PySpark Meetup Talk

Python API for Spark
Josh Rosen UC#BERKELEY#

What is Spark?
Fast and expressive cluster computing system
Compatible with Hadoop-‐supported ﬁle systems

and data formats (HDFS, S3, SequenceFile, ...)
Improves efficiency through in-
memory computing primitives
and general computation
graphs
As much as 30x faster

Improves usability through rich
APIs in Scala, Python, and
Java, and an interactive shell
Often 2-10x less code

RDDs
Resilient Distributed Datasets
Immutable, partitioned
collections of objects
Transformations
map
filter
groupBy
join
...
Actions
count
collect
save
...
Example: Log Mining
val lines = spark.textFile(“hdfs://...”)

val errors = lines.filter(_.startsWith(“ERROR”))
val messages = errors.map(_.split(‘\t’)(2))
messages.filter(_.contains(“foo”)).count
What is PySpark?
PySpark at a Glance
Write Spark jobs Run interactive Supports C

in Python jobs in the shell extensions
Previewed at AMP Camp 2012
Available now in 0.7 release

Example: Word Count
from pyspark.context import SparkContext
sc = SparkContext(...)
lines = sc.textFile(sys.argv[2], 1)
counts = lines.flatMap(lambda x: x.split(' ')) \
.map(lambda x: (x, 1)) \
.reduceByKey(lambda x, y: x + y)
for (word, count) in counts.collect():

print "%s : %i" % (word, count)
Demo
Implementation
Built on top of Java API
PySpark
Java API
Spark
Local
Mesos Standalone YARN
Mode
Process data
in Python
and persist /
transfer it in Java
Re-uses Spark’s scheduling
broadcast
checkpointing
networking
fault-recovery
HDFS access
PySpark has a small codebase:
-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐
File blank comment code
-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐
python/pyspark/rdd.py 115 345 302
core/src/main/scala/spark/api/python/PythonRDD.scala 33 45 231
python/pyspark/context.py 32 101 133
python/pyspark/tests.py 26 11 84
python/pyspark/accumulators.py 37 91 70
python/pyspark/serializers.py 21 7 55
python/pyspark/join.py 15 27 50
python/pyspark/worker.py 8 7 44
core/src/main/scala/spark/api/python/PythonPartitioner.scala 5 9 34
pyspark 9 8 27
python/pyspark/java_gateway.py 5 7 26
python/pyspark/files.py 7 14 17
python/pyspark/broadcast.py 8 16 15
python/pyspark/shell.py 4 6 8
python/pyspark/__init__.py 6 14 7
-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐
SUM: 331 708 1103
-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐
< 2K lines, including comments

Data Flow
Local Cluster
Python JVM
Data Flow
Spark
Context
Local Cluster
Python JVM
Data Flow
Py4J
Socket Spark
Context
Spark
Context
Local Cluster
Python JVM
Data Flow
Py4J
Socket Spark
Context
Spark
Context
Local
FS
Local Cluster
Python JVM
Data Flow
Py4J
Spark
Socket Spark
Worker
Context
Spark
Context
Spark
Local
Worker
FS
Local Cluster
Python JVM
Data Flow
Pipe
Py4J Python
Spark
Socket Spark Python
Worker
Context Python
Spark
Context
Python
Spark
Local Python
Worker
FS Python
Local Cluster
Python JVM
Data is stored as Pickled
objects in RDD[Array[Byte]]
Storing batches of Python
objects in one Scala object
reduces overhead
When possible, RDD
transformations are pipelined:
lines.flatMap(lambda x: x.split(' ')) \
.map(lambda x: (x, 1))
MappedRDD MappedRDD
func(x) = x.split(‘’) func(x) = (x, 1)
MappedRDD
func(x) = (x.split(‘’), 1)
Python functions and closures
are serialized using PiCloud’s
CloudPickle module
Roadmap
Available in Spark 0.7
Thanks!
Bonus Slides
Pickle is a miniature stack
language
>>> x = ["Hello", "World!"]
>>> pickletools.dis(cPickle.dumps(x, 2))
0: \x80 PROTO 2
2: ] EMPTY_LIST
3: q BINPUT 1
5: ( MARK
6: U SHORT_BINSTRING 'Hello'
13: q BINPUT 2
15: U SHORT_BINSTRING 'World!'
23: q BINPUT 3
25: e APPENDS (MARK at 5)
26: . STOP
highest protocol among opcodes = 2
You can do crazy stuff, like
converting a collection of
pickled objects into a pickled
collection.
https://gist.github.com/JoshRosen/3384191
Bulk depickling can be faster
even if it involves Pickle
opcode manipulation:
10000 integers:
Bulk depickle (chunk size = 2): 0.266709804535
Individual depickle: 0.0540158748627
10000 dicts (dict([ (str(n), n) for n in range(100) ])):

Individual depickle: 2.44124102592
https://gist.github.com/JoshRosen/3401373

PySpark Meetup Talk

Uploaded by

Copyright:

Available Formats

PySpark Meetup Talk

Uploaded by

Document Information

Original Description:

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

PySpark Meetup Talk

Uploaded by

Copyright:

Available Formats

Python API for Spark

Josh Rosen UC#BERKELEY#

Compatible with Hadoop-‐supported ﬁle systems

As much as 30x faster

Often 2-10x less code

val lines = spark.textFile(“hdfs://...”)

Write Spark jobs Run interactive Supports C

Available now in 0.7 release

for (word, count) in counts.collect():

< 2K lines, including comments

10000 dicts (dict([ (str(n), n) for n in range(100) ])):

You might also like

PySpark Meetup Talk

Uploaded by

Copyright:

Available Formats

PySpark Meetup Talk

Uploaded by

Document Information

Original Description:

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

PySpark Meetup Talk

Uploaded by

Copyright:

Available Formats

Python API for Spark

Josh Rosen UC#BERKELEY#

Compatible with Hadoop-­‐supported ﬁle systems

As much as 30x faster

Often 2-10x less code

val lines = spark.textFile(“hdfs://...”)

Write Spark jobs Run interactive Supports C

Available now in 0.7 release

for (word, count) in counts.collect():

< 2K lines, including comments

10000 dicts (dict([ (str(n), n) for n in range(100) ])):

You might also like

Compatible with Hadoop-‐supported ﬁle systems