The developers of the ScyllaDB database have announced the project's transition to a proprietary license, which limits its scope of use. The ScyllaDB 6.2.x branch will be the last available under the AGPL license. Interested parties can fork the project and continue developing the ScyllaDB codebase under the old license if they wish. <\/p>\n
Previously, the ScyllaDB project followed an Open Core development model, where the core part of the product is developed under a free license, while the extended functionalities are closed and provided only to commercial license holders. Under the new licensing scheme, the publication of the open edition, ScyllaDB OSS, will cease, and the previously closed commercial product, ScyllaDB Enterprise, will start being distributed with source code under a proprietary license. Users will be able to use the full version of ScyllaDB Enterprise for free under certain conditions, instead of ScyllaDB OSS. <\/p>\n
The new license prohibits the use of the software to compete with the products and services of ScyllaDB developers, as well as for creating cloud systems 'application as a service' (SaaS, software-as-a-service) and commercial systems 'DB as a service' (dBaaS, database-as-a-service). The license does not restrict the use of the free product for commercial purposes and in working environments, provided that the total storage size across all nodes in the cluster does not exceed 10TB, and no more than 50 vCPUs are used for data processing. Users who do not meet the specified criteria are advised to purchase a commercial license. <\/p>\n
For example, ScyllaDB can be used for free on a cluster of three nodes, if each node contains 16 logical CPU cores and disk storage of 3TB. This configuration can handle 100-200 thousand operations per second. Distribution of ScyllaDB and modification of the code is allowed, provided that it is supplied under the same license, that a complete list of all changes is provided, and that the original author information is cited. <\/p>\n
The reason for changing ScyllaDB's distribution policy was the desire to unify and optimize the development of competing commercial and open-source versions, as maintaining them separately created complexities, consumed substantial effort, and drained resources. Due to the complexity of its internal architecture, ScyllaDB's codebase developed exclusively through the company's employees, and the open-source project never had external contributors passing along their modifications. A turning point was the implementation of the Raft consensus algorithm, tables, and built-in support for the S3 API, resulting in many auxiliary functions being transferred from external applications to the core of the database system. <\/p>\n
Ultimately, it was decided to combine the open-source and commercial products, which on one hand simplifies development, while on the other expands the capabilities of the free version. For example, users who fall within the free limits will gain access to features such as LDAP support, PGO optimizations (reducing text latency by 33% and increasing performance by up to 50%), file-level streaming mode (accelerating node addition\/removal by up to 30 times), the ability to set priorities for different workloads, support for compressing RPC traffic between nodes using the ZSTD algorithm, improved data packaging strategies (reducing storage size by up to 35%), support for encryption, a Kubernetes operator for ScyllaDB, and a long release support cycle. <\/p>\n
ScyllaDB allows for the creation of distributed, scalable, and fault-tolerant NoSQL systems, storing data in the form of associative arrays (hashes) with multiple levels of nesting. A SQL-like language called CQL (Cassandra Query Language) can be used for structured queries. Clusters based on ScyllaDB provide linear scalability, where performance directly correlates with the number of CPU cores. Data placed in the database is automatically replicated across multiple nodes, and in the event of a node failure, its functions are seamlessly taken over by other nodes. Adding, updating, and deleting nodes in the cluster occurs without interrupting operations and without reconfiguring other nodes. <\/p>\n
The ScyllaDB database was founded ten years ago by Avi Kiviti and Dor Laor, who at the time created a hypervisor. KVM<\/a> The project was initiated as an attempt to rewrite Apache Cassandra from Java to C++ to achieve higher performance. The database is API compatible with Apache Cassandra and Amazon DynamoDB. Tests have shown that ScyllaDB demonstrates a throughput increase of 2-5 times compared to Apache Cassandra. It is noted that a cluster based on Apache Cassandra can be replaced by a ScyllaDB cluster containing 10 times fewer nodes, yet outperforming it by 42% despite the smaller number of nodes. <\/p>\n