April 2017 Summaries

ScyllaDB 1.6.4, a bugfix release for the ScyllaDB 1.6 stable branch, has been announced, offering backward compatibility and support for rolling upgrades. Users transitioning from version 1.5.x to 1.6.4 are advised to review the release notes due to changes in certain metrics. The update addresses several critical bugs, including issues with CQL SELECT queries from large partitions, streaming problems during Leveled Compaction Strategy repairs that could lead to out-of-memory errors, and a rare gossip bug causing node state changes to result in system exits. Users are encouraged to report any issues encountered during the upgrade process.

In a comparative analysis of data storage capabilities, ScyllaDB 1.6 outperforms Apache Cassandra 3.0.9 by storing and retrieving at least double the amount of data per node, resulting in reduced operational complexity and expenses. Recent benchmarks demonstrated that ScyllaDB managed to ingest 1.3TB of data in under 12 hours, including compaction, whereas Apache Cassandra required over 28 hours for data ingestion and an additional 37 hours for compaction. This performance difference, in which ScyllaDB proved to be five times faster, was tested using Amazon Web Services EC2 instances with i2.8xlarge specifications, highlighting ScyllaDB's ability to condense cluster footprints and maintain high availability without detriment to performance. ScyllaDB's efficiency in handling large datasets offers a significant advantage to developers and database operators looking to optimize their data infrastructure.

ScyllaDB 1.6.3 is a bugfix release for the ScyllaDB 1.6 stable branch, focusing on backward compatibility and supporting rolling upgrades, with essential updates for users transitioning from version 1.5.x. Key fixes include renaming duplicate storage_proxy metrics, resolving commitlog replay errors after unexpected restarts, opening the Prometheus port by default in the Docker image, and preventing crashes in the golang driver, gocql, when certain logging features are enabled. Additionally, it addresses issues with node-to-node SSL logging messages and startup failures of the ScyllaDB service after setup and server restarts, ensuring a smoother experience for users and administrators.

ScyllaDB 1.7 introduces experimental support for counters, a special type of column designed for efficient counting operations with atomic updates, compatible with Cassandra 2.1 or later. Counters only allow incrementing, decrementing, reading, or deleting their values and have unique limitations, such as not being able to be set to a specific value once deleted and not being used in primary keys. Implemented using state-based conflict-free replicated data types (CRDT), counters enable local atomic operations without node synchronization, though initial updates remain non-idempotent, preventing safe client retries in case of failure. The internal mechanism involves counter shards, owned by specific nodes, and synchronized through a leader-based update process that minimizes the need for complex consensus protocols. During reads, counter values are aggregated from all counter shards, and standard eventual consistency limitations apply. ScyllaDB plans a more efficient counter implementation in its upcoming release 1.8, which is expected to resolve current inefficiencies.