AWS CDK Redshfit Demo

Abstract

For getting started with AWS Redshift, this post guides you to provide Redshift cluster and all necessary other resources such as VPC, redshift IAM role, S3 bucket where we load data to Amazon Redshift and a small EC2 instance as PostgreSQL client using AWS CDK typescript.

AWS Redshift overview
Redshift cluster stacks
Deploy stacks
Working with Redshift Cluster
Conclusion

🚀 AWS Redshift overview

AWS Redshift is a cloud-based petabyte-scale data warehouse service offered as one of Amazon’s ecosystem of data solutions.
Based on PostgreSQL, the platform integrates with most third-party applications by applying its ODBC and JDBC drivers.
Amazon Redshift delivers fast query performance by using columnar storage technology to improve I/O efficiency and parallelise queries across multiple nodes
Redshit cluster overview
Node slices in compute nodes

🚀 Redshift cluster stacks

The Redshift cluster is in VPC and under a private subnet and security group, so we first create the VPC stack, and for saving cost, max availabilityZones is set to 1
```
  const vpc = new Vpc(this, `${prefix}-vpc`, {
      vpcName: `${prefix}-vpc`,
      maxAzs: 1
    });
```

We will use S3 bucket to store JSON/parquet data and then load data from S3 bucket to redshift cluster as the following flow.

Redshift needs a role so that it can consume that role to download data from S3 bucket

  const s3 = new Bucket(this, `${prefix}-data-ingest`, {
    bucketName: `${prefix}-data-ingest`,
    encryption: BucketEncryption.S3_MANAGED,
    removalPolicy: RemovalPolicy.DESTROY,
    enforceSSL: true,
    blockPublicAccess: BlockPublicAccess.BLOCK_ALL
  });

  const role = new Role(this, `${prefix}-role`, {
    roleName: `${prefix}-role`,
    assumedBy: new ServicePrincipal('redshift.amazonaws.com')
  });
  s3.grantRead(role);

In this post, the redshift cluster is in multi-node cluster with 2 compute nodes type DC2_LARGE (as default). masterUser is created with password stored in the secret manager (as the default option)

    const cluster = new Cluster(this, `${prefix}-cluster-demo`, {
      clusterName: `${prefix}-demo`,
      vpc: vpc,
      masterUser: {
        masterUsername: 'admin'
      },
      numberOfNodes: 2,
      clusterType: ClusterType.MULTI_NODE,
      removalPolicy: RemovalPolicy.DESTROY,
      roles: [role]
    });

Finally, it's optional to create an EC2 instance as postgresql client to access redshift database (as we can use Amazon Redshift query editor). In practice, the application will use JDBC URL to connect to Redshift cluster in a private network. Back to the EC2 instance, it requires the following:
- Only small instance types such as t3.small
- Allocated in the private subnet of redshift VPC
- Attached Security Group of the redshift cluster which already allows internal communication
- Instance profile with only AmazonSSMManagedInstanceCore to start Amazon SSM agent
- And user-data.sh script to install postgresql at first start

      const clusterSg = cluster.connections.securityGroups[0];
      clusterSg.addIngressRule(clusterSg, Port.allTcp(), "Allow internal access Redshift");

      const ec2 = new Instance(this, `${prefix}-psql`, {
        instanceName: `${prefix}-psql`,
        vpc: vpc,
        securityGroup: clusterSg,
        instanceType: InstanceType.of(InstanceClass.T3, InstanceSize.SMALL),
        machineImage: new AmazonLinuxImage({generation: AmazonLinuxGeneration.AMAZON_LINUX_2}),
        role: new Role(this, `${prefix}-ec2-ssm`, {roleName: `${prefix}-ec2-ssm`, assumedBy: new ServicePrincipal('ec2.amazonaws.com'), managedPolicies: [{managedPolicyArn: 'arn:aws:iam::aws:policy/AmazonSSMManagedInstanceCore'}]})
      });

      const userData = readFileSync(resolve(__dirname, './user-data.sh'), 'utf8');
      ec2.addUserData(userData);

🚀 Deploy stacks

The infrastructure of code for this project is ready we now deploy the stacks. It's up to you to use concurrency option for fasten deployment and require-approval to bypass confirmation of creating/updating/removing sensitive things.
```
  cdk deploy --concurrency 2 --require-approval never
```
Check redshift cluster
Go to the secret manager to get the master user password
Overview of all components

🚀 Working with Redshift Cluster

We use sample data in data_sample/amzn_reviews_en.json, you can find more in Amazon review data (2018)
From the JSON file, we use pyspark to convert it to parquet format
- Overview of Parquet: Parquet follows the Columnar Storage Model and is available to any project in the Hadoop Ecosystem. Unlike the traditional Sequential Storage Model where data is written in sequence, the Columnar Storage Model stores column values together. Although the Sequential Storage Model has advantages in processing transactions, it is not suitable for running Analytical Queries on Big Data.
- Install pyspark by running pip install pyspark. Note that it also requires java openjdk to use the tool
- Covert JSON file to parquet.
```
  ⚡ $ cd data_sample/
  ⚡ $ pyspark
  Python 3.10.4 (main, Jun 29 2022, 12:14:53) [GCC 11.2.0] on linux
  Welcome to
        ____              __
      / __/__  ___ _____/ /__
      _\ \/ _ \/ _ `/ __/  '_/
    /__ / .__/\_,_/_/ /_/\_\   version 3.3.0
        /_/
  >>> df = spark.read.json("amzn_reviews_en.json")
  >>> print("Schema: {}".format(df.schema))
  >>> df.show()
  >>> df.write.parquet("amzn_reviews_en.parquet")
```

