The Apex Runner executes Apache Beam pipelines using Apache Apex as an underlying engine. The runner has broad support for the Beam model and supports streaming and batch pipelines.
Apache Apex is a stream processing platform and framework for low-latency, high-throughput and fault-tolerant analytics applications on Apache Hadoop. Apex has a unified streaming architecture and can be used for real-time and batch processing.
You may set up your own Hadoop cluster. Beam does not require anything extra to launch the pipelines on YARN. An optional Apex installation may be useful for monitoring and troubleshooting. The Apex CLI can be built or obtained as binary build. For more download options see distribution information on the Apache Apex website.
Put data for processing into HDFS:
hdfs dfs -mkdir -p /tmp/input/ hdfs dfs -put pom.xml /tmp/input/
The output directory should not exist on HDFS:
hdfs dfs -rm -r -f /tmp/output/
Run the wordcount example (example project needs to be modified to include HDFS file provider)
mvn compile exec:java -Dexec.mainClass=org.apache.beam.examples.WordCount -Dexec.args="--inputFile=/tmp/input/pom.xml --output=/tmp/output/ --runner=ApexRunner --embeddedExecution=false --configFile=beam-runners-apex.properties" -Papex-runner
The application will run asynchronously. Check status with
yarn application -list -appStates ALL
The configuration file is optional, it can be used to influence how Apex operators are deployed into YARN containers. The following example will reduce the number of required containers by collocating the operators into the same container and lower the heap memory per operator - suitable for execution in a single node Hadoop sandbox.
dt.application.*.operator.*.attr.MEMORY_MB=64 dt.stream.*.prop.locality=CONTAINER_LOCAL dt.application.*.operator.*.attr.TIMEOUT_WINDOW_COUNT=1200
Check the output of the pipeline in the HDFS location.
hdfs dfs -ls /tmp/output/
Depending on your installation, you may be able to monitor the progress of your job on the Hadoop cluster. Alternatively, you have following options: