inc.
 
Folding Cheat Sheet #4 - fourth in a series
Folding Cheat Sheet #4 - fourth in a seriesFolding Cheat Sheet #4 - fourth in a series
Folding Cheat Sheet #4 - fourth in a seriesPhilip Schwarz
 
Balasore Best It Company|| Top 10 IT Company || Balasore Software company Odisha
Balasore Best It Company|| Top 10 IT Company || Balasore Software company OdishaBalasore Best It Company|| Top 10 IT Company || Balasore Software company Odisha
Balasore Best It Company|| Top 10 IT Company || Balasore Software company Odishasmiwainfosol
 
A healthy diet for your Java application Devoxx France.pdf
A healthy diet for your Java application Devoxx France.pdfA healthy diet for your Java application Devoxx France.pdf
A healthy diet for your Java application Devoxx France.pdfMarharyta Nedzelska
 
Global Identity Enrolment and Verification Pro Solution - Cizo Technology Ser...
Global Identity Enrolment and Verification Pro Solution - Cizo Technology Ser...Global Identity Enrolment and Verification Pro Solution - Cizo Technology Ser...
Global Identity Enrolment and Verification Pro Solution - Cizo Technology Ser...Cizo Technology Services
 
Intelligent Home Wi-Fi Solutions | ThinkPalm
Intelligent Home Wi-Fi Solutions | ThinkPalmIntelligent Home Wi-Fi Solutions | ThinkPalm
Intelligent Home Wi-Fi Solutions | ThinkPalmSujith Sukumaran
 
Maximizing Efficiency and Profitability with OnePlan’s Professional Service A...
Maximizing Efficiency and Profitability with OnePlan’s Professional Service A...Maximizing Efficiency and Profitability with OnePlan’s Professional Service A...
Maximizing Efficiency and Profitability with OnePlan’s Professional Service A...OnePlan Solutions
 
Open Source Summit NA 2024: Open Source Cloud Costs - OpenCost's Impact on En...
Open Source Summit NA 2024: Open Source Cloud Costs - OpenCost's Impact on En...Open Source Summit NA 2024: Open Source Cloud Costs - OpenCost's Impact on En...
Open Source Summit NA 2024: Open Source Cloud Costs - OpenCost's Impact on En...Matt Ray
 
What is Fashion PLM and Why Do You Need It
What is Fashion PLM and Why Do You Need ItWhat is Fashion PLM and Why Do You Need It
What is Fashion PLM and Why Do You Need ItWave PLM
 
React Server Component in Next.js by Hanief Utama
React Server Component in Next.js by Hanief UtamaReact Server Component in Next.js by Hanief Utama
React Server Component in Next.js by Hanief UtamaHanief Utama
 
MYjobs Presentation Django-based project
MYjobs Presentation Django-based projectMYjobs Presentation Django-based project
MYjobs Presentation Django-based projectAnoyGreter
 
Cloud Data Center Network Construction - IEEE
Cloud Data Center Network Construction - IEEECloud Data Center Network Construction - IEEE
Cloud Data Center Network Construction - IEEEVICTOR MAESTRE RAMIREZ
 
英国UN学位证,北安普顿大学毕业证书1:1制作
英国UN学位证,北安普顿大学毕业证书1:1制作英国UN学位证,北安普顿大学毕业证书1:1制作
英国UN学位证,北安普顿大学毕业证书1:1制作qr0udbr0
 
Introduction Computer Science - Software Design.pdf
Introduction Computer Science - Software Design.pdfIntroduction Computer Science - Software Design.pdf
Introduction Computer Science - Software Design.pdfFerryKemperman
 
How to Track Employee Performance A Comprehensive Guide.pdf
How to Track Employee Performance A Comprehensive Guide.pdfHow to Track Employee Performance A Comprehensive Guide.pdf
How to Track Employee Performance A Comprehensive Guide.pdfLivetecs LLC
 
Xen Safety Embedded OSS Summit April 2024 v4.pdf
Xen Safety Embedded OSS Summit April 2024 v4.pdfXen Safety Embedded OSS Summit April 2024 v4.pdf
Xen Safety Embedded OSS Summit April 2024 v4.pdfStefano Stabellini
 
