SystemML Release Process

Release Candidate Build and Deployment

To be written. (Describe how the release candidate is built, including checksums. Describe how the release candidate is deployed to servers for review.)

Release Candidate Checklist

All Artifacts and Checksums Present

Up to Checklist

Verify that each expected artifact is present at and that each artifact has accompanying checksums (such as .asc and .md5).

Release Candidate Build

Up to Checklist

The release candidate should build on Windows, OS X, and Linux. To do this cleanly, the following procedure can be performed.

Clone the Apache SystemML GitHub repository to an empty location. Next, check out the release tag. Following this, build the distributions using Maven. This should be performed with an empty local Maven repository.

Here is an example:

$ git clone
$ cd incubator-systemml
$ git tag -l
$ git checkout tags/0.11.0-incubating-rc1 -b 0.11.0-incubating-rc1
$ mvn -Dmaven.repo.local=$HOME/.m2/temp-repo clean package -P distribution

Test Suite Passes

Up to Checklist

The entire test suite should pass on Windows, OS X, and Linux. The test suite can be run using:

$ mvn clean verify

All Binaries Execute

Up to Checklist

Validate that all of the binary artifacts can execute, including those artifacts packaged in other artifacts (in the tgz and zip artifacts).

The build artifacts should be downloaded from and these artifacts should be tested, as in this OS X example.

# download artifacts
wget -r -nH -nd -np -R 'index.html*'

# verify standalone tgz works
tar -xvzf systemml-0.13.0-incubating-bin.tgz
cd systemml-0.13.0-incubating-bin
echo "print('hello world');" > hello.dml
./ hello.dml
cd ..

# verify standalon zip works
rm -rf systemml-0.13.0-incubating-bin
cd systemml-0.13.0-incubating-bin
echo "print('hello world');" > hello.dml
./ hello.dml
cd ..

# verify src works
tar -xvzf systemml-0.13.0-incubating-src.tgz
cd systemml-0.13.0-incubating-src
mvn clean package -P distribution
cd target/
java -cp "./lib/*:systemml-0.13.0-incubating.jar" org.apache.sysml.api.DMLScript -s "print('hello world');"
java -cp "./lib/*:SystemML.jar" org.apache.sysml.api.DMLScript -s "print('hello world');"
cd ../..

# verify spark batch mode
export SPARK_HOME=~/spark-2.1.0-bin-hadoop2.7
cd systemml-0.13.0-incubating-bin/target/lib
$SPARK_HOME/bin/spark-submit systemml-0.13.0-incubating.jar -s "print('hello world');" -exec hybrid_spark

# verify hadoop batch mode
hadoop jar systemml-0.13.0-incubating.jar -s "print('hello world');"

# verify python artifact
# install numpy, pandas, scipy & set SPARK_HOME
pip install numpy
pip install pandas
pip install scipy
export SPARK_HOME=~/spark-2.1.0-bin-hadoop2.7
# get into the pyspark prompt
cd systemml-0.13.0
$SPARK_HOME/bin/pyspark --driver-class-path systemml-java/systemml-0.13.0-incubating.jar
# Use this program at the prompt:
import systemml as sml
import numpy as np
m1 = sml.matrix(np.ones((3,3)) + 2)
m2 = sml.matrix(np.ones((3,3)) + 3)
m2 = m1 * (m2 + m1)
m4 = 1.0 - m2

# This should be printed
# array([[-60.],
#       [-60.],
#       [-60.]])

Python Tests

For Spark 1.*, the Python tests at (src/main/python/tests) can be executed in the following manner:

PYSPARK_PYTHON=python3 pyspark --driver-class-path SystemML.jar
PYSPARK_PYTHON=python3 pyspark --driver-class-path SystemML.jar
PYSPARK_PYTHON=python3 pyspark --driver-class-path SystemML.jar
PYSPARK_PYTHON=python3 pyspark --driver-class-path SystemML.jar
PYSPARK_PYTHON=python3 pyspark --driver-class-path SystemML.jar

For Spark 2.*, pyspark can’t be used to run the Python tests, so they can be executed using spark-submit:

spark-submit --driver-class-path SystemML.jar
spark-submit --driver-class-path SystemML.jar
spark-submit --driver-class-path SystemML.jar
spark-submit --driver-class-path SystemML.jar
spark-submit --driver-class-path SystemML.jar

Check LICENSE and NOTICE Files

Up to Checklist

Each artifact must contain LICENSE and NOTICE files. These files must reflect the contents of the artifacts. If the project dependencies (ie, libraries) have changed since the last release, the LICENSE and NOTICE files must be updated to reflect these changes.

Each artifact should contain a DISCLAIMER file.

For more information, see:


Src Artifact Builds and Tests Pass

Up to Checklist

The project should be built using the src (tgz and zip) artifacts. In addition, the test suite should be run using an src artifact and the tests should pass.

tar -xvzf systemml-0.13.0-incubating-src.tgz
cd systemml-0.13.0-incubating-src
mvn clean package -P distribution
mvn verify

Single-Node Standalone

Up to Checklist

The standalone tgz and zip artifacts contain and runStandaloneSystemML.bat files. Verify that one or more algorithms can be run on a single node using these standalone distributions.

