ScyllaDB review

ScyllaDB is a distributed NoSQL database server that stores application data across a cluster. It sits behind application services as a persistence layer, rather than acting as a standalone app, an ORM or a hosted-only API. ScyllaDB Cloud is the vendor’s database-as-a-service option; ScyllaDB Enterprise is the self-managed option for teams running their own infrastructure.

The product targets high-throughput, low-latency workloads, including feature stores, inference data and contextual vector search. The homepage describes workloads spanning millions of operations or features and billions of embeddings, with storage scaling to petabytes and predictable single-digit millisecond latencies. It also positions ScyllaDB as a replacement for DynamoDB and Cassandra. Those are workload claims, not a guarantee that an existing schema or application can move without changes. Treat it as a purpose-built distributed data store: validate access patterns, partitioning and operational requirements before adopting it.

The homepage presents ScyllaDB as a database, not a general-purpose web API or application framework. It describes ScyllaDB Cloud as a DBaaS and points prospective users to getting started, comparison and use-case material, but it does not name a programming language, SDK, client protocol or CLI. Build against the documented database client and driver interfaces, and confirm supported languages, protocol versions and driver compatibility in ScyllaDB’s docs before choosing an application stack.

The integration work is therefore more than creating a connection string. Your team needs to model queries and partitions around the workload, then verify how client retries, consistency choices and cluster topology fit the application. ScyllaDB offers workshops on getting started and DynamoDB data modeling, according to its homepage. Those resources can help frame an evaluation, but teams should check the technical documentation for deployment instructions, migration paths and supported integrations. The homepage does not establish that a DynamoDB or Cassandra replacement is drop-in compatible.

ScyllaDB lists a 30-day Developer Free Trial and a 48-hour Production Evaluation, but no free plan. Cloud Standard uses resource-based pricing with no per-request charges. The vendor does not publish monthly or annual plan prices, so teams need a quote based on their deployment. Standard lists one daily backup and 8x5 P1 support with a two-hour response; its listed uptime SLA conflicts with the FAQ, which gives different figures. Confirm the SLA before relying on it.

Professional adds Bring Your Own Account deployment, multi-region active-active HA/DR and 24x7 P1 support with a one-hour response. Premium adds a 15-minute 24x7 P1 response, Bring Your Own Encryption Key and a dedicated technical account manager. Self-managed ScyllaDB Enterprise requires a custom quote, with licensing based on provisioned vCPUs and cluster sizing. On-demand hourly rates or Flex Credits, cross-AZ or cross-region transfer and backup storage beyond the included daily backup can add cost. The pricing page lists savings of up to 70% or more for one- and three-year contracts versus on-demand rates.

ScyllaDB offers a self-managed Enterprise product, but self-hosting does not mean the commercial license is free. Current license terms matter: a commenter in one Reddit thread says ScyllaDB changed its license and that a commercial license is required above 50 vCPUs, including hyper-threading, or 10TB of available storage. Treat that as a user report, not a substitute for the current license; ask ScyllaDB which terms apply to your version and deployment.

The available pricing details identify ScyllaDB Enterprise as custom-quoted and do not establish a current open-source license for every edition. Before adopting it, review the license and contribution terms with your legal team. The vendor’s homepage and pricing details do not describe a data-export workflow or migration tooling. Ask for supported export formats, restore procedures and a practical exit path, then test them against your schema and workload. Switching from a distributed NoSQL database may require application and data-model changes, not just moving stored data.

ScyllaDB’s homepage claims predictable single-digit millisecond latencies and scale to petabytes. Those claims depend on workload and design; confirm them against your own query patterns and failure requirements. Rate limits are not published in the pricing details. For Cloud, verify the applicable SLA: Standard’s plan listing says 99.99% uptime, while its FAQ lists 99.9%. Ask how backups, restore times, maintenance and incident response work for the deployment you are buying.

Developers on Reddit report that ScyllaDB requires specific schemas and partitioning to meet performance expectations, and that maintenance operations need planning across replicas. A commenter also points to the absence of joins as a reason to separate analytics from active records. Other Reddit users describe Kubernetes operations as difficult, while another argues that the product’s quality and testing improved over time. These are individual reports, not a representative survey. Before production, review deployment guidance, security and compliance controls with the vendor, and test failover, recovery and capacity changes with your actual data model.