Apache Beam Portability Demo

For all runners, batch demo shows:

• Graph structure
• Proof of parallelism
• Input data size
• Metric: parseErrors

Streaming demo adds:

• Watermarks

Apache Kafka cluster in Google Cloud Dataproc

Set firewall rules for your GCP project to open tcp:2181, tcp:2888, tcp:3888 for Zookeeper and tcp:9092 for Kafka.

Upload the Dataproc config script that will install Kafka on Dataproc:

gsutil cp dataproc-config/dataproc-kafka-init.sh gs://apache-beam-demo/config/

Start the Kafka cluster:

gcloud dataproc clusters create kafka \
    --zone=us-central1-f \
    --initialization-actions gs://apache-beam-demo/config/dataproc-kafka-init.sh \
    --initialization-action-timeout 5m \
    --num-workers=3 \
    --scopes=https://www.googleapis.com/auth/cloud-platform \
    --worker-boot-disk-size=100gb \
    --master-boot-disk-size=500gb \
    --master-machine-type n1-standard-4 \
    --worker-machine-type n1-standard-4       

Create a topic to use for the game:

bin/kafka-topics.sh --create --zookeeper <external ip for kafka-m>:2181 --replication-factor 1 --partitions 3 --topic game

Note: The project pom.xml currently hard codes the external IP address for kafka-m, so you'll need to edit it by hand.

Injector

On a new "injector VM", install Maven (minimum 3.3.1), git, and OpenJDK 7.

git clone https://github.com/davorbonaci/beam-portability-demo.git
cd beam-portability-demo

screen
mvn clean compile exec:java@injector

Press Ctrl-A, Ctrl-D (later screen -r to resume).

Google Cloud Dataflow

HourlyTeamScore:

mvn clean compile exec:java@HourlyTeamScore-Dataflow -Pdataflow-runner

Leaderboard (requires the injector):

mvn clean compile exec:java@LeaderBoard-Dataflow -Pdataflow-runner

Apache Flink cluster in Google Cloud Dataproc

gsutil cp dataproc-config/dataproc-flink-init.sh gs://apache-beam-demo/config/

gcloud dataproc clusters create gaming-flink \
    --zone=us-central1-f \
    --initialization-actions gs://apache-beam-demo/config/dataproc-flink-init.sh \
    --initialization-action-timeout 5m \
    --num-workers=20 \
    --scopes=https://www.googleapis.com/auth/cloud-platform \
    --worker-boot-disk-size=100gb \
    --master-boot-disk-size=100gb

To view the Apache Flink UI, in another terminal:

gcloud compute ssh --zone=us-central1-f --ssh-flag="-D 1081" --ssh-flag="-N" --ssh-flag="-n" gaming-flink-m

Launch magic Google Chrome window and, if applicable, set BeyondCorp to System/Alternative:

/Applications/Google\ Chrome.app/Contents/MacOS/Google\ Chrome \
  --proxy-server="socks5://localhost:1081" \
  --host-resolver-rules="MAP * 0.0.0.0 , EXCLUDE localhost" \
  --user-data-dir=/tmp/

Open the UI:

http://gaming-flink-m:8088/

In the Flink UI, capture values of jobmanager.rpc.address (i.e. gaming-flink-w-5) and jobmanager.rpc.port (i.e. 58268) in the Job Manager configuration.

In separate terminals, use those two values to set up two SSH tunnels to the machine in the cluster running the Flink Job Manager:

gcloud compute ssh gaming-flink-w-10 --zone us-central1-f --ssh-flag="-D 1082"

gcloud compute ssh gaming-flink-w-10 --zone us-central1-f --ssh-flag="-L 40007:localhost:40007"

Submit HourlyTeamScore to the cluster:

mvn clean package exec:java -Pflink-runner \
    -DsocksProxyHost=localhost \
    -DsocksProxyPort=1082 \
    -Dexec.mainClass="demo.HourlyTeamScore" \
    -Dexec.args="--runner=flink \
                 --input=gs://apache-beam-demo/data/gaming* \
                 --outputPrefix=gs://apache-beam-demo-fjp/flink/hourly/scores \
                 --filesToStage=target/portability-demo-bundled-flink.jar \
                 --flinkMaster=gaming-flink-w-10.c.apache-beam-demo.internal:40007"

Submit LeaderBoard to the cluster:

mvn clean package exec:java -Pflink-runner \
    -DsocksProxyHost=localhost \
    -DsocksProxyPort=1082 \
    -Dexec.mainClass="demo.LeaderBoard" \
    -Dexec.args="--runner=flink \
                 --kafkaBootstrapServer=35.184.132.47:9092 \
                 --topic=game \
                 --outputPrefix=gs://apache-beam-demo-fjp/flink/leader/scores \
                 --filesToStage=target/portability-demo-bundled-flink.jar \
                 --flinkMaster=gaming-flink-w-10.c.apache-beam-demo.internal:40007"                     

If you receive an error saying Cannot resolve the JobManager hostname, you may need to modify your /etc/hosts file to include an entry like this:

127.0.0.1 gaming-flink-w-5.c.apache-beam-demo.internal

Apache Spark cluster in Google Cloud Dataproc

gcloud dataproc clusters create gaming-spark \
    --image-version=1.0 \
    --zone=us-central1-f \
    --num-workers=25 \
    --worker-machine-type=n1-standard-8 \
    --master-machine-type=n1-standard-8 \
    --worker-boot-disk-size=100gb \
    --master-boot-disk-size=100gb

To view the Apache Spark UI, in another terminal:

gcloud compute ssh --zone=us-central1-f --ssh-flag="-D 1080" --ssh-flag="-N" --ssh-flag="-n" gaming-spark-m

Launch magic Google Chrome window and, if applicable, set BeyondCorp to System/Alternative:

/Applications/Google\ Chrome.app/Contents/MacOS/Google\ Chrome \
    --proxy-server="socks5://localhost:1080" \
    --host-resolver-rules="MAP * 0.0.0.0 , EXCLUDE localhost" \
    --user-data-dir=/tmp/

Open the UI:

http://gaming-spark-m:18080/

Submit the job to the cluster:

mvn clean package -Pspark-runner

gcloud dataproc jobs submit spark \
    --cluster gaming-spark \
    --properties spark.default.parallelism=200 \
    --class demo.HourlyTeamScore \
    --jars ./target/portability-demo-bundled-spark.jar \
    -- \
    --runner=spark \
    --outputPrefix=gs://apache-beam-demo-fjp/spark/hourly/scores \
    --input=gs://apache-beam-demo/data/gaming*