This document provides a summary of a presentation on scaling Apache Spark. It discusses techniques for reusing RDDs through caching, persistence levels and checkpointing. It also covers best practices for working with key-value data to avoid problems from groupByKey, and using Spark SQL and accumulators. Finally, it previews bringing code generation to Spark ML to improve performance.

1. Beyond Shuffling
tips & tricks for scaling Apache Spark
Vancouver
Spark 2015

2. Who am I?
● My name is Holden Karau
● Prefered pronouns are she/her
● I’m a Software Engineer at IBM
● previously Alpine, Databricks, Google, Foursquare & Amazon
● co-author of Learning Spark & Fast Data processing with Spark
○ co-author of a new book focused on Spark performance coming out next year*
● @holdenkarau
● Slide share http://www.slideshare.net/hkarau
● Linkedin https://www.linkedin.com/in/holdenkarau
● Github https://github.com/holdenk
● Spark Videos http://bit.ly/holdenSparkVideos

3. What is going to be covered:
● What I think I might know about you
● RDD re-use (caching, persistence levels, and checkpointing)
● Working with key/value data
○ Why group key is evil and what we can do about it
● Best practices for Spark accumulators*
● When Spark SQL can be amazing and wonderful
● A quick detour into some future performance work in Spark MLLib

4. Who I think you wonderful humans are?
● Nice* people
● Know some Apache Spark
● Want to scale your Apache Spark jobs
Lori Erickson

5. Cat photo from http://galato901.deviantart.com/art/Cat-on-Work-Break-173043455
Photo from Cocoa Dream

6. RDD re-use - sadly not magic
● If we know we are going to re-use the RDD what should we do?
○ If it fits nicely in memory caching in memory
○ persisting at another level
■ MEMORY, MEMORY_ONLY_SER, MEMORY_AND_DISK,
MEMORY_AND_DISK_SER
○ checkpointing
● Noisey clusters
○ _2 & checkpointing can help
Richard Gillin

7. Considerations for Key/Value Data
● What does the distribution of keys look like?
● What type of aggregations do we need to do?
● Do we want our data in any particular order?
● Are we joining with another RDD?
● Whats our partitioner?
○ If we don’t have an explicit one: what is the partition structure?
eleda 1

8. What is key skew and why do we care?
● Keys aren’t evenly distributed
○ Sales by zip code, or records by city, etc.
● groupByKey will explode (but it's pretty easy to break)
● We can have really unbalanced partitions
○ If we have enough key skew sortByKey could even fail
○ Stragglers (uneven sharding can make some tasks take much longer)
Mitchell
Joyce

9. groupByKey - just how evil is it?
● Pretty evil
● Groups all of the records with the same key into a single record
○ Even if we immediately reduce it (e.g. sum it or similar)
○ This can be too big to fit in memory, then our job fails
● Unless we are in SQL then happy pandas
PROgeckoam

10. So what does that look like?
(94110, A, B)
(94110, A, C)
(10003, D, E)
(94110, E, F)
(94110, A, R)
(10003, A, R)
(94110, D, R)
(94110, E, R)
(94110, E, R)
(67843, T, R)
(94110, T, R)
(94110, T, R)
(67843, T, R)(10003, A, R)
(94110, [(A, B), (A, C), (E, F), (A, R), (D, R), (E, R), (E, R), (T, R) (T, R)]

11. Let’s revisit wordcount with groupByKey
val words = rdd.flatMap(_.split(" "))
val wordPairs = words.map((_, 1))
val grouped = wordPairs.groupByKey()
grouped.mapValues(_.sum)

12. And now back to the “normal” version
val words = rdd.flatMap(_.split(" "))
val wordPairs = words.map((_, 1))
val wordCounts = wordPairs.reduceByKey(_ + _)
wordCounts

13. Let’s see what it looks like when we run the two
Quick pastebin of the code for the two: http://pastebin.com/CKn0bsqp
val rdd = sc.textFile("python/pyspark/*.py", 20) // Make sure we have many partitions
// Evil group by key version
val words = rdd.flatMap(_.split(" "))
val wordPairs = words.map((_, 1))
val grouped = wordPairs.groupByKey()
val evilWordCounts = grouped.mapValues(_.sum)
evilWordCounts.take(5)
// Less evil version
val wordCounts = wordPairs.reduceByKey(_ + _)
wordCounts.take(5)

14. GroupByKey

15. reduceByKey

16. So what did we do instead?
● reduceByKey
○ Works when the types are the same (e.g. in our summing version)
● aggregateByKey
○ Doesn’t require the types to be the same (e.g. computing stats model or similar)
Allows Spark to pipeline the reduction & skip making the list
We also got a map-side reduction (note the difference in shuffled read)

17. So why did we read in python/*.py
If we just read in the standard README.md file there aren’t enough duplicated
keys for the reduceByKey & groupByKey difference to be really apparent
Which is why groupByKey can be safe sometimes

18. Can just the shuffle cause problems?
● Sorting by key can put all of the records in the same partition
● We can run into partition size limits (around 2GB)
● Or just get bad performance
● So we can handle data like the above we can add some “junk” to our key
(94110, A, B)
(94110, A, C)
(10003, D, E)
(94110, E, F)
(94110, A, R)
(10003, A, R)
(94110, D, R)
(94110, E, R)
(94110, E, R)
(67843, T, R)
(94110, T, R)
(94110, T, R)
PROTodd
Klassy

