davidwrede commented on a change in pull request #13317:
URL: https://github.com/apache/beam/pull/13317#discussion_r523302017
##########
File path: website/www/site/content/en/documentation/programming-guide.md
##########
@@ -5431,4 +5431,280 @@ use case.
{{< highlight py >}}
{{< code_sample "sdks/python/apache_beam/examples/snippets/snippets.py"
BundleFinalize >}}
-{{< /highlight >}}
\ No newline at end of file
+{{< /highlight >}}
+
+## 13. Multi-language pipelines {#mulit-language-pipelines}
+
+Beam allows you to combine transforms written in any supported SDK language
(currently, Java and Python) and use them in one multi-language pipeline. This
capability makes it easy to provide new functionality simultaneously in
different Apache Beam SDKs through a single cross-language transform. For
example, the [Apache Kafka
connector](https://github.com/apache/beam/blob/master/sdks/python/apache_beam/io/kafka.py)
and [SQL
transform](https://github.com/apache/beam/blob/master/sdks/python/apache_beam/transforms/sql.py)
from the Java SDK can be used in Python streaming pipelines.
+
+Pipelines that use transforms from more than one SDK-language are known as
*multi-language pipelines*.
+
+### 13.1. Creating cross-language transforms {#create-x-lang-transforms}
+
+To make transforms written in one language available to pipelines written in
another language, an *expansion service* for that transform is used to create
and inject the appropriate language-specific pipeline fragments into your
pipeline.
+
+In the following example, a Python pipeline written the Apache Beam SDK for
Python starts up a local Java expansion service on your computer to create and
inject the appropriate Java pipeline fragments for executing the Java Kafka
cross-language transform into your Python pipeline. The SDK then downloads and
stages the necessary Java dependencies needed to execute these transforms.
+
+
+
+At runtime, the Beam runner will execute both Python and Java transforms to
execute your pipeline.
+
+In this section, we will use
[KafkaIO.Read](https://beam.apache.org/releases/javadoc/current/org/apache/beam/sdk/io/kafka/KafkaIO.Read.html)
to illustrate how to create a cross-language transform for Java and a test
example for Python.
+
+#### 13.1.1. Creating cross-language Java transforms
+
+To make your Apache Beam Java SDK transform portable across SDK languages, you
must implement two interfaces:
[ExternalTransformBuilder](https://github.com/apache/beam/blob/master/sdks/java/core/src/main/java/org/apache/beam/sdk/transforms/ExternalTransformBuilder.java)
and
[ExternalTransformRegistrar](https://github.com/apache/beam/blob/master/sdks/java/core/src/main/java/org/apache/beam/sdk/expansion/ExternalTransformRegistrar.java).
The `ExternalTransformBuilder` interface constructs the cross-language
transform using configuration values passed in from the pipeline and the
`ExternalTransformRegistrar` interface registers the cross-language transform
for use with the expansion service.
+
+**Implementing the interfaces**
+
+1. Define a Builder class for your transform that implements the
`ExternalTransformBuilder` interface and overrides the `buildExternal` method
that will be used to build your transform object. Initial configuration values
for your transform should be defined in the `buildExternal` method. In most
cases, it is convenient to make the Java transform builder class implement
`ExternalTransformBuilder`.
+
+ > **Note:** `ExternalTransformBuilder` requires you to define a
configuration object (a simple POJO) to capture a set of parameters sent by
external SDKs to initiate the Java transform. Usually these parameters directly
map to constructor parameters of the Java transform.
+
+ {{< highlight >}}
+@AutoValue.Builder
+abstract static class Builder<K, V>
+ implements ExternalTransformBuilder<External.Configuration, PBegin,
PCollection<KV<K, V>>> {
+ abstract Builder<K, V> setConsumerConfig(Map<String, Object> config);
+
+ abstract Builder<K, V> setTopics(List<String> topics);
+
+ /** Remaining property declarations omitted for clarity. */
+
+ abstract Read<K, V> build();
+
+ @Override
+ public PTransform<PBegin, PCollection<KV<K, V>>> buildExternal(
+ External.Configuration config) {
+ ImmutableList.Builder<String> listBuilder = ImmutableList.builder();
+ for (String topic : config.topics) {
+ listBuilder.add(topic);
+ }
+ setTopics(listBuilder.build());
+
+ /** Remaining property defaults omitted for clarity. */
+ }
+}
+ {{< /highlight >}}
+2. Register the transform as an external cross-language transform by defining
a class that implements `ExternalTransformRegistrar`. You must annotate your
class with the `AutoService` annotation to ensure that your transform is
registered and instantiated properly by the expansion service.