Building a General PDE Solving Framework with Symbolic-Numeric Scientific Mac...
Building a General PDE Solving Framework with Symbolic-Numeric Scientific Mac...Building a General PDE Solving Framework with Symbolic-Numeric Scientific Mac...
Building a General PDE Solving Framework with Symbolic-Numeric Scientific Mac...stazi3110
 
Buds n Tech IT Solutions: Top-Notch Web Services in Noida
Buds n Tech IT Solutions: Top-Notch Web Services in NoidaBuds n Tech IT Solutions: Top-Notch Web Services in Noida
Buds n Tech IT Solutions: Top-Notch Web Services in Noidabntitsolutionsrishis
 
BATTLEFIELD ORM: TIPS, TACTICS AND STRATEGIES FOR CONQUERING YOUR DATABASE
BATTLEFIELD ORM: TIPS, TACTICS AND STRATEGIES FOR CONQUERING YOUR DATABASEBATTLEFIELD ORM: TIPS, TACTICS AND STRATEGIES FOR CONQUERING YOUR DATABASE
BATTLEFIELD ORM: TIPS, TACTICS AND STRATEGIES FOR CONQUERING YOUR DATABASEOrtus Solutions, Corp
 

Kürzlich hochgeladen (20)

Alluxio Monthly Webinar | Cloud-Native Model Training on Distributed Data
Alluxio Monthly Webinar | Cloud-Native Model Training on Distributed DataAlluxio Monthly Webinar | Cloud-Native Model Training on Distributed Data
Alluxio Monthly Webinar | Cloud-Native Model Training on Distributed Data
 
Folding Cheat Sheet #4 - fourth in a series
Folding Cheat Sheet #4 - fourth in a seriesFolding Cheat Sheet #4 - fourth in a series
Folding Cheat Sheet #4 - fourth in a series
 
2.pdf Ejercicios de programación competitiva
2.pdf Ejercicios de programación competitiva2.pdf Ejercicios de programación competitiva
2.pdf Ejercicios de programación competitiva
 
Balasore Best It Company|| Top 10 IT Company || Balasore Software company Odisha
Balasore Best It Company|| Top 10 IT Company || Balasore Software company OdishaBalasore Best It Company|| Top 10 IT Company || Balasore Software company Odisha
Balasore Best It Company|| Top 10 IT Company || Balasore Software company Odisha
 
A healthy diet for your Java application Devoxx France.pdf
A healthy diet for your Java application Devoxx France.pdfA healthy diet for your Java application Devoxx France.pdf
A healthy diet for your Java application Devoxx France.pdf
 
Global Identity Enrolment and Verification Pro Solution - Cizo Technology Ser...
Global Identity Enrolment and Verification Pro Solution - Cizo Technology Ser...Global Identity Enrolment and Verification Pro Solution - Cizo Technology Ser...
Global Identity Enrolment and Verification Pro Solution - Cizo Technology Ser...
 
Intelligent Home Wi-Fi Solutions | ThinkPalm
Intelligent Home Wi-Fi Solutions | ThinkPalmIntelligent Home Wi-Fi Solutions | ThinkPalm
Intelligent Home Wi-Fi Solutions | ThinkPalm
 
Maximizing Efficiency and Profitability with OnePlan’s Professional Service A...
Maximizing Efficiency and Profitability with OnePlan’s Professional Service A...Maximizing Efficiency and Profitability with OnePlan’s Professional Service A...
Maximizing Efficiency and Profitability with OnePlan’s Professional Service A...
 
Open Source Summit NA 2024: Open Source Cloud Costs - OpenCost's Impact on En...
Open Source Summit NA 2024: Open Source Cloud Costs - OpenCost's Impact on En...Open Source Summit NA 2024: Open Source Cloud Costs - OpenCost's Impact on En...
Open Source Summit NA 2024: Open Source Cloud Costs - OpenCost's Impact on En...
 
