Deploy ScalarDB Analytics in Public Cloud Environments

This guide explains how to deploy ScalarDB Analytics in a public cloud environment. ScalarDB Analytics consists of two main components: a ScalarDB Analytics server and Apache Spark. In this guide, you can choose either Amazon EMR or Databricks for the Spark environment. For details about ScalarDB Analytics, refer to ScalarDB Analytics Design.

Deploy ScalarDB Analytics catalog server

ScalarDB Analytics requires a catalog server to manage metadata and data source connections. The catalog server should be deployed by using Helm charts on a Kubernetes cluster.

For detailed deployment instructions, see [TBD - Helm chart deployment guide].

After deploying the catalog server, note the following information for Spark configuration:

• Catalog server host address
• Catalog port (default: 11051)
• Metering port (default: 11052)

Deploy Spark with ScalarDB Analytics

After deploying the catalog server, you can configure and deploy Spark with ScalarDB Analytics by using managed Spark services.

Supported managed Spark services and their application types

ScalarDB Analytics supports the following managed Spark services and application types.

Public Cloud Service	Spark Driver	Spark Connect	JDBC
Amazon EMR (EMR on EC2)	✅	✅	❌
Databricks	✅	❌	✅

Configure and deploy

Select your public cloud environment, and follow the instructions to set up and deploy Spark with ScalarDB Analytics.

Amazon EMR

Use Amazon EMR

You can use Amazon EMR (EMR on EC2) to run analytical queries through ScalarDB Analytics. For the basics to launch an EMR cluster, please refer to the AWS EMR on EC2 documentation.

ScalarDB Analytics configuration

To enable ScalarDB Analytics, you need to add the following configuration to the Software setting when you launch an EMR cluster. Be sure to replace the content in the angle brackets:

[
  {
    "Classification": "spark-defaults",
    "Properties": {
      "spark.jars.packages": "com.scalar-labs:scalardb-analytics-spark-all-<SPARK_VERSION>_<SCALA_VERSION>:<SCALARDB_ANALYTICS_VERSION>",
      "spark.extraListeners": "com.scalar.db.analytics.spark.metering.ScalarDbAnalyticsListener",
      "spark.sql.catalog.<CATALOG_NAME>": "com.scalar.db.analytics.spark.catalog.ScalarDBAnalyticsCatalog",
      "spark.sql.catalog.<CATALOG_NAME>.server.host": "<CATALOG_SERVER_HOST>",
      "spark.sql.catalog.<CATALOG_NAME>.server.catalog.port": "11051",
      "spark.sql.catalog.<CATALOG_NAME>.server.metering.port": "11052"
    }
  }
]

The following describes what you should change the content in the angle brackets to:

• <SPARK_VERSION>: The version of Spark (e.g., 3.5 or 3.4).
• <SCALA_VERSION>: The version of Scala used to build Spark (for example, 2.13 or 2.12).
• <SCALARDB_ANALYTICS_VERSION>: The version of ScalarDB Analytics (for example, 3.16.0).
• <CATALOG_NAME>: The name of the catalog. This must match a catalog created on the ScalarDB Analytics server.
• <CATALOG_SERVER_HOST>: The host address of your ScalarDB Analytics server.

For more details, refer to Set up ScalarDB Analytics in the Spark configuration.

Run analytical queries via the Spark driver

After the EMR Spark cluster has launched, you can use ssh to connect to the primary node of the EMR cluster and run your Spark application. For details on how to create a Spark driver application, refer to Spark driver application.

Run analytical queries via Spark Connect

You can use Spark Connect to run your Spark application remotely by using the EMR cluster that you launched.

You first need to configure the Software setting in the same way as the Spark driver application. You also need to set the following configuration to enable Spark Connect.

Allow inbound traffic for a Spark Connect server

1. Create a security group to allow inbound traffic for a Spark Connect server. (Port 15001 is the default).
2. Allow the role of "Amazon EMR service role" to attach the security group to the primary node of the EMR cluster.
3. Add the security group to the primary node of the EMR cluster as "Additional security groups" when you launch the EMR cluster.

Launch the Spark Connect server via a bootstrap action

1. Create a script file to launch the Spark Connect server as follows:

#!/usr/bin/env bash

set -eu -o pipefail

cd /var/lib/spark

sudo -u spark /usr/lib/spark/sbin/start-connect-server.sh --packages org.apache.spark:spark-connect_<SCALA_VERSION>:<SPARK_FULL_VERSION>,com.scalar-labs:scalardb-analytics-spark-all-<SPARK_VERSION>_<SCALA_VERSION>:<SCALARDB_ANALYTICS_VERSION>

The following describes what you should change the content in the angle brackets to:

• <SCALA_VERSION>: The major and minor version of Scala that matches your Spark installation (such as 2.12 or 2.13)
• <SPARK_FULL_VERSION>: The full version of Spark you are using (such as 3.5.3)
• <SPARK_VERSION>: The major and minor version of Spark you are using (such as 3.5)
• <SCALARDB_ANALYTICS_VERSION>: The version of ScalarDB Analytics

2. Upload the script file to S3.
3. Allow the role of "EC2 instance profile for Amazon EMR" to access the uploaded script file in S3.
4. Add the uploaded script file to "Bootstrap actions" when you launch the EMR cluster.