+3. In your registrar class, define a Uniform Resource Name (URN) for your
transform. The URN must be a unique string that identifies your transform with
the expansion service.
+4. From within your registrar class, define a configuration class for the
parameters used during the initialization of your transform by the external SDK.
+
+ The following example from the KafkaIO transform shows how to implement
steps two through four:
+
+ {{< highlight >}}
+@AutoService(ExternalTransformRegistrar.class)
+public static class External implements ExternalTransformRegistrar {
+
+ public static final String URN = "beam:external:java:kafka:read:v1";
+
+ @Override
+ public Map<String, Class<? extends ExternalTransformBuilder<?, ?, ?>>>
knownBuilders() {
+ return ImmutableMap.of(
+ URN,
+ (Class<? extends ExternalTransformBuilder<?, ?, ?>>)
+ (Class<?>) AutoValue_KafkaIO_Read.Builder.class);
+ }
+
+ /** Parameters class to expose the Read transform to an external SDK. */
+ public static class Configuration {
+ private Map<String, String> consumerConfig;
+ private List<String> topics;
+
+
+ public void setConsumerConfig(Map<String, String> consumerConfig) {
+ this.consumerConfig = consumerConfig;
+ }
+
+ public void setTopics(List<String> topics) {
+ this.topics = topics;
+ }
+
+ /** Remaining properties omitted for clarity. */
+ }
+}
+ {{< /highlight >}}
+
+After you have implemented the `ExternalTransformBuilder` and
`ExternalTransformRegistrar` interfaces, your transform can be registered and
created successfully by the default Java expansion service.
+
+**Using the expansion service**
+
+Java has a default expansion service included and available in the Apache Beam
Java SDK. You can write your own expansion service, but that is generally not
needed, so it is not covered in this section.
+
+Perform the following steps to start up a Java expansion service directly:
+
+{{< highlight >}}
+// Path to a JAR file that contains the transform to expand, cross-language
specific utilities (builder, registrar, etc.), and dependencies.
+$ export EXPANSION_SERVICE_JAR=<My_expansion_service_JAR>
+
+$ export PORT_FOR_EXPANSION_SERVICE=12345
+
+$ jar -jar $EXPANSION_SERVICE_JAR $PORT_FOR_EXPANSION_SERVICE
+{{< /highlight >}}
+
+When creating SDK-specific wrappers for your transform, SDKs may provide
utilities that are readily available for easily starting up an expansion
service. For example, the Python SDK provides the
[BeamJarExpansionService](https://github.com/apache/beam/blob/master/sdks/python/apache_beam/transforms/external.py)
utility for starting up a Java expansion service using a JAR file.
+
+**Including dependencies**
+
+If your transform requires external libraries, you can include them by adding
them to the classpath of the expansion service. After they are included in the
classpath, they will be staged when your transform is expanded by the expansion
service.
+
+**Writing SDK-specific wrappers**
+
+Your cross-language Java transform can be called through the lower-level
[ExternalTransform](https://github.com/apache/beam/blob/master/sdks/python/apache_beam/transforms/external.py)
class in a multi-language pipeline (as described in the next section);
however, if possible, you should create a SDK-specific wrapper written in the
programming language of the pipeline (such as Python) to access the transform
instead. This higher-level abstraction will make it easier for pipeline authors
to use your transform.
+
+To create an SDK wrapper for use in a Python pipeline, do the following:
+
+1. Create a Python module for your cross-language transform.
+2. In the module, build the payload that should be used to initiate the
cross-language transform expansion request using one of the available
[PayloadBuilder](https://github.com/apache/beam/blob/master/sdks/python/apache_beam/transforms/external.py)
classes.
