Migrating Critical Workloads from MongoDB to Amazon DocumentDB with Hudl

Founded in 2006 in Lincoln, Nebraska, Hudl provides smart cameras and associated software to help sports teams capture their games, annotate and analyze them, and use data to improve their play. More than 300,000 teams across more than 40 sports globally depend on Hudl’s services, which are powered by document databases that store sports footage; metadata, such as camera angles or team formations; and other pertinent information, such as box scores and notes.

Hudl had been hosting its MongoDB databases on Amazon Elastic Compute Cloud (Amazon EC2), which provides secure and resizable compute capacity for virtually any workload. But the databases couldn’t accommodate the latest in Amazon EC2 processors, were no longer supported, and received sparse attention. Engineers felt that disaster recovery was unreliable, and the company faced scalability challenges during times of high traffic due to poorly understood custom tooling. “When we needed a bigger cluster, it required specialized knowledge and an understanding of infrastructure at a deep level,” says Aaron Kalair, lead infrastructure engineer at Hudl. “The process would take days, and it came with risk.”

As an AWS customer, Hudl set out to migrate an initial set of 33 MongoDB databases to Amazon DocumentDB, which lets engineers run the same application code and use the same drivers and tools they use with MongoDB. In January 2023, the company began to research how to migrate safely, quickly, and with minimal involvement from the product teams. It turned to AWS Database Migration Service (AWS DMS), which has been trusted by customers globally to securely migrate more than 1 million databases with minimal downtime.

At first, Hudl outsourced the migration to a contractor and asked the Hudl product teams to perform quality assurance before the databases flipped to production. But because the contractor’s work required more input from the Hudl team than expected, the company discarded this approach and worked on the migrations directly. The first five migrations took 3 months, leading Hudl to reevaluate its process.

The Hudl team refined how it carried out the steps, batching them across multiple databases rather than doing them sequentially. To verify quality, engineers wrote tooling that captured queries on MongoDB and replayed them against Amazon DocumentDB, visualizing the results as histograms. The migration of the next 22 databases took just 6 months instead of the 2 years the platform team had projected, with virtually zero downtime.

Along the way, Hudl met regularly with the Amazon DocumentDB team to review the road map and get expert technical guidance. For example, during the migration, the team recommended that Hudl upgrade to Amazon DocumentDB 5.0, which has cutting-edge features such as client-side field level encryption and enhanced operator support. “That was a big win that saved us a significant amount of work,” says Nestor Rodriguez, a senior engineer at Hudl. “Working alongside the Amazon DocumentDB team massively helped speed up our migration and remove some blockers.”

Using Amazon DocumentDB, engineers have cut the time to resize compute nodes from days to just 30 minutes, and teams no longer require special knowledge. Instead, Hudl uses Terraform templates to spin up new databases. “It’s very mechanical; just copy and paste the same things to repeat the process,” says Kalair. “It’s fairly low effort.” Now, Hudl’s platform is well equipped to accommodate instances based on AWS Graviton processors, a family designed to deliver the best price performance for cloud workloads running in Amazon EC2.

Hudl also benefits from automated alerts through Amazon CloudWatch, a service that monitors applications, responds to performance changes, optimizes resource use, and provides insights into operational health. For example, engineers can set alarms for slow queries or COLLSCANS. In fact, Hudl has repaired more than 10 COLLSCANs as a result of the new alerting functionality. “We have solved dozens of queries,” says Kalair. “They had always been happening before, but no one knew about them. Using Amazon CloudWatch, we’re now alerted to them.”

Metrics from Amazon CloudWatch load automatically into Amazon DocumentDB Performance Insights, a feature that adds to the existing Amazon DocumentDB monitoring features with a dashboard, helping Hudl engineers monitor load, analyze queries, and identify performance issues.

Because the Amazon DocumentDB architecture separates compute from storage, Hudl runs two compute nodes instead of three while maintaining better data durability than it previously had done. “By using Amazon DocumentDB, we’re doubling down on our core competency—creating unparalleled sports video and data experiences for coaches and athletes,” says Andy Pryor, Hudl’s senior director of engineering. “This shift allows us to focus on innovating tools that help our users succeed, while AWS handles the reliability and scalability of the infrastructure we rely on.”

By the first half of 2025, Hudl plans to complete the migration of its sharded databases. Additionally, it plans to significantly reduce the complexity of its setup by migrating its five sharded clusters to unsharded clusters on Amazon DocumentDB, further simplifying database management. After it completes its migration, Hudl estimates it will save 37 percent annually in operational cost using Amazon DocumentDB. “It’s a no-stress migration,” says Kalair. “We are at the point where Amazon DocumentDB is a boring, reliable database for us, which is what we want. For our mission-critical databases, it just works.”