Skip to content

Latest commit

 

History

History
169 lines (120 loc) · 7.97 KB

apache-spark-with-minio.md

File metadata and controls

169 lines (120 loc) · 7.97 KB
Raw

Apache Spark with MinIO Server Slack

Apache Spark is a fast and general engine for large-scale data processing. In this recipe we'll see how to launch jobs on Apache Spark-Shell that reads/writes data to a MinIO server.

1. Prerequisites

2. Installation

  • Extract the spark-2.4.4-bin-without-hadoop tar ball in the directory where you're planning to install Spark.
  • Extract the hadoop-2.8.2 tar ball in a separate directory. Copy the path to this directory.
  • Move all the dependency jar files (downloaded in previous step) in /path/to/.spark-2.4.4-bin-without-hadoop/jars directory.

3. Start Spark-Shell

Navigate to the directory where you extracted spark-2.4.4-bin-without-hadoop, and set the following environment variables:

export SPARK_HOME=/path/to/spark-2.4.4-bin-without-hadoop
export PATH=$PATH:$SPARK_HOME/bin
export HADOOP_HOME=/path/to/hadoop-2.8.2
export PATH=$PATH:$HADOOP_HOME/bin
export LD_LIBRARY_PATH=$HADOOP_HOME/lib/native
export SPARK_DIST_CLASSPATH=$(hadoop classpath)

Now, assuming your MinIO Server is running at http://localhost:9000, and Access Key, Secret Key are minio and minio123 respectively, start Spark-Shell

./bin/spark-shell \
--conf spark.hadoop.fs.s3a.endpoint=http://localhost:9000 \
--conf spark.hadoop.fs.s3a.access.key=minio \
--conf spark.hadoop.fs.s3a.secret.key=minio123 \
--conf spark.hadoop.fs.s3a.path.style.access=true \
--conf spark.hadoop.fs.s3a.impl=org.apache.hadoop.fs.s3a.S3AFileSystem

You should see the prompt

Welcome to
      ____              __
     / __/__  ___ _____/ /__
    _\ \/ _ \/ _ `/ __/  '_/
   /___/ .__/\_,_/_/ /_/\_\   version 2.4.4
      /_/

Using Scala version 2.11.12 (Java HotSpot(TM) 64-Bit Server VM, Java 1.8.0_221)
Type in expressions to have them evaluated.
Type :help for more information.

scala>

Remember to update the s3a related args as per your actual MinIO deployment values.

4. Test if Spark-Shell can communicate with MinIO server

Read Test

For this recipe, MinIO server is running on http://127.0.0.1:9000. To test if read works on Spark-Shell, create a bucket called spark-test on your MinIO server and upload a test file. Here is how to do this with mc

mc config host add myminio http://127.0.0.1:9000 minio minio123
mc mb myminio/spark-test
mc cp test.json myminio/spark-test/test.json

Now, switch to the Spark-Shell terminal and type

val b1 = sc.textFile("s3a://spark-test/test.json")
b1.collect().foreach(println)

You should be able to see the text file you uploaded to MinIO server.

If you configured Spark history server as described in step 4, you can see the event logs on the console here http://localhost:18080/.

Write Test

To test if Spark-Shell can write back to MinIO server, switch to Spark-Shell terminal and run

import spark.implicits._
val data = Array(1, 2, 3, 4, 5)
val distData = sc.parallelize(data)
distData.saveAsTextFile("s3a://spark-test/test-write")

You should see a prefix called test-write created inside the bucket spark-test. The data is written under the test-write prefix.

MinIO server configured with HTTPS

In case you are planning to use self-signed certificates for testing purposes, you'll need to add these certificates to local JRE cacerts. You can do this using

keytool -import -trustcacerts -keystore $JAVA_HOME/jre/lib/security/cacerts -storepass changeit -noprompt -alias mycert -file /home/username/.minio/certs/public.crt

Also, change the scheme to https in fs.s3a.endpoint in file $HADOOP_HOME/etc/hadoop/core-site.xml.

5. (Optional) Start Spark-History server

Spark History Server provides web UI for completed and running Spark applications. Once Spark jobs are configured to log events, the history server displays both completed and incomplete Spark jobs. If an application makes multiple attempts after failures, the failed attempts will be displayed, as well as any ongoing incomplete attempt or the final successful attempt.

MinIO can be used as the storage back-end for Spark history using the s3a file system. To enable this, setup the conf/spark-defaults.conf file so history server uses s3a to store the files. By default the conf directory has a spark-defaults.conf.template file, make a copy of the template file and rename it to spark-defaults.conf. Then add the below content to the file

spark.jars.packages                 net.java.dev.jets3t:jets3t:0.9.4,com.google.guava:guava:14.0.1,com.amazonaws:aws-java-sdk:1.11.712,org.apache.hadoop:hadoop-aws:2.8.2
spark.eventLog.enabled              true
spark.eventLog.dir                  s3a://spark/
spark.history.fs.logDirectory       s3a://spark/

Next step is to add jar files specified under the spark.jars.packages section to the jars directory. Once the files are added, create a new bucket called spark in the MinIO instance. This is because we specified the log directory as s3a://spark/. Finally start Spark history server using

./sbin/start-history-server.sh

If everything works fine you should be able to see the console on http://localhost:18080/.

6. (Optional) MinIO with DeltaLake

Delta Lake provides ACID transactions, scalable metadata handling, and unifies streaming and batch data processing. Delta Lake runs on top of existing data lake (like MinIO) and is fully compatible with Apache Spark APIs. Specifically, Delta Lake offers ACID transactions on Spark, Schema enforcement, Time travel and more. Read details here.

You can create, read, and write Delta tables on MinIO using the s3a connector. To setup Delta Lake with MinIO, you can use the below command

./bin/spark-shell --packages io.delta:delta-core_2.11:0.5.0 \
--conf spark.delta.logStore.class=org.apache.spark.sql.delta.storage.S3SingleDriverLogStore \
--conf spark.hadoop.fs.s3a.endpoint=http://localhost:9000 \
--conf spark.hadoop.fs.s3a.access.key=minio \
--conf spark.hadoop.fs.s3a.secret.key=minio123 \
--conf spark.hadoop.fs.s3a.path.style.access=true \
--conf spark.hadoop.fs.s3a.impl=org.apache.hadoop.fs.s3a.S3AFileSystem

Here, we're using the Delta Lake release 0.5.0 with Scala 2.11 release, while the s3a fields based on the MinIO cluster. After you run this and get the Spark console, you can run delta and Spark APIs. Refer Delta Lake storage configuration docs here.

Write Test

To test if you can use the delta API, use the below commands in spark console

spark.range(5).write.format("delta").save("s3a://spark-test/delta")

Read Test

To check if the reads are working fine, attempt reading the file we created in write test.

val df = spark.read.format("delta").load("s3a://spark-test/delta")
df.show()