What is Fashion PLM and Why Do You Need It
What is Fashion PLM and Why Do You Need ItWhat is Fashion PLM and Why Do You Need It
What is Fashion PLM and Why Do You Need It
 
React Server Component in Next.js by Hanief Utama
React Server Component in Next.js by Hanief UtamaReact Server Component in Next.js by Hanief Utama
React Server Component in Next.js by Hanief Utama
 
MYjobs Presentation Django-based project
MYjobs Presentation Django-based projectMYjobs Presentation Django-based project
MYjobs Presentation Django-based project
 
Cloud Data Center Network Construction - IEEE
Cloud Data Center Network Construction - IEEECloud Data Center Network Construction - IEEE
Cloud Data Center Network Construction - IEEE
 
英国UN学位证,北安普顿大学毕业证书1:1制作
英国UN学位证,北安普顿大学毕业证书1:1制作英国UN学位证,北安普顿大学毕业证书1:1制作
英国UN学位证,北安普顿大学毕业证书1:1制作
 
Introduction Computer Science - Software Design.pdf
Introduction Computer Science - Software Design.pdfIntroduction Computer Science - Software Design.pdf
Introduction Computer Science - Software Design.pdf
 
How to Track Employee Performance A Comprehensive Guide.pdf
How to Track Employee Performance A Comprehensive Guide.pdfHow to Track Employee Performance A Comprehensive Guide.pdf
How to Track Employee Performance A Comprehensive Guide.pdf
 
Xen Safety Embedded OSS Summit April 2024 v4.pdf
Xen Safety Embedded OSS Summit April 2024 v4.pdfXen Safety Embedded OSS Summit April 2024 v4.pdf
Xen Safety Embedded OSS Summit April 2024 v4.pdf
 
Building a General PDE Solving Framework with Symbolic-Numeric Scientific Mac...
Building a General PDE Solving Framework with Symbolic-Numeric Scientific Mac...Building a General PDE Solving Framework with Symbolic-Numeric Scientific Mac...
Building a General PDE Solving Framework with Symbolic-Numeric Scientific Mac...
 
Buds n Tech IT Solutions: Top-Notch Web Services in Noida
Buds n Tech IT Solutions: Top-Notch Web Services in NoidaBuds n Tech IT Solutions: Top-Notch Web Services in Noida
Buds n Tech IT Solutions: Top-Notch Web Services in Noida
 
BATTLEFIELD ORM: TIPS, TACTICS AND STRATEGIES FOR CONQUERING YOUR DATABASE
BATTLEFIELD ORM: TIPS, TACTICS AND STRATEGIES FOR CONQUERING YOUR DATABASEBATTLEFIELD ORM: TIPS, TACTICS AND STRATEGIES FOR CONQUERING YOUR DATABASE
BATTLEFIELD ORM: TIPS, TACTICS AND STRATEGIES FOR CONQUERING YOUR DATABASE
 