Access the EC2 instance by using the SSM session

  ~ $ aws ssm start-session --target i-0be265f7c54177548 --region ap-southeast-1

Connect redshift cluster database

  [root@ip-10-0-147-8 bin]# psql -h sin-d1-redshift-demo.cnozo5w39dmk.ap-southeast-1.redshift.amazonaws.com -U admin -p 5439 -d default_db
  Password for user admin:
  psql (10.17, server 8.0.2)
  SSL connection (protocol: TLSv1.2, cipher: ECDHE-RSA-AES256-GCM-SHA384, bits: 256, compression: off)
  Type "help" for help.

  default_db=#
  default_db=# \c prodreview
  prodreview=#

Create JSON table view

  CREATE TABLE IF NOT EXISTS product_reviews_json(
  review_id           varchar(100)        NOT NULL distkey sortkey,
  product_id          varchar(100)        NOT NULL,
  stars               varchar(18)         NOT NULL,
  review_body         varchar(10000)      NOT NULL,
  review_title        varchar(1000)       NOT NULL,
  reviewer_id         varchar(128)        NOT NULL,
  language            varchar(20)         NOT NULL,
  product_category    varchar(100)        NOT NULL,
  primary key(review_id)
  );

Create a parquet table view

  CREATE TABLE IF NOT EXISTS product_reviews_parquet(
  language            varchar(20)         NOT NULL ENCODE lzo,
  product_category    varchar(100)        NOT NULL ENCODE lzo,
  product_id          varchar(100)        NOT NULL ENCODE lzo,
  review_body         varchar(10000)      NOT NULL ENCODE lzo,
  review_id           varchar(100)        NOT NULL distkey sortkey ENCODE lzo,
  review_title        varchar(1000)       NOT NULL ENCODE lzo,
  reviewer_id         varchar(128)        NOT NULL ENCODE lzo,
  stars               varchar(18)         NOT NULL ENCODE lzo,
  primary key(review_id)
  );

Upload data amzn_reviews_en.json and amzn_reviews_en.parquet to S3 and then load them to redshift database

Load amzn_reviews_en.json to the JSON table

  prodreview=# copy product_reviews_json
  prodreview-# FROM 's3://sin-d1-redshift-data-ingest/amzn_reviews_en.json'
  prodreview-# IAM_ROLE 'arn:aws:iam::107858015234:role/sin-d1-redshift-role'
  prodreview-# json 'auto ignorecase';
  INFO:  Load into table 'product_reviews_json' completed, 5000 record(s) loaded successfully.
  COPY

  prodreview=# SELECT COUNT(*) FROM product_reviews_json;
  count
  -------
    5000
  (1 row)

Load amzn_reviews_en.parquet to parquet table

  prodreview=# copy product_reviews_parquet
  FROM 's3://sin-d1-redshift-data-ingest/amzn_reviews_en.parquet'
  IAM_ROLE 'arn:aws:iam::107858015234:role/sin-d1-redshift-role'
  format as parquet;
  INFO:  Load into table 'product_reviews_parquet' completed, 5000 record(s) loaded successfully.
  COPY

  prodreview=# SELECT COUNT(*) FROM product_reviews_parquet;
  count
  -------
    5000
  (1 row)

Compare loading time

  select datediff(s,starttime,endtime) as duration,
  *
  from
  stl_query
  where query in (
      4231, /*  query id of json copy */
      4521 /*  query id of parquet copy */
  );

RUN QUERIES: query data for the question: "Do people who buy kitchen and grocery items leave higher ratings?"

  prodreview=# SELECT stars, COUNT(stars) total_ratings FROM product_reviews_json WHERE product_category='kitchen' or product_category='grocery'
  prodreview-# GROUP BY stars;
  stars | total_ratings
  -------+---------------
  5     |            92
  2     |            78
  4     |            63
  1     |            72
  3     |            66
  (5 rows)

  prodreview=# SELECT stars, COUNT(stars) total_ratings FROM product_reviews_parquet WHERE product_category='kitchen' or product_category='grocery'
  GROUP BY stars;
  stars | total_ratings
  -------+---------------
  3     |            66
  5     |            92
  2     |            78
  4     |            63
  1     |            72
  (5 rows)

🚀 Conclusion

Here we have demonstrated how to create a redshift cluster and its ETL using CDK typescript, you can update the cluster as well as add more S3 buckets or attach more roles to the redshift cluster for separate ETL processes through CDK stacks.
If you want to destroy all the resources created by the stack, simply run cdk destroy --all

Vu Dao