DataStax is taking the next step in the build-out of its managed Astra cloud service by making it serverless as the new default. They claim that with serverless, DataStax customers could realize 3 – 5x better TCO on a "typical consumer application" compared to provisioning for peak usage. For now, Astra is the only pure Apache Cassandra managed cloud service that is delivered as serverless and runs on a choice of public clouds: AWS, Azure, and GCP. The announcement is being made today in advance of the transition to serverless, which will activate next week.
With serverless, customers pay only for what they use; they do not need to specify or provision a set capacity of nodes. Serverless is best suited for scenarios where traffic is either likely to fluctuate or not be easily predictable; conversely, reserving set capacity is the more economical route for workloads that are predictable and stable.
As noted above, serverless will be the new default for Astra customers; however, those preferring the traditional provisioned capacity option, which will be called Reserved Capacity, will still have that option available.
With serverless, DataStax will provide free monthly credits up to a certain level of capacity, which will allow Astra customers to scale up and scale down subject to the upper limits that they set. As noted above, DataStax claims that this will deliver significant savings compared to existing approaches that provision for peak capacity.
By comparison, Amazon Keyspaces originally launched as a serverless offering. But there are major differences between Keyspaces and Astra. Keyspaces is an API-compatible implementation of Cassandra that uses a different storage engine, whereas Astra is a full implementation of open source Apache Cassandra. While Keyspaces only runs within a single region on AWS, Astra now runs on all three major clouds and can run across multiple regions. At launch, Astra serverless will be restricted to a single region, but we expect that that multi-region deployment will coming sooner rather than later.
Additionally, Astra has been implemented with a Kubernetes (K8s) operator to make it cloud-agnostic; that is currently unique among all cloud Cassandra implementations. Astra's K8s implementation in turn spurred a byproduct: the K8ssandra distribution that is being offered to the Cassandra community and as an option to customers seeking to run their own private clouds.
Originally released last spring, DataStax has been steadily adding enhancements since day one. Among them are support of Stargate endpoints that could turn Cassandra into a multi-model database; multi-region replication for high availability scenarios; and a choice multi-tenant or dedicated cluster support. While serverless is the new default option, customers can still opt for provisioned, reserved capacity, which is best suited for workloads with stable, predictable throughput. Astra supports a variety of APIs including REST, GraphQL, JSON, and of course, CQL (Cassandra Query Language).
But the nature of Cassandra, as a distributed, multi-master global-scaled database that promises near 100% uptime, is that it will tend to attract write-heavy workloads such as recording IoT and other sensory data; messaging systems; e-commerce product catalogs, recommendation and personalization engines; and media/gaming website user tracking. These workloads are likely to be highly variable. Guaranteeing reads and writes across multiple nodes, or in some cases, geographic regions, with always-on availability, is a challenge for capacity planning. So peak provisioning should be an issue for most customers. With serverless, customers only pay for what they use.
Astra Serverless will go live next week.
Disclosure: DataStax is a dbInsight client.