Here is an example based on the Standalone Guide demonstrating the execution of an algorithm (on OS X).

tar -xvzf systemml-0.13.0-incubating-bin.tgz
cd systemml-0.13.0-incubating-bin
wget -P data/
echo '{"rows": 306, "cols": 4, "format": "csv"}' > data/
echo '1,1,1,2' > data/types.csv
echo '{"rows": 1, "cols": 4, "format": "csv"}' > data/types.csv.mtd
./ scripts/algorithms/Univar-Stats.dml -nvargs X=data/ TYPES=data/types.csv STATS=data/univarOut.mtx CONSOLE_OUTPUT=TRUE
cd ..

Single-Node Spark

Up to Checklist

Verify that SystemML runs algorithms on Spark locally.

Here is an example of running the Univar-Stats.dml algorithm on random generated data.

cd systemml-0.13.0-incubating-bin/lib
export SPARK_HOME=~/spark-2.1.0-bin-hadoop2.7
$SPARK_HOME/bin/spark-submit systemml-0.13.0-incubating.jar -f ../scripts/datagen/genRandData4Univariate.dml -exec hybrid_spark -args 1000000 100 10 1 2 3 4 uni.mtx
echo '1' > uni-types.csv
echo '{"rows": 1, "cols": 1, "format": "csv"}' > uni-types.csv.mtd
$SPARK_HOME/bin/spark-submit systemml-0.13.0-incubating.jar -f ../scripts/algorithms/Univar-Stats.dml -exec hybrid_spark -nvargs X=uni.mtx TYPES=uni-types.csv STATS=uni-stats.txt CONSOLE_OUTPUT=TRUE
cd ..

Single-Node Hadoop

Up to Checklist

Verify that SystemML runs algorithms on Hadoop locally.

Based on the “Single-Node Spark” setup above, the Univar-Stats.dml algorithm could be run as follows:

cd systemml-0.13.0-incubating-bin/lib
hadoop jar systemml-0.13.0-incubating.jar -f ../scripts/algorithms/Univar-Stats.dml -nvargs X=uni.mtx TYPES=uni-types.csv STATS=uni-stats.txt CONSOLE_OUTPUT=TRUE


Up to Checklist

Verify that SystemML can be executed from Jupyter and Zeppelin notebooks. For examples, see the Spark MLContext Programming Guide.

Performance Suite

Up to Checklist

Verify that the performance suite located at scripts/perftest/ executes on Spark and Hadoop. Testing should include 80MB, 800MB, 8GB, and 80GB data sizes.


Following a successful release candidate vote by SystemML PMC members on the SystemML mailing list, the release candidate is voted on by Incubator PMC members on the general incubator mailing list. If this vote succeeds, the release candidate has been approved.


Release Deployment

To be written. (What steps need to be done? How is the release deployed to Apache dist and the central maven repo? Where do the release notes for the release go?)

Documentation Deployment

This section describes how to deploy versioned project documentation to the main website. Note that versioned project documentation is committed directly to the svn project’s docs folder. The versioned project documentation is not committed to the website’s git project.

Checkout branch in main project (incubator-systemml).

$ git checkout branch-0.13.0

In incubator-systemml/docs/_config.yml, set:

Generate docs/_site by running bundle exec jekyll serve in incubator-systemml/docs.

$ bundle exec jekyll serve

Verify documentation site looks correct.

In website svn project, create incubator-systemml-website-site/docs/0.13.0 folder.

Copy contents of incubator-systemml/docs/_site to incubator-systemml-website-site/docs/0.13.0.

Delete any unnecessary files (Gemfile, Gemfile.lock).

Create incubator-systemml-website-site/docs/0.13.0/api/java folder for javadocs.

Update incubator-systemml/pom.xml project version to what should be displayed in javadocs (such as 0.13.0).

Build project (which generates javadocs).

$ mvn clean package -P distribution

Copy contents of incubator-systemml/target/apidocs to incubator-systemml-website-site/docs/0.13.0/api/java.

Open up file:///.../incubator-systemml-website-site/docs/0.13.0/index.html and verify API DocsJavadoc link works and that the correct Javadoc version is displayed. Verify feedback links under Issues menu are not present.

Clean up any unnecessary files (such as deleting .DS_Store files on OS X).

$ find . -name '.DS_Store' -type f -delete

Commit the versioned project documentation to svn:

$ svn status
$ svn add docs/0.13.0
$ svn commit -m "Add 0.13.0 docs to website"

Update incubator-systemml-website/_src/documentation.html to include 0.13.0 link.

Start main website site by running gulp in incubator-systemml-website:

$ gulp

Commit and push the update to git project.

$ git add -u
$ git commit -m "Add 0.13.0 link to documentation page"
$ git push
$ git push apache master

Copy contents of incubator-systemml-website/_site (generated by gulp) to incubator-systemml-website-site. After doing so, we should see that incubator-systemml-website-site/documentation.html has been updated.

$ svn status
$ svn diff

Commit the update to documentation.html to publish the website update.

$ svn commit -m "Add 0.13.0 link to documentation page"

The versioned project documentation is now deployed to the main website, and the Documentation Page contains a link to the versioned documentation.