Building Flowman from Sources

Since Flowman depends on libraries like Spark and Hadoop, which are commonly provided by a platform environment like Cloudera or EMR, you currently need to build Flowman yourself to match the correct versions.

Download Prebuilt Distribution

The simplest way to get started with Flowman is to download a prebuilt distribution, which is provided at GitHub. This probably is the simplest way to grab a working Flowman package. Note that for each release, there are different packages being provided, for different Spark and Hadoop versions. The naming is very simple:

flowman-dist-<version>-oss-spark<spark-version>-hadoop<hadoop-version>-bin.tar.gz

You simply have to use the package which fits to the Spark and Hadoop versions of your environment. For example the package of Flowman 0.24.1 and for Spark 3.1 and Hadoop 3.2 would be

flowman-dist-0.24.1-oss-spark3.1-hadoop3.2-bin.tar.gz

and the full URL then would be

https://github.com/dimajix/flowman/releases/download/0.24.1/flowman-dist-0.24.1-oss-spark3.1-hadoop3.2-bin.tar.gz

Build with Maven

When you decide against downloading a prebuilt Flowman distribution, you can simply build it yourself with Maven. As a prerequisite, you need * Java (1.8 for Spark <= 2.4 and 11 for Spark >= 3.0) * Apache Maven * On Windows: Hadoop libraries

Building Flowman with the default settings (i.e. Hadoop and Spark version) is as easy as

mvn clean install

Skip Tests

In case you don’t want to run tests, you can simply append -DskipTests

mvn clean install -DskipTests

Skip Docker Image

In case you don’t want to build the Docker image (for example when the build itself is done within a Docker container), you can simply append -Ddockerfile.skip

mvn clean install -Ddockerfile.skip

Main Artifacts

The main artifacts will be a Docker image ‘dimajix/flowman’ and additionally a tar.gz file containing a runnable version of Flowman for direct installation in cases where Docker is not available or when you want to run Flowman in a complex environment with Kerberos. You can find the tar.gz file in the directory flowman-dist/target

Custom Builds

Flowman supports various versions of Spark and Hadoop to match your requirements and your environment. By providing appropriate build profiles, you can easily create a custom build.

Build on Windows

Although you can normally build Flowman on Windows, you will need the Hadoop WinUtils installed. You can download the binaries from https://github.com/steveloughran/winutils and install an appropriate version somewhere onto your machine. Do not forget to set the HADOOP_HOME environment variable to the installation directory of these utils!

You should also configure git such that all files are checked out using “LF” endings instead of “CRLF”, otherwise some unittests may fail and Docker images might not be useable. This can be done by setting the git configuration value “core.autocrlf” to “input”

git config --global core.autocrlf input

You might also want to skip unittests (the HBase plugin is currently failing under windows)

mvn clean install -DskipTests

Build for Custom Spark / Hadoop Version

Per default, Flowman will be built for fairly recent versions of Spark (3.0.2 as of this writing) and Hadoop (3.2.0). But of course, you can also build for a different version by either using a profile

mvn install -Pspark2.4 -Phadoop2.7 -DskipTests

This will always select the latest bugfix version within the minor version. You can also specify versions explicitly as follows:

mvn install -Dspark.version=2.4.1 -Dhadoop.version=2.7.3

Note that using profiles is the preferred way, as this guarantees that also dependencies are selected using the correct version. The following profiles are available:

spark-2.4
spark-3.0
spark-3.1
spark-3.2
spark-3.3
hadoop-2.6
hadoop-2.7
hadoop-2.8
hadoop-2.9
hadoop-3.1
hadoop-3.2
hadoop-3.3
EMR-6.10
synapse-3.3
CDH-6.3
CDP-7.1
CDP-7.1-spark-3.2
CDP-7.1-spark-3.3

With these profiles it is easy to build Flowman to match your environment.

Building for specific Java Version

If nothing else is set on the command line, Flowman will now build for Java 11 (except when building the profile CDH-6.3, where Java 1.8 is used). If you are still stuck on Java 1.8, you can simply override the Java version by specifying the property java.version

mvn install -Djava.version=1.8

Building for Open Source Hadoop and Spark

Spark 2.4 and Hadoop 2.6:

mvn clean install -Pspark-2.4 -Phadoop-2.6 -DskipTests

Spark 2.4 and Hadoop 2.7:

mvn clean install -Pspark-2.4 -Phadoop-2.7 -DskipTests

Spark 2.4 and Hadoop 2.8:

mvn clean install -Pspark-2.4 -Phadoop-2.8 -DskipTests

Spark 2.4 and Hadoop 2.9:

mvn clean install -Pspark-2.4 -Phadoop-2.9 -DskipTests

Spark 3.0 and Hadoop 3.1

mvn clean install -Pspark-3.0 -Phadoop-3.1 -DskipTests

Spark 3.0 and Hadoop 3.2

mvn clean install -Pspark-3.0 -Phadoop-3.2 -DskipTests

Spark 3.1 and Hadoop 3.2

mvn clean install -Pspark-3.1 -Phadoop-3.2 -DskipTests

Spark 3.2 and Hadoop 3.3

mvn clean install -Pspark-3.2 -Phadoop-3.3 -Phadoop.version=3.3.1 -DskipTests

Spark 3.3 and Hadoop 3.3

mvn clean install -Pspark-3.3 -Phadoop-3.3 -Phadoop.version=3.3.2 -DskipTests

Building for Cloudera

The Maven project also contains preconfigured profiles for Cloudera CDH 6.3 and for CDP 7.1 and also supports optional Spark 3.2 and Spark 3.3 parcels for CDP 7.1:

mvn clean install -PCDH-6.3 -DskipTests

mvn clean install -PCDP-7.1 -DskipTests

mvn clean install -PCDP-7.1-spark-3.2 -DskipTests

mvn clean install -PCDP-7.1-spark-3.3 -DskipTests

Building for AWS EMR

The Maven project also contains preconfigured profiles for AWS EMR 6.10.

mvn clean install -PEMR-6.10 -DskipTests

Building for Azure Synapse

Flowman also provides a build profile for Azure Synapse 3.3:

mvn clean install -Psynapase-3.3 -DskipTests

Coverage Analysis

Flowman also now supports creating a coverage analysis via the scoverage Maven plugin. It is not part of the default build and has to be triggered explicitly:

mvn scoverage:report

Building Documentation

Flowman also contains Markdown documentation which is processed by Sphinx to generate the online HTML documentation.

cd docs
make html