Automated deployment for AWS via CloudFormation with Databricks

Applies to: Kyvos Enterprise  Kyvos Cloud (SaaS on AWS) Kyvos AWS Marketplace

Kyvos Azure Marketplace   Kyvos GCP Marketplace Kyvos Single Node Installation (Kyvos SNI)

Prerequisites

Before starting the deployment for AWS, you must have the following.

1. AWS CloudFormation template. Contact Kyvos support to get your custom template. Alternatively, download the default template Kyvos_AWS_Default_Template_Databricks2023.2.json file, or create a template as per your requirements.

2. The CloudFormation template can be deployed through the logged-in user or a role. The logged-in user must have the required policies given in the aws-console-user-iam-policy.json file.

3. EC2 key pair, consisting of a private key and a public key. You can create the key pair if needed.

4. Networking requirements:

   1. Use the Network CloudFormation template to create network resources (VPC, Subnet, and Security Group) automatically. 

      • If you want to deploy your network with NAT Gateway, use the NATGateway Template (vpc_nat.json file) . 
        OR

   2. If you want to use existing network resources, perform the following steps in your VPC. 

      1. You must create VPC Endpoints within your VPC, to connect with the AWS services. Else, you must have the internet and NAT Gateway in the subnet.

         List of VPC Endpoints for AWS services required by Kyvos:
         
         

5. Permission requirements:

   1. You can create IAM roles using the CloudFormation template (automated_deployment_iam_role.json file).
      OR

   2. Create IAM Role for:
      Refer to the section Creating IAM Roles for EC2 and Lambda to create new roles.

      1. EC2 that will be attached to all Kyvos instances. This role contains all the permissions required by Kyvos Services and Kyvos Manager.
         Details for permissions required for EC2.

      2. Lambda that will be attached to the Kyvos created Lambda functions. This role contains all the permissions required by lambda functions to run.

6. Port 443 of the Databricks cluster should be accessible by Kyvos.

7. Create Databricks-instanceprofile-role with the following permissions:

   { "Version": "2012-10-17", "Statement": [ { "Sid": "GrantCatalogAccessToGlue", "Effect": "Allow", "Action": [ "glue:BatchCreatePartition", "glue:BatchDeletePartition", "glue:BatchGetPartition", "glue:CreateDatabase", "glue:CreateTable", "glue:CreateUserDefinedFunction", "glue:DeleteDatabase", "glue:DeletePartition", "glue:DeleteTable", "glue:DeleteUserDefinedFunction", "glue:GetDatabase", "glue:GetDatabases", "glue:GetPartition", "glue:GetPartitions", "glue:GetTable", "glue:GetTables", "glue:GetUserDefinedFunction", "glue:GetUserDefinedFunctions", "glue:UpdateDatabase", "glue:UpdatePartition", "glue:UpdateTable", "glue:UpdateUserDefinedFunction" ], "Resource": [ "*" ] }, { "Sid": "VisualEditor0", "Effect": "Allow", "Action": [ "s3:GetBucketLocation", "s3:GetObject", "s3:ListBucket" ], "Resource": "*" } ] }

8. S3 Bucket permissions

   If you want to use an existing S3 bucket and IAM role, or if you want to read data from an S3 bucket other than where Kyvos is deployed, then the IAM role must have the following permissions on the S3 bucket.

   Here, replace:

   <Bucket Name> with the name of your bucket name.

   <Lambda Role> with the name of your Lambda Role.

   <EC2 Role> with the name of your EC2 Role.

   { "Version": "2008-10-17", "Statement": [ { "Sid": "Ec2LambdaRoleBucketPolicy", "Effect": "Allow", "Principal": { "AWS": [ "arn:aws:iam::456531263183:role/EC2-Role", "arn:aws:iam::<AWS Accout ID>:role/<Lambda Role>", "arn:aws:iam::456531263183:role/Databricks-instanceprofile-role" ] }, "Action": [ "s3:PutAnalyticsConfiguration", "s3:GetObjectVersionTagging", "s3:ReplicateObject", "s3:GetObjectAcl", "s3:GetBucketObjectLockConfiguration", "s3:DeleteBucketWebsite", "s3:PutLifecycleConfiguration", "s3:GetObjectVersionAcl", "s3:DeleteObject", "s3:GetBucketPolicyStatus", "s3:GetObjectRetention", "s3:GetBucketWebsite", "s3:PutReplicationConfiguration", "s3:PutObjectLegalHold", "s3:GetObjectLegalHold", "s3:GetBucketNotification", "s3:PutBucketCORS", "s3:GetReplicationConfiguration", "s3:ListMultipartUploadParts", "s3:PutObject", "s3:GetObject", "s3:PutBucketNotification", "s3:PutBucketLogging", "s3:GetAnalyticsConfiguration", "s3:PutBucketObjectLockConfiguration", "s3:GetObjectVersionForReplication", "s3:GetLifecycleConfiguration", "s3:GetInventoryConfiguration", "s3:GetBucketTagging", "s3:PutAccelerateConfiguration", "s3:DeleteObjectVersion", "s3:GetBucketLogging", "s3:ListBucketVersions", "s3:RestoreObject", "s3:ListBucket", "s3:GetAccelerateConfiguration", "s3:GetBucketPolicy", "s3:PutEncryptionConfiguration", "s3:GetEncryptionConfiguration", "s3:GetObjectVersionTorrent", "s3:AbortMultipartUpload", "s3:GetBucketRequestPayment", "s3:GetObjectTagging", "s3:GetMetricsConfiguration", "s3:DeleteBucket", "s3:PutBucketVersioning", "s3:GetBucketPublicAccessBlock", "s3:ListBucketMultipartUploads", "s3:PutMetricsConfiguration", "s3:GetBucketVersioning", "s3:GetBucketAcl", "s3:PutInventoryConfiguration", "s3:GetObjectTorrent", "s3:PutBucketWebsite", "s3:PutBucketRequestPayment", "s3:PutObjectRetention", "s3:GetBucketCORS", "s3:GetBucketLocation", "s3:ReplicateDelete", "s3:GetObjectVersion", "s3:PutBucketTagging" ], "Resource": [ "arn:aws:s3:::bucket-name/*", "arn:aws:s3:::bucket-name" ] } ] }

