How Gupshup built their multi-tenant messaging analytics platform on Amazon Redshift

Gupshup is a leading conversational messaging platform, powering over 10 billion messages per month. Across verticals, thousands of large and small businesses in emerging markets use Gupshup to build conversational experiences across marketing, sales, and support. Gupshup’s carrier-grade platform provides a single messaging API for 30+ channels, a rich conversational experience-building tool kit for any use case, and a network of emerging market partnerships across messaging channels, device manufacturers, ISVs, and operators.

Objective

Gupshup wanted to build a messaging analytics platform that provided:

• Build a platform to get detailed insights, data, and reports about WhatsApp/SMS campaigns and track the success of every text message sent by the end customers.
• Easily gain insight into trends, delivery rates, and speed.
• Save time and eliminate unnecessary processes.

About Redshift and some relevant features for the use case

Amazon Redshift is a fully managed, petabyte-scale, massively parallel data warehouse that offers simple operations and high performance. It makes it fast, simple, and cost-effective to analyze all your data using standard SQL and your existing business intelligence (BI) tools. Amazon Redshift extends beyond traditional data warehousing workloads, by integrating with the AWS cloud with features such as querying the data lake with Spectrum, semistructured data ingestion and querying with PartiQL, streaming ingestion from Amazon Kinesis and Amazon MSK, Redshift ML, federated queries to Amazon Aurora and Amazon RDS operational databases, and federated materialized views.

In this use case, Gupshup is heavily relying on Amazon Redshift as their data warehouse to process billions of streaming events every month, performing intricate data-pipeline-like operations on such data and incrementally maintaining a hierarchy of aggregations on top of raw data. They have been enjoying the flexibility and convenience that Amazon Redshift has brought to their business. By leveraging the Amazon Redshift materialized views, Gupshup has been able to dramatically improve query performance on recurring and predictable workloads, such as dashboard queries from Business Intelligence (BI) tools. Additionally, extract, load, and transform (ELT) data processing is sped up and made easier. To store commonly used pre-computations and seamlessly utilize them to reduce latency on ensuing analytical queries, Redshift materialized views feature incremental refresh capability which enables Gupshup to be more agile while using less code. Without writing complicated code for incremental updates, they were able to deliver data latency of roughly 15 minutes for some use cases.

Overall architecture and implementation details with Redshift Materialized views

Gupshup uses a CDC mechanism to extract data from their source systems and persist it in S3 in order to meet these needs. A series of materialized view refreshes are used to calculate metrics, after which the incremental data from S3 is loaded into Redshift. This compiled data is then imported into Aurora PostgreSQL Serverless for operational reporting. The ability of Redshift to incrementally refresh materialized views, enabling it to process massive amounts of data progressively, the capacity for scaling, which utilizes concurrency and elastic resizing for vertical scaling, as well as the RA3 architecture, delivers the separation of storage and compute to scale one without worrying about the other, led Gupshup to make this choice. Gupshup chose Aurora PostgreSQL as the operational reporting layer due to its anticipated increase in concurrency and cost-effectiveness for queries that retrieve only precalculated metrics.

Incremental analytics is the main reason for Gupshup to use Redshift. The diagram shows a simplified version of a typical data processing pipeline where data comes via multiple streams. The streams need to be joined together, then enriched by joining with master data tables. This is followed by series of joins and aggregations. All this needs to be performed in incremental manner, providing 30 minutes of latency.

Gupshup uses Redshift’s incremental materialized view feature to accomplish this. All of the join, enrich, and aggregation statements are written using sql statements. The stream-to-stream joins are performed by ingesting both streams in a table sorted by the key fields. Then an incremental MV aggregates data by the key fields. Redshift then automatically takes care of keeping the MVs refreshed incrementally with incoming data. The incremental view maintenance feature works even for hierarchical aggregations with MVs based on other MVs. This allows Gupshup to build an entire processing pipeline incrementally. It has actually helped Gupshup reduce cycle time during the POC and prototyping phases. Moreover, no separate effort is required to process historical data versus live streaming data.

Apart from incremental analytics, Redshift simplifies a lot of operational aspects. E.g., use the snapshot-restore feature to quickly create a green experimental cluster from an existing blue serving cluster. In case the processing logic changes (which happens quite often in prototyping stages), they need to reprocess all historical data. Gupshup uses Redshift’s elastic scaling feature to temporarily scale the cluster up and then scale it down when done. They also use Redshift to directly power some of their high-concurrency dashboards. For such cases, the concurrency scaling feature of Redshift really comes in handy. Apart from this, they have a lot of in-house data analysts who need to run ad hoc queries on live production data. They use the workload management features of Redshift to make sure their analysts can run queries while ensuring that production queries do not get affected.

Benefits realized with Amazon Redshift

• On-Demand Scaling
• Ease of use and maintenance with less code
• Performance benefits with an incremental MV refresh

Conclusion

Gupshup, an enterprise messaging company, needed a scalable data warehouse solution to analyze billions of events generated each month. They chose Amazon Redshift to build a cloud data warehouse that could handle this scale of data and enable fast analytics.

By combining Redshift’s scalability, snapshots, workload management, and low-operational approach, Gupshup provides data-driven insights in less than 15 minutes analytics refresh rate.

Overall, Redshift’s scalability, performance, ease of management, and cost effectiveness have allowed Gupshup to gain data-driven insights from billions of events in near real-time. A scalable and robust data foundation is enabling Gupshup to build innovative messaging products and a competitive advantage.

The incremental refresh of materialized views feature of Redshift allowed us to be more agile with less code:

• For some use cases, we are able to provide data latency of about 15 minutes, without having to write complex code for incremental updates.
• The incremental refresh feature is a main differentiating factor that gives Redshift an edge over some of its competitors. I request that you keep improving and enhancing it.

“The incremental refresh of materialized views feature of Redshift allowed us to be more agile with less code”

– Pankaj Bisen, Director of AI and Analytics at Gupshup.

About the Authors

Shabi Abbas Sayed is a Senior Technical Account Manager at AWS. He is passionate about building scalable data warehouses and big data solutions working closely with the customers. He works with large ISVs customers, in helping them build and operate secure, resilient, scalable, and high-performance SaaS applications in the cloud.

Gaurav Singh is a Senior Solutions Architect at AWS, specializing in AI/ML and Generative AI. Based in Pune, India, he focuses on helping customers build, deploy, and migrate ML production workloads to SageMaker at scale. In his spare time, Gaurav loves to explore nature, read, and run.