This documentation is for an out-of-date version of Apache Flink. We recommend you use the latest stable version.

Python Programming Guide (Streaming) Beta

Attention This API is based on Jython, which is not a full Python replacement and may restrict the libraries you are able to use with your application (see below for more information).

Analysis streaming programs in Flink are regular programs that implement transformations on streaming data sets (e.g., filtering, mapping, joining, grouping). The streaming data sets are initially created from certain sources (e.g., by reading files, or from collections). Results are returned via sinks, which may for example write the data to (distributed) files, or to standard output (for example the command line terminal). Flink streaming programs run in a variety of contexts, standalone, or embedded in other programs. The execution can happen in a local JVM, or on clusters of many machines.

In order to create your own Flink streaming program, we encourage you to start with the program skeleton and gradually add your own transformations. The remaining sections act as references for additional operations and advanced features.

Jython Framework

Flink Python streaming API uses Jython framework (see to drive the execution of a given script. The Python streaming layer, is actually a thin wrapper layer for the existing Java streaming APIs.


There are two main constraints for using Jython:

  • The latest Python supported version is 2.7
  • It is not straightforward to use Python C extensions, which may prevent use of some libraries

(For more information please see

One possible alternative for streaming that allows for native Python execution would be the Apache Beam portability framework with the Flink runner.

Streaming Program Example

The following streaming program is a complete, working example of WordCount. You can copy & paste the code to run it locally (see notes later in this section). It counts the number of each word (case insensitive) in a stream of sentences, on a window size of 50 milliseconds and prints the results into the standard output.

from org.apache.flink.streaming.api.functions.source import SourceFunction
from org.apache.flink.api.common.functions import FlatMapFunction, ReduceFunction
from import KeySelector
from org.apache.flink.streaming.api.windowing.time.Time import milliseconds

class Generator(SourceFunction):
    def __init__(self, num_iters):
        self._running = True
        self._num_iters = num_iters

    def run(self, ctx):
        counter = 0
        while self._running and counter < self._num_iters:
            ctx.collect('Hello World')
            counter += 1

    def cancel(self):
        self._running = False

class Tokenizer(FlatMapFunction):
    def flatMap(self, value, collector):
        for word in value.lower().split():
            collector.collect((1, word))

class Selector(KeySelector):
    def getKey(self, input):
        return input[1]

class Sum(ReduceFunction):
    def reduce(self, input1, input2):
        count1, word1 = input1
        count2, word2 = input2
        return (count1 + count2, word1)

def main(factory):
    env = factory.get_execution_environment()
    env.create_python_source(Generator(num_iters=1000)) \
        .flat_map(Tokenizer()) \
        .key_by(Selector()) \
        .time_window(milliseconds(50)) \
        .reduce(Sum()) \


  • Execution on a multi-node cluster requires a shared medium storage, which needs to be configured (.e.g HDFS) upfront.
  • The output from of the given script is directed to the standard output. Consequently, the output is written to the corresponding worker .out file. If the script is executed inside the IntelliJ IDE, then the output will be displayed in the console tab.

Back to top

Program Skeleton

As we already saw in the example, Flink streaming programs look like regular Python programs. Each program consists of the same basic parts:

  1. A main(factory) function definition, with an environment factory argument - the program entry point,
  2. Obtain an Environment from the factory,
  3. Load/create the initial data,
  4. Specify transformations on this data,
  5. Specify where to put the results of your computations, and
  6. Execute your program.

We will now give an overview of each of those steps but please refer to the respective sections for more details.

The main(factory) function is a must and it is used by Flink execution layer to run the given Python streaming program.

The Environment is the basis for all Flink programs. You can obtain one using the factory methods provided by the factory:


For specifying data sources the streaming execution environment has several methods. To just read a text file as a sequence of lines, you can use:

env = factory.get_execution_environment()
text = env.read_text_file("file:///path/to/file")

This will give you a DataStream on which you can then apply transformations. For more information on data sources and input formats, please refer to Data Sources.

Once you have a DataStream you can apply transformations to create a new DataStream which you can then write to a file, transform again, or combine with other DataStreams. You apply transformations by calling methods on DataStream with your own custom transformation function. For example, a map transformation looks like this:

class Doubler(MapFunction):
    def map(self, value):
        return value * 2

This will create a new DataStream by doubling every value in the original DataStream. For more information and a list of all the transformations, please refer to Transformations.

Once you have a DataStream that needs to be written to disk you can call one of these methods on DataStream:

data.write_as_text("<file-path>", mode=WriteMode.OVERWRITE)

The last method is only useful for developing/debugging on a local machine, it will output the contents of the DataSet to standard output. (Note that in a cluster, the result goes to the standard out stream of the cluster nodes and ends up in the .out files of the workers). The first two do as the name suggests. Please refer to Data Sinks for more information on writing to files.