+
+ The parameter names and types of the payload should map to parameter names
and types of the configuration POJO provided to the Java
`ExternalTransformBuilder`. Parameter types are mapped across SDKs using a
[Beam
schema](https://github.com/apache/beam/blob/master/model/pipeline/src/main/proto/schema.proto).
Parameter names are mapped by simply converting Python underscore-separated
variable names to camel-case (Java standard).
+
+ In the following example,
[kafka.py](https://github.com/apache/beam/blob/master/sdks/python/apache_beam/io/kafka.py)
uses `NamedTupleBasedPayloadBuilder` to build the payload. The parameters map
to the Java
[KafkaIO.External.Configuration](https://github.com/apache/beam/blob/master/sdks/java/io/kafka/src/main/java/org/apache/beam/sdk/io/kafka/KafkaIO.java)
config object defined previously in the **Implementing the interfaces**
section.
+
+ {{< highlight >}}
+class ReadFromKafkaSchema(typing.NamedTuple):
+ consumer_config: typing.Mapping[str, str]
+ topics: typing.List[str]
+ # Other properties omitted for clarity.
+
+payload = NamedTupleBasedPayloadBuilder(ReadFromKafkaSchema(...))
+ {{< /highlight >}}
+3. Start an expansion service unless one is specified by the pipeline creator.
The Apache Beam Python SDK provides a utility, `BeamJarExpansionService`, for
easily starting an expansion service based on a JAR released with Beam. To use
this, do the following:
+
+ 1. Add a Gradle target to Beam that can be used to build a shaded
expansion service JAR for the target Java transform. This target should produce
a Beam JAR that contains all dependencies needed for expanding the Java
transform and the JAR should be released with Beam.
+ 2. In your Python module, instantiate `BeamJarExpansionService` with the
Gradle target.
+
+ {{< highlight >}}
+ expansion_service =
BeamJarExpansionService('sdks:java:io:expansion-service:shadowJar')
+ {{< /highlight >}}
+4. Add a Python wrapper transform class that extends
[ExternalTransform](https://github.com/apache/beam/blob/master/sdks/python/apache_beam/transforms/external.py).
Pass the payload and expansion service defined above as parameters to the
constructor of the `ExternalTransform` parent class.
+
+#### 13.1.2. Creating cross-language Python transforms
+
+To make your Python transform usable with different SDK languages, you must
create a Python module that registers an existing Python transform as a
cross-language transform for use with the Python expansion service and calls
into that existing transform to perform its intended operation.
+
+**Defining the Python module**
+
+1. Define a Uniform Resource Name (URN) for your transform. The URN must be a
unique string that identifies your transform with the expansion service.
+
+ {{< highlight >}}
+TEST_COMPK_URN = "beam:transforms:xlang:test:compk"
+ {{< /highlight >}}
+2. For an existing Python transform, create a new class to register the URN
with the Python expansion service.
+
+ {{< highlight >}}
+@ptransform.PTransform.register_urn(TEST_COMPK_URN, None)
+class CombinePerKeyTransform(ptransform.PTransform):
+ {{< /highlight >}}
+3. From within the class, define an expand method that takes an input
PCollection, runs the Python transform, and then returns the output PCollection.
+
+ {{< highlight >}}
+def expand(self, pcoll):
+ return pcoll \
+ | beam.CombinePerKey(sum).with_output_types(
+ typing.Tuple[unicode, int])
+ {{< /highlight >}}
+4. As with other Python transforms, define a `to_runner_api_parameter` method
that returns the URN.
+
+ {{< highlight >}}
+def to_runner_api_parameter(self, unused_context):
+ return TEST_COMPK_URN, None
+ {{< /highlight >}}
+5. Define a static `from_runner_api_parameter` method that returns an
instantiation of the cross-language Python transform.
+
+ {{< highlight >}}
+@staticmethod
+def from_runner_api_parameter(
+ unused_ptransform, unused_parameter, unused_context):
+ return CombinePerKeyTransform()
+ {{< /highlight >}}
+
+**Using the expansion service**
Review comment:
Done.
----------------------------------------------------------------
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
For queries about this service, please contact Infrastructure at:
us...@infra.apache.org
