No matter the industry or level of maturity within AWS, our customers require better visibility into their AWS Glue usage. Better visibility can lend itself to gains in operational efficiency, informed business decisions, and further transparency into your return on investment (ROI) when using the various features available through AWS Glue.
As your company grows, you should be able to answer simple questions about your AWS Glue usage, such as the following:
- Where am I spending the most with AWS Glue?
- Where can I save the most by taking advantage of new AWS Glue features?
- What does my overall usage look like using AWS Glue?
AWS offers services such as Amazon QuickSight, a serverless business intelligence (BI) service that lets you centralize this view and even ask natural language questions of your data, using Amazon QuickSight Q. QuickSight can give business leaders and their technology counterparts a common landscape for reporting important details of their usage, providing automated narratives to bridge communication gaps.
In this post, we explore how to combine AWS Glue usage information and metrics with centralized reporting and visualization using QuickSight. This can provide you with a more comprehensive view of your usage and tools to help you dive deep into your AWS Glue job run environment. You have metrics available per job run within the AWS Glue console, but they don’t cover all available AWS Glue job metrics, and the visuals aren’t as interactive compared to the QuickSight dashboard.
Although we don’t cover optimizing your jobs for costs in this post, you can refer to Monitor and optimize cost on AWS Glue for Apache Spark to learn how to fine-tune your AWS Glue jobs for performance, efficiency ,and cost-optimization.
Let’s dive in!
The following diagram illustrates the architecture for the given solution. At a high level, a scheduled event triggers an orchestration flow consisting of multiple data, compute, and analytics resources—the output of which culminates as a set of visuals in a BI dashboard.
Now let’s dig into the technical details involved in this solution.
An AWS Step Functions workflow is scheduled to run once per hour through Amazon EventBridge, which triggers an AWS Lambda function that calls the AWS Glue
GetJobRun APIs. We parse this data to check for jobs that have succeeded, stopped, or failed in the past hour, as well as any streaming jobs. The metadata is extracted from each job run, including information like runtime, start time, end time, auto scaling, number of workers, and worker type, and is written to an Amazon DynamoDB table with TTL (time to live) enabled to ensure the table doesn’t grow too large.
We move into a parallel state to check two tables that Amazon Athena writes the output of the federated queries to. Athena first checks to make sure the tables exist in Amazon Simple Storage Service (Amazon S3), where the data will be stored. If the tables don’t exist, Athena creates them. One federated query gathers AWS Glue metric data from Amazon CloudWatch metrics; the other gathers data from the DynamoDB table where Lambda writes the AWS Glue job metadata it’s collecting. Both federated queries utilize appropriate filtering in order to only scan the necessary data from each source.
There is a choice state for each branch. If there is no new data to be added to a table in Amazon S3, the state ends and waits for the other to complete. For example, there could be an AWS Glue job that is running while the step is evaluating. In this case, the metrics for the job would be inserted in the table on Amazon S3, but the metadata from DynamoDB wouldn’t arrive until the following hour after the job has succeeded, stopped, or failed.
When new metrics or metadata are found, Athena inserts this data to the metrics or metadata tables in Amazon S3, which are both partitioned by the hour. After the data is inserted, the final steps call the QuickSight CreateIngestion API, which triggers data ingestion into QuickSight SPICE to power interactive analysis. At this point, the workflow has finished running and will run again the following hour.
In the following sections, we show you how to set up the solution, explore the dashboards, and configure alarms.
The code for this solution can be found at the AWS samples GitHub repository.
You should have the following prerequisites:
- An AWS account with AWS Identity and Access Management (IAM) privileges sufficient to create the solution resources
- QuickSight Standard or Enterprise Edition with a QuickSight user created
- An AWS Cloud9 integrated development environment (IDE) or your local machine using your preferred IDE with the following packages installed:
- The AWS Cloud Development Kit (AWS CDK) bootstrapped in your target AWS account and Region
Deploy solution resources with the AWS CDK
To provision the resources that build the dashboard and keep it up to date, we provide steps to download and deploy the solution via the AWS CDK. The solution was developed with cost-optimization as a priority, but some resources in the stack will incur costs once deployed.
