layout | title |
---|---|
global |
Python Programming Guide |
The Spark Python API (PySpark) exposes the Spark programming model to Python. To learn the basics of Spark, we recommend reading through the Scala programming guide first; it should be easy to follow even if you don't know Scala. This guide will show how to use the Spark features described there in Python.
There are a few key differences between the Python and Scala APIs:
- Python is dynamically typed, so RDDs can hold objects of multiple types.
- PySpark does not yet support a few API calls, such as
lookup
and non-text input files, though these will be added in future releases.
In PySpark, RDDs support the same methods as their Scala counterparts but take Python functions and return Python collection types.
Short functions can be passed to RDD methods using Python's lambda
syntax:
{% highlight python %} logData = sc.textFile(logFile).cache() errors = logData.filter(lambda line: "ERROR" in line) {% endhighlight %}
You can also pass functions that are defined with the def
keyword; this is useful for longer functions that can't be expressed using lambda
:
{% highlight python %} def is_error(line): return "ERROR" in line errors = logData.filter(is_error) {% endhighlight %}
Functions can access objects in enclosing scopes, although modifications to those objects within RDD methods will not be propagated back:
{% highlight python %} error_keywords = ["Exception", "Error"] def is_error(line): return any(keyword in line for keyword in error_keywords) errors = logData.filter(is_error) {% endhighlight %}
PySpark will automatically ship these functions to executors, along with any objects that they reference. Instances of classes will be serialized and shipped to executors by PySpark, but classes themselves cannot be automatically distributed to executors. The Standalone Use section describes how to ship code dependencies to executors.
In addition, PySpark fully supports interactive use---simply run ./bin/pyspark
to launch an interactive shell.
PySpark requires Python 2.6 or higher. PySpark applications are executed using a standard CPython interpreter in order to support Python modules that use C extensions. We have not tested PySpark with Python 3 or with alternative Python interpreters, such as PyPy or Jython.
By default, PySpark requires python
to be available on the system PATH
and use it to run programs; an alternate Python executable may be specified by setting the PYSPARK_PYTHON
environment variable in conf/spark-env.sh
(or .cmd
on Windows).
All of PySpark's library dependencies, including Py4J, are bundled with PySpark and automatically imported.
Standalone PySpark applications should be run using the bin/spark-submit
script, which automatically
configures the Java and Python environment for running Spark.
The bin/pyspark
script launches a Python interpreter that is configured to run PySpark applications. To use pyspark
interactively, first build Spark, then launch it directly from the command line without any options:
{% highlight bash %} $ sbt/sbt assembly $ ./bin/pyspark {% endhighlight %}
The Python shell can be used explore data interactively and is a simple way to learn the API:
{% highlight python %}
words = sc.textFile("/usr/share/dict/words") words.filter(lambda w: w.startswith("spar")).take(5) [u'spar', u'sparable', u'sparada', u'sparadrap', u'sparagrass'] help(pyspark) # Show all pyspark functions {% endhighlight %}
By default, the bin/pyspark
shell creates SparkContext that runs applications locally on all of
your machine's logical cores.
To connect to a non-local cluster, or to specify a number of cores, set the MASTER
environment variable.
For example, to use the bin/pyspark
shell with a standalone Spark cluster:
{% highlight bash %} $ MASTER=spark://IP:PORT ./bin/pyspark {% endhighlight %}
Or, to use exactly four cores on the local machine:
{% highlight bash %} $ MASTER=local[4] ./bin/pyspark {% endhighlight %}
It is also possible to launch the PySpark shell in IPython, the
enhanced Python interpreter. PySpark works with IPython 1.0.0 and later. To
use IPython, set the IPYTHON
variable to 1
when running bin/pyspark
:
{% highlight bash %} $ IPYTHON=1 ./bin/pyspark {% endhighlight %}
Alternatively, you can customize the ipython
command by setting IPYTHON_OPTS
. For example, to launch
the IPython Notebook with PyLab graphing support:
{% highlight bash %} $ IPYTHON_OPTS="notebook --pylab inline" ./bin/pyspark {% endhighlight %}
IPython also works on a cluster or on multiple cores if you set the MASTER
environment variable.
PySpark can also be used from standalone Python scripts by creating a SparkContext in your script and running the script using bin/spark-submit
.
The Quick Start guide includes a complete example of a standalone Python application.
Code dependencies can be deployed by passing .zip or .egg files in the --py-files
option of spark-submit
:
{% highlight bash %} ./bin/spark-submit --py-files lib1.zip,lib2.zip my_script.py {% endhighlight %}
Files listed here will be added to the PYTHONPATH
and shipped to remote worker machines.
Code dependencies can also be added to an existing SparkContext at runtime using its addPyFile()
method.
You can set configuration properties by passing a SparkConf object to SparkContext:
{% highlight python %} from pyspark import SparkConf, SparkContext conf = (SparkConf() .setAppName("My app") .set("spark.executor.memory", "1g")) sc = SparkContext(conf = conf) {% endhighlight %}
spark-submit
supports launching Python applications on standalone, Mesos or YARN clusters, through
its --master
argument. However, it currently requires the Python driver program to run on the local
machine, not the cluster (i.e. the --deploy-mode
parameter cannot be cluster
).
API documentation for PySpark is available as Epydoc. Many of the methods also contain doctests that provide additional usage examples.
MLlib is also available in PySpark. To use it, you'll need NumPy version 1.4 or newer. The MLlib guide contains some example applications.
PySpark also includes several sample programs in the examples/src/main/python
folder.
You can run them by passing the files to pyspark
; e.g.:
./bin/spark-submit examples/src/main/python/wordcount.py
Each program prints usage help when run without arguments.