Once you specified the complete program you need to call execute on the Environment. This will either execute on your local machine or submit your program for execution on a cluster, depending on how Flink was started.

Back to top

Project setup

Apart from setting up Flink, no additional work is required. Using Jython to execute the Python script, means that no external packages are needed and the program is executed as if it was a jar file.

The Python API was tested on Windows/Linux/OSX systems.

Back to top

Lazy Evaluation

All Flink programs are executed lazily: When the program’s main method is executed, the data loading and transformations do not happen directly. Rather, each operation is created and added to the program’s plan. The operations are actually executed when one of the execute() methods is invoked on the Environment object. Whether the program is executed locally or on a cluster depends on the environment of the program.

The lazy evaluation lets you construct sophisticated programs that Flink executes as one holistically planned unit.

Back to top


Data transformations transform one or more DataStreams into a new DataStream. Programs can combine multiple transformations into sophisticated assemblies.

This section gives a brief overview of the available transformations. The transformations documentation has a full description of all transformations with examples.

Transformation Description
PythonDataStream → PythonDataStream

Takes one element and produces one element.

class Doubler(MapFunction):
    def map(self, value):
        return value * 2
PythonDataStream → PythonDataStream

Takes one element and produces zero, one, or more elements.

class Tokenizer(FlatMapFunction):
    def flatMap(self, word, collector):
        collector.collect((1, word))

PythonDataStream → PythonDataStream

Evaluates a boolean function for each element and retains those for which the function returns true.

class GreaterThen1000(FilterFunction):
    def filter(self, value):
        return value > 1000

PythonDataStream → PythonKeyedStream

Logically partitions a stream into disjoint partitions, each partition containing elements of the same key. Internally, this is implemented with hash partitioning. See keys on how to specify keys. This transformation returns a PythonKeyedDataStream.

class Selector(KeySelector):
    def getKey(self, input):
        return input[1]  # Key by the second element in a tuple

data_stream.key_by(Selector()) // Key by field "someKey"
PythonKeyedStream → PythonDataStream

A "rolling" reduce on a keyed data stream. Combines the current element with the last reduced value and emits the new value.

class Sum(ReduceFunction):
    def reduce(self, input1, input2):
        count1, val1 = input1
        count2, val2 = input2
        return (count1 + count2, val1)

PythonKeyedStream → PythonWindowedStream

Windows can be defined on already partitioned KeyedStreams. Windows group the data in each key according to some characteristic (e.g., the data that arrived within the last 5 seconds). See windows for a complete description of windows.

keyed_stream.count_window(10, 5)  # Last 10 elements, sliding (jumping) by 5 elements

keyed_stream.time_window(milliseconds(30))  # Last 30 milliseconds of data

keted_stream.time_window(milliseconds(100), milliseconds(20))  # Last 100 milliseconds of data, sliding (jumping) by 20 milliseconds

Window Apply
PythonWindowedStream → PythonDataStream

Applies a general function to the window as a whole. Below is a function that manually sums the elements of a window.

class WindowSum(WindowFunction):
    def apply(self, key, window, values, collector):
        sum = 0
        for value in values:
            sum += value[0]
        collector.collect((key, sum))

Window Reduce
PythonWindowedStream → PythonDataStream

Applies a functional reduce function to the window and returns the reduced value.

class Sum(ReduceFunction):
    def reduce(self, input1, input2):
        count1, val1 = input1
        count2, val2 = input2
        return (count1 + count2, val1)

PythonDataStream* → PythonDataStream

Union of two or more data streams creating a new stream containing all the elements from all the streams. Note: If you union a data stream with itself you will get each element twice in the resulting stream.

data_stream.union(other_stream1, other_stream2, ...);
PythonDataStream → PythonSplitStream

Split the stream into two or more streams according to some criterion.

class StreamSelector(OutputSelector):
    def select(self, value):
        return ["even"] if value % 2 == 0 else ["odd"]

splited_stream = data_stream.split(StreamSelector())

SplitStream → DataStream

Select one or more streams from a split stream.

even_data_stream ="even")
odd_data_stream ="odd")
all_data_stream ="even", "odd")

PythonDataStream → PythonIterativeStream → PythonDataStream

Creates a "feedback" loop in the flow, by redirecting the output of one operator to some previous operator. This is especially useful for defining algorithms that continuously update a model. The following code starts with a stream and applies the iteration body continuously. Elements that are greater than 0 are sent back to the feedback channel, and the rest of the elements are forwarded downstream. See iterations for a complete description.