19. Shuffle explosions :(
(94110, A, B)
(94110, A, C)
(10003, D, E)
(94110, E, F)
(94110, A, R)
(10003, A, R)
(94110, D, R)
(94110, E, R)
(94110, E, R)
(67843, T, R)
(94110, T, R)
(94110, T, R)
(94110, A, B)
(94110, A, C)
(94110, E, F)
(94110, A, R)
(94110, D, R)
(94110, E, R)
(94110, E, R)
(94110, T, R)
(94110, T, R)
(67843, T, R)(10003, A, R)

20. Spark accumulators
● Really “great” way for keeping track of failed records
● Double counting makes things really tricky
○ Jobs which worked “fine” don’t continue to work “fine” when minor changes happen
● Relative rules can save us* under certain conditions
Found Animals Foundation Follow

21. Using an accumulator for validation:
val (ok, bad) = (sc.accumulator(0), sc.accumulator(0))
val records = input.map{ x => if (isValid(x)) ok +=1 else bad += 1
// Actual parse logic here
}
// An action (e.g. count, save, etc.)
if (bad.value > 0.1* ok.value) {
throw Exception("bad data - do not use results")
// Optional cleanup
}
// Mark as safe
P.S: If you are interested in this check out spark-validator (still early stages).
Found Animals Foundation Follow

22. Using a library: simple historic validation Photo by Dvortygirl
val vc = new ValidationConf(jobHistoryPath, "1", true,
List[ValidationRule](new AvgRule("acc", 0.001, Some(200))))
val v = Validation(sc, vc)
// Some job logic
// Register an accumulator (optional)
val acc = sc.accumulator(0)
v.registerAccumulator(acc, "acc")
// More Job logic goes here
if (v.validate(jobId)) {
// Success logic goes here
} else sadness()

23. With a Spark internal counter...
val vc = new ValidationConf(tempPath, "1", true,
List[ValidationRule](
new AbsoluteSparkCounterValidationRule("recordsRead", Some(30), Some
(1000)))
)
val sqlCtx = new SQLContext(sc)
val v = Validation(sc, sqlCtx, vc)
//Do work here....
assert(v.validate(5) === true)
}
Photo by Dvortygirl

24. Where can Spark SQL benefit perf?
● Structured or semi-structured data
● OK with having less* complex operations available to us
● We may only need to operate on a subset of the data
○ The fastest data to process isn’t even read
● Remember that non-magic cat? Its got some magic** now
○ In part from peeking inside of boxes
● non-JVM (aka Python & R) users: saved from double serialization cost! :)
**Magic may cause stack overflow. Not valid in all states. Consult local magic bureau before attempting
magic
Matti Mattila

25. Why is Spark SQL good for those things?
● Space efficient columnar cached representation
● Able to push down operations to the data store
● Optimizer is able to look inside of our operations
○ Regular spark can’t see inside our operations to spot the difference between (min(_, _)) and
(append(_, _))
Matti Mattila

26. Preview: bringing codegen to Spark ML
● Based on Spark SQL’s code generation
○ First draft using quasiquotes
○ Switch to janino for Java compilation
● Initial draft for Gradient Boosted Trees
○ Based on DB’s work
○ First draft with QuasiQuotes
■ Moved to Java for speed
○ See SPARK-10387 for the details
Jon

27. @Override
public double call(Vector input) throws
Exception {
if (input.apply(1) <= 1.0) {
return 0.1;
} else {
if (input.apply(0) <= 0.5) {
return 0.0;
} else {
return 2.0;
}
}
}
(1, 1.0)
0.1 (0, 0.5)
0.0 2.0
What the generated code looks like: Glenn Simmons

28. Everyone* needs reduce, let’s make it faster!
● reduce & aggregate have “tree” versions
● we already had free map-side reduction
● but now we can get even better!**
**And we might be able to make even cooler versions

29. Additional Resources
● Programming guide (along with JavaDoc, PyDoc,
ScalaDoc, etc.)
○ http://spark.apache.org/docs/latest/
● Books
● Videos
● Denny’s meetup on Wednesday :)
● Spark Office Hours
○ follow me on twitter for future ones - https://twitter.com/holdenkarau
○ fill out this survey to choose the next date - http://bit.ly/spOffice1
raider of gin

30. Learning Spark
Fast Data
Processing with
Spark
(Out of Date)
Fast Data
Processing with
Spark
(2nd edition)
Advanced
Analytics with
Spark
Coming soon:
Spark in Action

31. And the next book…..
Still being written - signup to be notified when it is available:
● http://www.highperformancespark.com
● https://twitter.com/highperfspark

32. Q&A OR A quick detour into spark testing?
● It's like a choose your own adventure novel, but with
voting
● But more like the voting in High School since if we are
running out of time we might just skip it

33. Spark Videos
● Apache Spark Youtube Channel
● My Spark videos on YouTube -
○ http://bit.ly/holdenSparkVideos
● Spark Summit 2014 training
● Paco’s Introduction to Apache Spark

34. Cat wave photo by Quinn Dombrowski
k thnx bye!
If you care about Spark testing and
don’t hate surveys: http://bit.
ly/holdenTestingSpark
Will tweet results
“eventually” @holdenkarau