This documentation is for an out-of-date version of Apache Flink.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

Connectors

This page describes how to use connectors in PyFlink and highlights the details to be aware of when using Flink connectors in Python programs.

Note For general connector information and common configuration, please refer to the corresponding Java/Scala documentation.

Download connector and format jars

Since Flink is a Java/Scala-based project, for both connectors and formats, implementations are available as jars that need to be specified as job dependencies.

table_env.get_config().get_configuration().set_string("pipeline.jars", "file:///my/jar/path/connector.jar;file:///my/jar/path/json.jar")

How to use connectors

In PyFink’s Table API, DDL is the recommended way to define sources and sinks, executed via the execute_sql() method on the TableEnvironment. This makes the table available for use by the application.

source_ddl = """
        CREATE TABLE source_table(
            a VARCHAR,
            b INT
        ) WITH (
          'connector' = 'kafka',
          'topic' = 'source_topic',
          'properties.bootstrap.servers' = 'kafka:9092',
          'properties.group.id' = 'test_3',
          'scan.startup.mode' = 'latest-offset',
          'format' = 'json'
        )
        """

sink_ddl = """
        CREATE TABLE sink_table(
            a VARCHAR
        ) WITH (
          'connector' = 'kafka',
          'topic' = 'sink_topic',
          'properties.bootstrap.servers' = 'kafka:9092',
          'format' = 'json'
        )
        """

t_env.execute_sql(source_ddl)
t_env.execute_sql(sink_ddl)

t_env.sql_query("SELECT a FROM source_table") \
    .insert_into("sink_table")

Below is a complete example of how to use a Kafka source/sink and the JSON format in PyFlink.

from pyflink.datastream import StreamExecutionEnvironment, TimeCharacteristic
from pyflink.table import StreamTableEnvironment, EnvironmentSettings


def log_processing():
    env = StreamExecutionEnvironment.get_execution_environment()
    env_settings = EnvironmentSettings.Builder().use_blink_planner().build()
    t_env = StreamTableEnvironment.create(stream_execution_environment=env, environment_settings=env_settings)
    t_env.get_config().get_configuration().set_boolean("python.fn-execution.memory.managed", True)
    # specify connector and format jars
    t_env.get_config().get_configuration().set_string("pipeline.jars", "file:///my/jar/path/connector.jar;file:///my/jar/path/json.jar")
    
    source_ddl = """
            CREATE TABLE source_table(
                a VARCHAR,
                b INT
            ) WITH (
              'connector' = 'kafka',
              'topic' = 'source_topic',
              'properties.bootstrap.servers' = 'kafka:9092',
              'properties.group.id' = 'test_3',
              'scan.startup.mode' = 'latest-offset',
              'format' = 'json'
            )
            """

    sink_ddl = """
            CREATE TABLE sink_table(
                a VARCHAR
            ) WITH (
              'connector' = 'kafka',
              'topic' = 'sink_topic',
              'properties.bootstrap.servers' = 'kafka:9092',
              'format' = 'json'
            )
            """

    t_env.execute_sql(source_ddl)
    t_env.execute_sql(sink_ddl)

    t_env.sql_query("SELECT a FROM source_table") \
        .insert_into("sink_table")

    t_env.execute("payment_demo")


if __name__ == '__main__':
    log_processing()

Predefined Sources and Sinks

Some data sources and sinks are built into Flink and are available out-of-the-box. These predefined data sources include reading from Pandas DataFrame, or ingesting data from collections. The predefined data sinks support writing to Pandas DataFrame.

from/to Pandas

PyFlink Tables support conversion to and from Pandas DataFrame.

import pandas as pd
import numpy as np

# Create a PyFlink Table
pdf = pd.DataFrame(np.random.rand(1000, 2))
table = t_env.from_pandas(pdf, ["a", "b"]).filter("a > 0.5")

# Convert the PyFlink Table to a Pandas DataFrame
pdf = table.to_pandas()

from_elements()

from_elements() is used to create a table from a collection of elements. The element types must be acceptable atomic types or acceptable composite types.

table_env.from_elements([(1, 'Hi'), (2, 'Hello')])

# use the second parameter to specify custom field names
table_env.from_elements([(1, 'Hi'), (2, 'Hello')], ['a', 'b'])

# use the second parameter to specify a custom table schema
table_env.from_elements([(1, 'Hi'), (2, 'Hello')],
                        DataTypes.ROW([DataTypes.FIELD("a", DataTypes.INT()),
                                       DataTypes.FIELD("b", DataTypes.STRING())]))

The above query returns a Table like:

+----+-------+
| a  |   b   |
+====+=======+
| 1  |  Hi   |
+----+-------+
| 2  | Hello |
+----+-------+

User-defined sources & sinks

In some cases, you may want to define custom sources and sinks. Currently, sources and sinks must be implemented in Java/Scala, but you can define a TableFactory to support their use via DDL. More details can be found in the Java/Scala documentation.