This solution generates the following resources:
- IAM role
- EventBridge rule
- Step Functions state machine
- Lambda function
- S3 bucket
- Two AWS Glue tables and one AWS Glue database
- DynamoDB table
- Athena queries invoked by Step Functions
- QuickSight data source, dataset, analysis, and dashboard
To deploy the solution, complete the following steps:
- Clone the source code from AWS samples GitHub repository to the client:
- Bootstrap your AWS CDK app:
- Deploy the solution with the required parameters:
- The first parameter is for a new S3 bucket to be created, which holds the AWS Glue metrics and metadata.
- The second parameter is required in order for QuickSight to assign permissions to the user who will manage the assets. Refer to Managing user access inside Amazon QuickSight to find your existing QuickSight users.
If your deployment fails, make sure you installed the AWS CDK library and rerun
cdk deploy after installing:
The deployment may take up to 10 minutes.
After the solution is deployed, the Step Functions state machine will evaluate once per hour if it should ingest data into QuickSight. You can run some AWS Glue jobs after the stack is deployed and check the QuickSight dashboard in the next hour or two, where the job metadata and metrics will be populated for your analysis.
Explore the dashboard
The dashboard contains two sheets: Glue Jobs and Glue Metrics.
The Glue Jobs sheet includes all of the metadata about your AWS Glue job runs, including AWS Glue for Apache Spark, AWS Glue for Ray, and AWS Glue streaming ETL. Most of the visuals also have a hierarchy that you can drill down into with QuickSight, going as low as each specific job run ID. You can use controls to filter by date, job name, and job run ID.
In the following demonstration, you will see the pivot table, which is a simple view of all our job metadata, including estimated cost per job and job run. We open up a job name and see the different job runs. There is one individual job run that we would like to inspect the metrics on, so we choose the job name and choose View metrics for job run id:
The Glue Metrics sheet is built to reflect the documentation we provide in AWS Glue resource monitoring. This documentation helps explain each visual in the dashboard. You can use the Glue Metrics sheet to view aggregated metrics across all jobs, a single job, or down to the job run ID.
To populate the Glue Metrics sheet, your AWS Glue jobs must be enabled to capture metrics in CloudWatch.
Set up alerts
Setting up alerts on measures is also straightforward to do in QuickSight. To do so, choose (right-click) one of the tracked measures on either worksheet and choose Create Alarm. This will bring you to the configuration page to set up the metric you’d like to be alerted on.
The dashboard is designed to give you the freedom to alter it and make your own visualizations with the metadata and metrics that are provided to you. If you want even more insight into cost, consider deploying the CUDOS dashboard as well!
If you no longer need the dashboard, delete the CDK app:
In this post, we talked about the importance of having observability of your AWS Glue jobs and provided an AWS CDK app that deploys a QuickSight dashboard for you. We hope this helps you optimize your AWS Glue environment using the insights the dashboard provides. To learn about event-based alerting for your AWS Glue for Apache Spark and Ray jobs, refer to Automate alerting and reporting for AWS Glue job resource usage.
About the authors
Michael Hamilton is a Sr Analytics Solutions Architect focusing on helping enterprise customers in the south east modernize and simplify their analytics workloads on AWS. He enjoys mountain biking and spending time with his wife and three children when not working.
Cody Penta is a Solutions Architect at Amazon Web Services and is based out of Charlotte, NC. He has a focus in security and CDK, and enjoys solving the really difficult problems in the technology world. Off the clock, he loves relaxing in the mountains, coding personal projects, and gaming.
Angus Ferguson is a Solutions Architect at AWS who is passionate about meeting customers across the world, helping them solve their technical challenges. Angus specializes in Data & Analytics with a focus on customers in the financial services industry.