9. You must have the Access Key and Secret Key to access the Kyvos bundle. Contact Kyvos Support for details.

10. Valid Kyvos license file.

11. Databricks cluster for semantic model processing and processing aggregations, with the following parameters:  

    1. Databricks Runtime Version: Select Version 10.4 LTS (includes Apache Spark 3.2.1, Scala 2.12)

    2. Autopilot Options: Select the following:  

       1. Enable autoscaling: Select this to enable autoscaling.  

       2. Terminate after ___ minutes of inactivity: Set the value as 30

    3. Worker type: Recommended value r5.4xlarge  

       1. Min Workers: Recommended value 1  

       2. Max Workers: Recommended value 10  

    4. Driver Type: Recommended value r 5.xlarge  

    5. Advanced options

       1. By default, the Spot fall back to On-demand checkbox is selected. Kyvos recommends clearing this checkbox.

       2. In the Spark Configurations define the below property in case of Glue-based deployment.

          • spark.databricks.hive.metastore.glueCatalog.enabled  true  

       3. If cross-account glue is to be used, define the below property to access cross-account glue: spark.hadoop.hive.metastore.glue.catalogid <GLUE_CATALOG_ID>  
          Now, set the below parquet-specific configuration properties:  

          1. spark.hadoop.spark.sql.parquet.int96AsTimestamp true  

          2. spark.sql.parquet.binaryAsString false

          3. spark.sql.parquet.int96AsTimestamp true  

          4. spark.hadoop.spark.sql.parquet.binaryAsString false

          5. spark.sql.caseSensitive false

          6. spark.hadoop.spark.sql.caseSensitive false

          7. spark.databricks.preemption.enabled false

       4. You must change Spark configurations to use managed disk. Ensure that you must not change the configuration in the default root (/tmp) volume.

          1. In the Spark Configurations, add the spark.local.dir /local_disk0 property where the local_disk0 is the managed disk.

          2. Optionally, you can execute the df -h command from a notebook for verification.

          3. Add the SPARK_WORKER_DIR=/local_disk0 value in the Environment variables.

    6. Tags: UsedBy tag with the value set as Kyvos is  required to run the cluster.  

    7. Instance profile: Copy the Instance Profile ARN of the role which is being created in Point 7 above.  

       1. In Databricks console, go to Admin Console > Instance Profile and click Add Instance Profile. Paste the Instance Profile ARN in the text box.  

       2. Select the Skip Validation checkbox and then click Add.  

       3. In Cluster settings, go to Advance Options, and select the instance profile created above in the Instance Profile field.

12. Databricks information:  

    1. Databricks Cluster Id: To obtain this ID, click the Cluster Name on the Clusters page in Databricks.  
       The page URL shows <https://<databricks-instance>/#/settings/clusters/<cluster-id>. The cluster ID is the number after the /clusters/ component in the URL of this page.  

    2. Databricks Cluster Organization ID:  To obtain this ID, click the Cluster Name on the Clusters page in Databricks.  
       The number after o= in the workspace URL is the organization ID. For example, if the workspace URL is https://westus.azuredatabricks.net/?o=7692xxxxxxxx , then the organization ID is 7692xxxxxxxx.  

    3. Databricks Role ARN: Use the ARN of the Databricks-instanceprofile-role created in point 7 above . 
       The ARN looks like this:  arn:aws:iam ::45653****** *:role /AssumeRoleTest.
       This Databricks Role should have  " iam:PassRole " permission in the role you have created for  Databricks workspace.

13. If using an existing Secrets Manager, ensure that the KYVOS-DATABRICKS-SERVICE-TOKEN-DefaultHadoopCluster01 key is added to it.

Creating CloudFormation template

The Kyvos CloudFormation template can create the following resources:

• EC2 instances for Kyvos services - BI Server, Query Engines, Kyvos Manager, Kyvos Web Portal, and Postgres.

• S3 for storing Kyvos semantic model

• RDS for use as Kyvos repository if you don't want to use the default Postgres database provided in the Kyvos package.

• Lambda to use the scheduling (cluster ON) features.

• API Gateway to get the Rest URL on the Lambda function.

• CloudWatch event for scheduling the Kyvos BI Server.

• Secrets Manager for storing passwords, like Kyvos DB password, Active Directory password, and SMTP password (if configured)

• SecurityGroup for Kyvos and Databricks instances.

Next: Deploy Kyvos using CloudFormation Template