logo

Transport UDFs

Transport is a framework for writing performant user-defined functions (UDFs) that are portable across a variety of engines including Apache Spark, Apache Hive, and Presto. Transport UDFs are also capable of directly processing data stored in serialization formats such as Apache Avro. With Transport, developers only need to implement their UDF logic once using the Transport API. Transport then takes care of translating the UDF to native UDF version targeted at various engines or formats. Currently, Transport is capable of generating engine-artifacts for Spark, Hive, and Presto, and format-artifacts for Avro. Further details on Transport can be found in this LinkedIn Engineering blog post.

Documentation

Example

This example shows how a portable UDF is written using the Transport APIs.

public class MapFromTwoArraysFunction extends StdUDF2<StdArray, StdArray, StdMap> implements TopLevelStdUDF {

  private StdType _mapType;

  @Override
  public List<String> getInputParameterSignatures() {
    return ImmutableList.of(
        "array(K)",
        "array(V)"
    );
  }

  @Override
  public String getOutputParameterSignature() {
    return "map(K,V)";
  }

  @Override
  public void init(StdFactory stdFactory) {
    super.init(stdFactory);
    _mapType = getStdFactory().createStdType(getOutputParameterSignature());
  }

  @Override
  public StdMap eval(StdArray a1, StdArray a2) {
    if (a1.size() != a2.size()) {
      return null;
    }
    StdMap map = getStdFactory().createMap(_mapType);
    for (int i = 0; i < a1.size(); i++) {
      map.put(a1.get(i), a2.get(i));
    }
    return map;
  }

  @Override
  public String getFunctionName() {
    return "map_from_two_arrays";
  }

  @Override
  public String getFunctionDescription() {
    return "A function to create a map out of two arrays";
  }
}

In the example above, StdMap and StdArray are interfaces that provide high-level map and array operations to their objects. Depending on the engine where this UDF is executed, those interfaces are implemented differently to deal with native data types used by that engine. getStdFactory() is a method used to create objects that conform to a given data type (such as a map whose keys are of the type of elements in the first array and values are of the type of elements in the second array). StdUDF2 is an abstract class to express a UDF that takes two parameters. It is parametrized by the UDF input types and the UDF output type. Please consult the Transport UDFs API for more details and examples.

How to Build

Clone the repository:

git clone https://github.com/linkedin/transport.git

Change directory to transport:

cd transport

Build:

./gradlew build

Please note that this project requires Java 1.8.0_151 or higher. Either set JAVA_HOME to the home of an appropriate version and then use ./gradlew build as described above, or set the org.gradle.java.home gradle property to the Java home of an appropriate version as below:

./gradlew -Dorg.gradle.java.home=/path/to/java/home build

How to Use

The project under the directory transportable-udfs-examples is a standalone Gradle project that shows how to setup a project that uses the Transport UDFs framework to write Transportable UDFs. You can model your project after that standalone project. It implements a number of example UDFs to showcase different features and aspects of the API. Basically, you need to check out three components:

Change directory to transportable-udfs-examples:

cd transportable-udfs-examples

Build transportable-udfs-examples:

gradle build

You will notice that the build process generates some code. This is the platform-specific versions of the UDFs. Once the build succeeds, check out the output artifacts:

ls transportable-udfs-example-udfs/build/libs/

The results should be like:

transportable-udfs-example-udfs-hive.jar
transportable-udfs-example-udfs-presto.jar
transportable-udfs-example-udfs-spark.jar
transportable-udfs-example-udfs.jar

That is it! While only one version of the UDFs is implemented, multiple jars are produced upon building the project. Each of those jars uses native platform APIs and data models to implement the UDFs. So from an execution engine's perspective, there is no data transformation needed for interoperability or portability. Only suitable classes are used for each engine.

To call those jars from your SQL engine (i.e., Hive, Spark, or Presto), the standard process for deploying UDF jars is followed for each engine. For example, in Hive, you add the jar to the classpath using the ADD JAR statement, and register the UDF using CREATE FUNCTION statement. In Presto, the jar is deployed to the plugin directory. However, a small patch is required for the Presto engine to recognize the jar as a plugin, since the generated Presto UDFs implement the SqlScalarFunction API, which is currently not part of Presto's SPI architecture. You can find the patch here and apply it before deploying your UDFs jar to the Presto engine.

Contributing

The project is under active development and we welcome contributions of different forms:

Please take a look at the Contribution Agreement.

Questions?

Please send any questions or discussion topics to [email protected]

License

BSD 2-CLAUSE LICENSE

Copyright 2018 LinkedIn Corporation.
All Rights Reserved.
Redistribution and use in source and binary forms, with or without
modification, are permitted provided that the following conditions are
met:

1. Redistributions of source code must retain the above copyright
   notice, this list of conditions and the following disclaimer.

2. Redistributions in binary form must reproduce the above copyright
   notice, this list of conditions and the following disclaimer in the
   documentation and/or other materials provided with the
   distribution.

THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
"AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
HOLDER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
(INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.