March 2016 Summaries

ScyllaDB 1.0, the first production-ready release of the open-source, Apache-Cassandra-compatible NoSQL database, has been announced by the ScyllaDB team. This release focuses on delivering superior performance and consistent low latency, with a commitment to fixing only critical bugs in the 1.0.x branch while continuing to develop new features for future versions. The release schedule includes monthly timed-based updates, with version 1.1 expected by the end of April. Although no new functionality has been added since the Release Candidate 2, significant testing efforts have been made to ensure reliability. Users are encouraged to report any issues and engage with the community through the scylladb-users mailing list and other channels for updates on future developments.

ScyllaDB 1.0 RC2 is a Release Candidate of the open-source, Apache-Cassandra-compatible NoSQL database, known for its high performance and low latency. The release focuses on critical bug fixes in branch-1.0, while new features and bug fixes continue on the master branch. Key updates include an enhanced Developer Mode that allows ScyllaDB to run without performance enhancements when necessary, although it is not recommended for production due to reduced performance. The release requires running scylla_io_setup to configure I/O queues, essential for new installations unless in Developer Mode, and introduces Controlled Repair, which allocates resources more fairly by separating repair writes from ordinary write requests. The update also includes a tcp keepalive feature for maintaining connections in certain networking environments and numerous bug fixes addressing node failure handling, shutdown and boot failures, and packaging issues on Ubuntu. ScyllaDB is progressing toward a General Availability release in the first quarter of 2016, with further details available on the ScyllaDB wiki and status page.

ScyllaDBTop is a diagnostic tool designed to provide real-time monitoring of ScyllaDB nodes directly from the terminal by connecting to the collectd data gathering service, allowing users to view a variety of metrics without the need for complex monitoring setups. This lightweight program requires no additional dependencies beyond the standard Linux distribution and can be configured to display both ScyllaDB-specific and general metrics by leveraging collectd plugins. ScyllaDBTop, included in the ScyllaDB package and preinstalled on its Amazon EC2 AMI, is not meant to replace comprehensive monitoring solutions like ScyllaDB Monitoring or Grafana but rather serves as a complementary tool for quick and straightforward node diagnostics. The program must be run on the same host as collectd, which can be configured to use various plugins to monitor network errors, disk I/O, and more, enhancing its flexibility and utility in managing and understanding system performance.

Nadav Har’El delivered an insightful presentation on the design and future plans of ScyllaDB during the Generalist Engineer meetup in Tel Aviv, highlighting the improvements that make ScyllaDB a significantly faster alternative to Apache Cassandra. The talk delved into the Seastar C++ framework that underpins ScyllaDB, illustrating how its architecture contributes to a tenfold increase in performance. The presentation, available on SlideShare by Tzach Livyatan, forms part of ScyllaDB's ongoing efforts to engage with the community and share advancements. Interested individuals are encouraged to subscribe to ScyllaDB’s newsletter, blog RSS feed, or follow their Twitter account to stay informed about future events and developments. It is important to note that while Apache and Apache Cassandra are referenced trademarks of the Apache Software Foundation, no endorsement is implied.

ScyllaDB's longevity testing, a crucial component of its testing suite, focuses on evaluating the stability and reliability of ScyllaDB clusters over extended periods, using a variety of stress tests and fault injections to identify potential issues in long-running deployments. This integration test, unlike unit tests, assesses the entire system as deployed on AWS, including a cluster of nodes subjected to stress via tools like cassandra-stress and disruptions orchestrated by a component called Nemesis. These disruptions mimic real-world failures and include actions such as stopping and restarting instances, data corruption, and node decommissioning. The testing process aims to ensure that the clusters can maintain functionality despite these induced failures, with successful tests indicating resilience and robustness of the system. However, the tests have revealed critical issues such as nodes running out of space and failures in node decommissioning. Future plans for the longevity test suite include expanding to multiple cloud providers and incorporating multi-datacenter testing capabilities, with the source code available on GitHub for community involvement and further development.

The blog post introduces a live blogging series by Ori Lahav, focusing on the deployment of ScyllaDB at OutBrain. It invites readers to follow the series for insights into the testing process and to stay updated by subscribing to the blog's RSS feed or following ScyllaDB on Twitter. The post is part of a broader set of content related to performance evaluations and database strategies, highlighting topics such as the impact of compaction strategies on performance and the use of ScyllaDB as a backend for JanusGraph and Zenly's database replacement journey.

ScyllaDB version 0.19, an open-source, Apache-Cassandra-compatible NoSQL database, has been released with a focus on stability and bug fixes, alongside several notable updates. This release introduces a program called iotune, which optimizes I/O settings during installation, and implements a formal specification of internal protocols to ensure backward compatibility and facilitate rolling upgrades. Enhancements also include a new I/O scheduler that improves performance on systems with slower disks by balancing requests and a CLI tool named scyllatop for monitoring key metrics. These updates aim to streamline ScyllaDB's performance and prepare it for a GA release in the first quarter of 2016.

The blog post by Shlomi Livne focuses on the distributed testing (dtest) framework for ScyllaDB, which extends upon the original dtest project developed for Apache Cassandra. It explains how dtests function as comprehensive black-box tests to ensure ScyllaDB's compatibility with Cassandra by validating cluster operations. The post details various testing scenarios covered by dtest, such as batch operations, paging, TTL, commit logs, server-client notifications, snapshot and restore, and consistency levels. It highlights the importance of automated testing in uncovering bugs, citing examples of issues resolved through dtest, and emphasizes the significance of both unit tests and dtest in ScyllaDB's development process. Additionally, it outlines the plan to add more tests for upgrades and migrations, and discusses the contribution of these tests to the original dtest project, underscoring the collaborative efforts in enhancing ScyllaDB's reliability and functionality.

The MIKELANGELO project, a three-year initiative funded by the European Commission's Horizon 2020 program, aims to enhance cloud computing by supporting high-performance computing (HPC) and I/O-intensive applications. During a recent meeting in Brussels, the project's progress was lauded, with several technical innovations highlighted. ScyllaDB showcased Seastar, a C++ framework that significantly boosts the performance of server applications, and OSv, a Linux-compatible kernel optimized for cloud VMs, which was used for aerodynamic simulations by Pipistrel. Other partners, including Huawei, IBM, and Intel, presented advancements such as RDMA virtualization, I/O bandwidth improvements, and a telemetry framework for cloud measurement analysis. The project's collaborative, open-source approach promises to yield a more efficient, secure, and high-performing cloud infrastructure in the future.

Nadav Har’El, a developer at ScyllaDB, will deliver a talk on March 15 at the Generalist Engineer meetup in Tel Aviv, focusing on modern server application development and performance optimization. With a diverse 20-year programming background, Har’El will discuss the limitations of traditional operating system APIs and programming practices on multi-core servers, emphasizing the need for new design trends like event-driven programming and fast user-space packet processing. He will introduce Seastar, an open-source C++ framework that enhances server application performance by utilizing a single thread per core and data sharding, enabling complex applications through futures and continuations. The talk will highlight how applications like Memcached and Cassandra have been rewritten using Seastar, achieving up to ten times better performance than their original versions. Additionally, Har’El will present ScyllaDB, a re-implementation of Cassandra in C++ and Seastar, offering significantly improved throughput and latency.