Scala Data Pipelines @ Spotify

  • 2. Who am I? ‣ SpotifyNYCsince2011 ‣ FormerlyYahoo!Search ‣ Musicrecommendations ‣ Datainfrastructure ‣ Scalasince2013
  • 3. Spotify in numbers • Started in 2006, 58 markets • 75M+ active users, 20M+ paying • 30M+ songs, 20K new per day • 1.5 billion playlists • 1 TB logs per day • 1200+ node Hadoop cluster • 10K+ Hadoop jobs per day
  • 4. Music recommendation @ Spotify • Discover Weekly • Radio • RelatedArtists • Discover Page
  • 6. A little teaser PGroupedTable<K,V>::combineValues(CombineFn<K,V> combineFn, CombineFn<K,V> reduceFn) Crunch: CombineFns are used to represent the associative operations… Grouped[K, +V]::reduce[U >: V](fn: (U, U) U) Scalding: reduce with fn which must be associative and commutative… PairRDDFunctions[K, V]::reduceByKey(fn: (V, V) => V) Spark: Merge the values for each key using an associative reduce function…
  • 7. Monoid! enables map side reduce Actually it’s a semigroup
  • 8. One more teaser Linear equation inAlternate Least Square (ALS) Matrix factorization xu = (YTY + YT(Cu − I)Y)−1YTCup(u) vectors.map { case (id, v) => (id, v * v) }.map(_._2).reduce(_ + _) // YtY ratings.keyBy(fixedKey).join(outerProducts) // YtCuIY .map { case (_, (r, op)) => (solveKey(r), op * (r.rating * alpha)) }.reduceByKey(_ + _) ratings.keyBy(fixedKey).join(vectors) // YtCupu .map { case (_, (r, v)) => val (Cui, pui) = (r.rating * alpha + 1, if (Cui > 0.0) 1.0 else 0.0) (solveKey(r), v * (Cui * pui)) }.reduceByKey(_ + _) http://www.slideshare.net/MrChrisJohnson/scala-data-pipelines-for-music-recommendations
  • 9. Success story • Mid 2013: 100+ Python Luigi M/R jobs, few tests • 10+ new hires since, most fresh grads • Few with Java experience, none with Scala • Now: 300+ Scalding jobs, 400+ tests • More ad-hoc jobs untracked • Spark also taking off
  • 12. Guess how many jobs written by yours truly?
  • 15. To join or not to join? val streams: TypedPipe[(String, String)] = _ // (track, user) val tgp: TypedPipe[(String, String)] = _ // (track, genre) streams .join(tgp) .values // (user, genre) .group .mapValueStream(vs => Iterator(vs.toSet)) // reducer-only
  • 16. Hash join val streams: TypedPipe[(String, String)] = _ // (track, user) val tgp: TypedPipe[(String, String)] = _ // (track, genre) streams .hashJoin(tgp.forceToDisk) // tgp replicated to all mappers .values // (user, genre) .group .mapValueStream(vs => Iterator(vs.toSet)) // reducer-only
  • 17. CoGroup val streams: TypedPipe[(String, String)] = _ // (track, user) val tgp: TypedPipe[(String, String)] = _ // (track, genre) streams .cogroup(tgp) { case (_, users, genres) => users.map((_, genres.toSet)) } // (track, (user, genres)) .values // (user, genres)
 .group .reduce(_ ++ _) // map-side reduce!
  • 18. CoGroup val streams: TypedPipe[(String, String)] = _ // (track, user) val tgp: TypedPipe[(String, String)] = _ // (track, genre) streams .cogroup(tgp) { case (_, users, genres) => users.map((_, genres.toSet)) } // (track, (user, genres)) .values // (user, genres)
 .group .sum // SetMonoid[Set[T]] from Algebird * sum[U >:V](implicit sg: Semigroup[U])
  • 19. Key-value file as distributed cache val streams: TypedPipe[(String, String)] = _ // (gid, user) val tgp: SparkeyManager = _ // tgp replicated to all mappers streams .map { case (track, user) => (user, tgp.get(track).split(",").toSet) } .group .sum https://github.com/spotify/sparkey SparkeyManagerwraps DistributedCacheFile
  • 20. Joins and CoGroups • Require shuffle and reduce step • Some ops force everything to reducers
 e.g. mapGroup, mapValueStream • CoGroup more flexible for complex logic • Scalding flattens a.join(b).join(c)…
 into MultiJoin(a, b, c, …)
  • 21. Distributed cache • Fasterwith off-heap binary files • Building cache = more wiring • Memory mapping may interfere withYARN • E.g. 64GB nodes with 48GB for containers (no cgroup) • 12 × 2GB containers each with 2GB JVM heap + mmap cache • OOM and swap! • Keep files small (< 1GB) or fallback to joins…
  • 22. Analyze your jobs • Concurrent Driven • Visualize job execution • Workflow optimization • Bottlenecks • Data skew
  • 24. Recommending tracks • User listened to Rammstein - Du Hast • Recommend 10 similartracks • 40 dimension feature vectors fortracks • Compute cosine similarity between all pairs • O(n) lookup per userwhere n ≈ 30m • Trythat with 50m users * 10 seed tracks each
  • 25. ANNOY - cheat by approximation • Approximate Nearest Neighbor OhYeah • Random projections and binarytree search • Build index on single machine • Load in mappers via distribute cache • O(log n) lookup https://github.com/spotify/annoy https://github.com/spotify/annoy-java