Run analytical queries

You can run your Spark application via Spark Connect from anywhere by using the remote URL of the Spark Connect server, which is sc://<PRIMARY_NODE_PUBLIC_HOSTNAME>:15001.

For details on how to create a Spark application by using Spark Connect, refer to Spark Connect application.

Databricks

Use Databricks

You can use Databricks to run analytical queries through ScalarDB Analytics.

note

Note that Databricks provides a modified version of Apache Spark, which works differently from the original Apache Spark.

Launch Databricks cluster

ScalarDB Analytics works with all-purpose and jobs-compute clusters on Databricks. When you launch the cluster, you need to configure the cluster to enable ScalarDB Analytics as follows:

1. Select "No isolation shared" for the cluster mode. (This is required. ScalarDB Analytics works only with this cluster mode.)
2. Select an appropriate Databricks runtime version that supports Spark 3.4 or later.
3. Configure "Advanced Options" > "Spark config" as follows:

spark.extraListeners com.scalar.db.analytics.spark.metering.ScalarDbAnalyticsListener
spark.sql.catalog.<CATALOG_NAME> com.scalar.db.analytics.spark.catalog.ScalarDBAnalyticsCatalog
spark.sql.catalog.<CATALOG_NAME>.server.host <CATALOG_SERVER_HOST>
spark.sql.catalog.<CATALOG_NAME>.server.catalog.port 11051
spark.sql.catalog.<CATALOG_NAME>.server.metering.port 11052

Replace the placeholders:

• <CATALOG_NAME>: The name of the catalog. This must match a catalog created on the ScalarDB Analytics server.
• <CATALOG_SERVER_HOST>: The host address of your ScalarDB Analytics catalog server.

4. Add the library of ScalarDB Analytics to the launched cluster as a Maven dependency. For details on how to add the library, refer to the Databricks cluster libraries documentation.

Run analytical queries via the Spark driver

You can run your Spark application on the properly configured Databricks cluster with Databricks Notebook or Databricks Jobs to access the tables in ScalarDB Analytics. To run the Spark application, you can migrate your Pyspark, Scala, or Spark SQL application to Databricks Notebook, or use Databricks Jobs to run your Spark application. ScalarDB Analytics works with task types for Notebook, Python, JAR, and SQL.

For more details on how to use Databricks Jobs, refer to the Databricks Jobs documentation

Run analytical queries via the JDBC driver

Databricks supports JDBC to run SQL jobs on the cluster. You can use this feature to run your Spark application in SQL with ScalarDB Analytics by configuring extra settings as follows:

1. Download the ScalarDB Analytics library JAR file from the Maven repository.
2. Upload the JAR file to the Databricks workspace.
3. Add the JAR file to the cluster as a library, instead of the Maven dependency.
4. Create an init script as follows, replacing <PATH_TO_YOUR_JAR_FILE_IN_WORKSPACE> with the path to your JAR file in the Databricks workspace:

#!/bin/bash

# Target directories
TARGET_DIRECTORIES=("/databricks/jars" "/databricks/hive_metastore_jars")
JAR_PATH="<PATH_TO_YOUR_JAR_FILE_IN_WORKSPACE>"

# Copy the JAR file to the target directories
for TARGET_DIR in "${TARGET_DIRECTORIES[@]}"; do
  mkdir -p "$TARGET_DIR"
  cp "$JAR_PATH" "$TARGET_DIR/"
done

5. Upload the init script to the Databricks workspace.
6. Add the init script to the cluster to "Advanced Options" > "Init scripts" when you launch the cluster.

After the cluster is launched, you can get the JDBC URL of the cluster in the "Advanced Options" > "JDBC/ODBC" tab on the cluster details page.

To connect to the Databricks cluster by using JDBC, you need to add the Databricks JDBC driver to your application dependencies. For example, if you are using Gradle, you can add the following dependency to your build.gradle file:

implementation("com.databricks:databricks-jdbc:0.9.6-oss")

Then, you can connect to the Databricks cluster by using JDBC with the JDBC URL (<YOUR_CLUSTERS_JDBC_URL>), as is common with JDBC applications.

Class.forName("com.databricks.client.jdbc.Driver");
String url = "<YOUR_CLUSTERS_JDBC_URL>";
Connection conn = DriverManager.getConnection(url)

For more details on how to use JDBC with Databricks, refer to the Databricks JDBC Driver documentation.