Yelp

Yelp Reroutes Cassandra Workloads to EKS Without Data Migration

Yelp's move to re-point existing EBS volumes at new pods on Amazon EKS is a refreshingly pragmatic approach to infrastructure migration.

3 min readInfoQ
Yelp Reroutes Cassandra Workloads to EKS Without Data Migration

Yelp's decision to move its Cassandra fleet from self-managed Kubernetes to Amazon EKS by re-pointing existing EBS volumes at new pods, rather than streaming terabytes of data between nodes, is a quiet masterclass in operational pragmatism. It's a reminder that not every migration demands a forklift or a firehose. By sidestepping the costly, slow, and risk-prone process of data streaming, Yelp proved that sometimes the smartest infrastructure play is the one that moves the least data. That insight matters far beyond their specific stack.

For teams wrestling with legacy Kubernetes clusters, the takeaway is direct: you may not need to rebuild your stateful workloads to modernize your control plane. Yelp's approach preserves the data where it already lives, attached to persistent volumes, and simply points new pods at it. This avoids the months of validation and downtime that typically accompany Cassandra migrations. It also aligns with a broader shift we're seeing across the industry, one where operational maturity means choosing the least disruptive path forward. Compare this to the approach in Qwik 2.0 RC Delivers Faster Reactivity With Script Blocks and Streamlined State, where performance gains come from fundamentally rethinking how reactivity is structured. Both innovations are about removing friction, but Yelp's is a systems-level version of the same principle: don't make users, or in this case, engineers, pay for complexity they don't need.

What makes Yelp's story especially instructive is that it challenges the default assumption that migrating stateful workloads to a new orchestration layer requires a data-level migration. That assumption has stalled countless projects. By detaching the control plane migration from the data plane, Yelp effectively decoupled two problems that are often treated as one. This is not a theoretical exercise; it's a pattern that any organization running Cassandra, Kafka, or similar stateful systems on self-managed Kubernetes can evaluate today. The practical consequence is a faster path to managed services without the operational drag of rebalancing rings or resyncing replicas. As teams explore how to Package coding agent skills as versioned plugins for smarter repo governance, they should ask whether their own infrastructure migrations are similarly overcomplicating what could be a simpler swap.

The open question Yelp leaves us with is whether this technique generalizes beyond EBS-backed Cassandra deployments. For teams using different storage backends or different NoSQL databases, the same principle may require more engineering to replicate. But that's exactly the kind of detail worth watching: if others adopt this pattern, we may see a quiet shift in how stateful workloads are migrated, less streaming, more re-pointing. For now, Yelp's move is proof that progress doesn't always require a dramatic rewrite. Sometimes it just requires the discipline to ask what work you can avoid doing.

From InfoQ

Yelp has moved its Cassandra fleet from legacy self-managed Kubernetes clusters to Amazon EKS by re-pointing existing EBS volumes at new pods, rather than streaming data between nodes.

Read the original at InfoQ