  • 27. Filtering candidates • Users don’t like seeing artist/album/tracks they already know • But may forget what they listened long ago • 50m * thousands of items each • Over 5 years of streaming logs • Need to update daily • Need to purge old items per user
  • 28. Options • Aggregate all logs daily • Aggregate last x days daily • CSVof artist/album/track ids • Bloom filters
  • 29. Decayed value with cutoff • Compute new user-item score daily • Weighted on context, e.g. radio, search, playlist • score’ = score + previous * 0.99 • half life = log0.99 0.5 = 69 days • Cut off at top 2000 • Items that users might remember seeing recently
  • 30. Bloom filters • Probabilistic data structure • Encoding set of items with m bits and k hash functions • No false negative • Tunable false positive probability • Size proportional to capacity & FP probability • Let’s build one per user-{artists,albums,tracks} • Algebird BloomFilterMonoid: z = all zero bits, + = bitwise OR
  • 31. Size versus max items & FP prob • User-item distribution is uneven • Assuming same setting for all users • # items << capacity → wasting space • # items > capacity → high FP rate
  • 32. Scalable Bloom Filter • Growing sequence of standard BFs • Increasing capacity and tighter FP probability • Most users have few BFs • Power users have many • Serialization and lookup overhead
  • 33. Scalable Bloom Filter • Growing sequence of standard BFs • Increasing capacity and tighter FP probability • Most users have few BFs • Power users have many • Serialization and lookup overhead n=1k item
  • 34. Scalable Bloom Filter • Growing sequence of standard BFs • Increasing capacity and tighter FP probability • Most users have few BFs • Power users have many • Serialization and lookup overhead n=1k n=10k item full
  • 35. Scalable Bloom Filter • Growing sequence of standard BFs • Increasing capacity and tighter FP probability • Most users have few BFs • Power users have many • Serialization and lookup overhead item n=1k n=10k n=100k fullfull
  • 36. Scalable Bloom Filter • Growing sequence of standard BFs • Increasing capacity and tighter FP probability • Most users have few BFs • Power users have many • Serialization and lookup overhead n=1k n=10k n=100k n=1m item fullfullfull
  • 37. Opportunistic Bloom Filter • Building n BFs of increasing capacity in parallel • Up to << N max possible items • Keep smallest one with capacity > items inserted • Expensive to build • Cheap to store and lookup
  • 38. Opportunistic Bloom Filter • Building n BFs of increasing capacity in parallel • Up to << N max possible items • Keep smallest one with capacity > items inserted • Expensive to build • Cheap to store and lookup n=1k
  • 43. Opportunistic Bloom Filter • Building n BFs of increasing capacity in parallel • Up to N max possible items • Keep smallest one with capacity items inserted • Expensive to build • Cheap to store and lookup n=1k
  • 48. Opportunistic Bloom Filter • Building n BFs of increasing capacity in parallel • Up to N max possible items • Keep smallest one with capacity items inserted • Expensive to build • Cheap to store and lookup n=1k
  • 53. Opportunistic Bloom Filter • Building n BFs of increasing capacity in parallel • Up to N max possible items • Keep smallest one with capacity items inserted • Expensive to build • Cheap to store and lookup n=1k
  • 60. Track metadata • Label dump → content ingestion • Third partytrack genres, e.g. GraceNote • Audio attributes, e.g. tempo, key, time signature • Cultural data, e.g. popularity, tags • Latent vectors from collaborative filtering • Many sources for album, artist, user metadata too
  • 61. Multiple data sources • Big joins • Complex dependencies • Wide rows with few columns accessed • Wasting I/O
  • 62. Apache Parquet • Pre-join sources into mega-datasets • Store as Parquet columnar storage • Column projection • Predicate pushdown • Avro within Scalding pipelines
  • 63. Projection pipe.map(a = (a.getName, a.getAmount)) versus Parquet.project[Account](name, amount) • Strings → unsafe and error prone • No IDE auto-completion → finger injury • my_fancy_field_name → .getMyFancyFieldName • Hard to migrate existing code
  • 64. Predicate pipe.filter(a = a.getName == Neville a.getAmount 100) versus FilterApi.and( FilterApi.eq(FilterApi.binaryColumn(name), Binary.fromString(Neville)), FilterApi.gt(FilterApi.floatColumn(amount), 100f.asInstnacesOf[java.lang.Float]))
  • 65. Macro to the rescue Code →AST→ (pattern matching) → (recursion) → (quasi-quotes) → Code Projection[Account](_.getName, _.getAmount) Predicate[Account](x = x.getName == “Neville x.getAmount 100) https://github.com/nevillelyh/parquet-avro-extra http://www.lyh.me/slides/macros.html
  • 66. What else? ‣ Analytics ‣ Adstargeting,prediction ‣ Metadataquality ‣ Zeppelin ‣ Morecoolstuffintheworks