class MinusOne(MapFunction):
    def map(self, value):
        return value - 1

class PositiveNumber(FilterFunction):
    def filter(self, value):
        return value > 0

class LessEquelToZero(FilterFunction):
    def filter(self, value):
        return value <= 0

iteration = initial_stream.iterate(5000)
iteration_body =
feedback = iteration_body.filter(PositiveNumber())
output = iteration_body.filter(LessEquelToZero())

Back to top

Certain operations require user-defined functions as arguments. All the functions should be defined as Python classes that derived from the relevant Flink function. User-defined functions are serialized and sent over to the TaskManagers for execution.

class Filter(FilterFunction):
    def filter(self, value):
        return value > 5


Rich functions (.e.g RichFilterFunction) enable to define (override) the optional operations: open & close. The user may use these functions for initialization and cleanups.

class Tokenizer(RichMapFunction):
    def open(self, config):
    def close(self):
    def map(self, value):

The open function is called by the worker before starting the streaming pipeline. The close function is called by the worker after the streaming pipeline is stopped.

Back to top

Data Types

Flink’s Python Streaming API offers support for primitive Python types (int, float, bool, string), as well as byte arrays and user-defined classes.

class Person:
    def __init__(self, name, age): = name
        self.age = age

class Tokenizer(MapFunction):
    def map(self, value):
        return (1, Person(*value))


You can use the tuples (or lists) for composite types. Python tuples are mapped to Jython native corresponding types, which are handled by the Python wrapper thin layer.

word_counts = env.from_elements(("hello", 1), ("world",2))

class Tokenizer(MapFunction):
    def map(self, value):
        return value[1]

counts =

Back to top

Data Sources

Data sources create the initial data streams, such as from files or from collections.


  • read_text_file(path) - reads files line wise and returns them as a stream of Strings.


  • from_elements(*args) - Creates a data stream from all the elements.
  • generate_sequence(from, to) - Generates the sequence of numbers in the given interval, in parallel.


env  = factory.get_execution_environment()

\# read text file from local files system
localLiens = env.read_text("file:///path/to/my/textfile")

\# read text file from a HDFS running at nnHost:nnPort
hdfsLines = env.read_text("hdfs://nnHost:nnPort/path/to/my/textfile")

\# create a set from some given elements
values = env.from_elements("Foo", "bar", "foobar", "fubar")

\# generate a number sequence
numbers = env.generate_sequence(1, 10000000)

Back to top

Data Sinks

Data sinks consume DataStreams and are used to store or return them:

  • write_as_text() - Writes elements line-wise as Strings. The Strings are obtained by calling the str() method of each element.
  • output() - Prints the str() value of each element on the standard out.
  • write_to_socket() - Writes the DataStream to a socket [host:port] as a byte array.

A DataStream can be input to multiple operations. Programs can write or print a data stream and at the same time run additional transformations on them.


Standard data sink methods:

 write DataStream to a file on the local file system

 write DataStream to a file on a HDFS with a namenode running at nnHost:nnPort

 write DataStream to a file and overwrite the file if it exists
textData.write_as_text("file:///my/result/on/localFS", WriteMode.OVERWRITE)

 this writes tuples in the text formatting "(a, b, c)", rather than as CSV lines

Back to top

Parallel Execution

This section describes how the parallel execution of programs can be configured in Flink. A Flink program consists of multiple tasks (operators, data sources, and sinks). A task is split into several parallel instances for execution and each parallel instance processes a subset of the task’s input data. The number of parallel instances of a task is called its parallelism or degree of parallelism (DOP).

The degree of parallelism of a task can be specified in Flink on different levels.

Execution Environment Level

Flink programs are executed in the context of an execution environment. An execution environment defines a default parallelism for all operators, data sources, and data sinks it executes. Execution environment parallelism can be overwritten by explicitly configuring the parallelism of an operator.

The default parallelism of an execution environment can be specified by calling the set_parallelism() method. To execute all operators, data sources, and data sinks of the WordCount example program with a parallelism of 3, set the default parallelism of the execution environment as follows:

env = factory.get_execution_environment()

text.flat_map(Tokenizer()) \
    .key_by(Selector()) \
    .time_window(milliseconds(30)) \
    .reduce(Sum()) \


System Level

A system-wide default parallelism for all execution environments can be defined by setting the parallelism.default property in ./conf/flink-conf.yaml. See the Configuration documentation for details.

Back to top

Executing Plans

To run the plan with Flink, go to your Flink distribution, and run the script from the /bin folder. The script containing the plan has to be passed as the first argument, followed by a number of additional Python packages, and finally, separated by -additional arguments that will be fed to the script.

./bin/ <Script>[ <pathToPackage1>[ <pathToPackageX]][ - <param1>[ <paramX>]]

Back to top