December 2016 Summaries

ScyllaDB 1.5 is a minor release of the open-source NoSQL database, ScyllaDB, which is compatible with Apache Cassandra and known for its high performance and low latency. This release emphasizes stability, bug fixes, and workload conditioning, a feature that allows ScyllaDB to self-tune to better handle various user workloads by automatically adjusting the rate of incoming requests based on disk performance. Key new features include the introduction of the "scylla-blocktune" script for optimizing data file and commitlog directories, and the disabling of Linux I/O scheduler settings to improve latency. The update also addresses notable bugs, such as unnecessary parallel queries in range scans, memory leaks during concurrent reads and writes, and startup issues on machines with many cores. Additionally, new metrics have been introduced to monitor performance under workload conditioning, specifically focusing on operations delayed due to disk speed limitations. The release encourages users of earlier versions to upgrade to maintain optimal performance and benefit from continued bug fixes and feature additions planned for future versions.

ScyllaDB 1.5 RC3, the third Release Candidate of the open-source, Apache-Cassandra-compatible NoSQL database, has been announced by the ScyllaDB team. Known for its superior performance and low latency, ScyllaDB 1.5 emphasizes stability, bug fixes, and workload conditioning, although it is not yet ready for production use. Users are encouraged to test this version and report any bugs to aid in stabilizing the final release. Recent bug fixes address issues such as improper memtable flushing criteria that could affect performance and potential exits during the initialization phase when encountering unrecognized sstable components.

ScyllaDB version 1.4.3, released by the ScyllaDB team in December 2016, is a bugfix update for the 1.4 stable branch, maintaining backward compatibility and supporting rolling upgrades. The release addresses several issues, including a bug with Systemd on CentOS 7.3 that prevented ScyllaDB from starting, a core dump problem during service shutdown, and a housekeeping service that failed to restart its timer upon ScyllaDB restarts. Additionally, the release corrects a kernel check documentation URL error. Contributions to this release come from multiple developers, each tackling specific areas such as systemd and component probing. The update process involved both bug fixes and code reverts to ensure system stability and functionality.

ScyllaDB introduces "Workload Conditioning," a concept designed to automate the tuning of NoSQL database systems by leveraging internal metrics to create feedback loops that ensure stable performance. This approach aims to eliminate common configuration questions related to request rates, disk bandwidth, and thread counts by implementing features that adjust system characteristics automatically. The article explains the write request rate adjustment feature in the upcoming ScyllaDB 1.5 release, which manages the balance between memory and disk speeds to prevent system overloads. By utilizing Direct I/O, ScyllaDB can precisely control data flow to the disk, aligning write and flush rates to optimize performance. The system employs Prometheus support for monitoring, offering insights into the internal processes and helping users maintain their SLAs without manual tuning. Future updates will continue to build on this foundation, enhancing features like memtable size determination and optimizing replica selection based on cache efficiency.

Eyal Gutkind and his colleague Nir Mayaan conducted a three-day ScyllaDB training session for developers, operations, and DBA professionals at Samsung SDS in Seoul, South Korea. The training included both lectures and hands-on practice sessions, covering the fundamentals of NoSQL databases and distributed systems, progressing to data modeling and analytics tools on ScyllaDB. The Samsung engineers demonstrated substantial expertise in NoSQL databases, leading to insightful discussions. The training took place at Samsung's new R&D center in the Seocho district, which provided an excellent environment for learning and collaboration. The experience was enriching, and the trainers look forward to future sessions with Samsung.

ScyllaDB 1.5 RC2, the second of three planned release candidates for ScyllaDB 1.5, has been announced by the ScyllaDB team, emphasizing stability, bug fixes, and workload conditioning. Although not yet production-ready, the release addresses several issues from previous versions, including a Thrift regression, service shutdown core dumps, incorrect documentation URLs, lengthy startup times with large data volumes, repository update complications, a Systemd bug on CentOS 7.3, and a non-restarting housekeeping timer. As an open-source, Apache-Cassandra-compatible NoSQL database, ScyllaDB is noted for its superior performance and low latency, and users are encouraged to report any bugs to aid in stabilizing version 1.5.