July 2016 Summaries

The blog post explores the integration of KairosDB with ScyllaDB, highlighting how KairosDB, a time series database built on Cassandra, can be redirected to work with ScyllaDB by utilizing the Thrift API. The process involves configuring ScyllaDB to manage CPU and memory usage alongside KairosDB, and using a Bash script to populate the database with time series data such as story views. The integration allows the use of KairosDB's API to query and manipulate data stored in ScyllaDB, including querying recent data points and deleting metrics. The post also mentions the ScyllaDB Summit, where Thrift and other topics related to ScyllaDB will be discussed, offering insights into new features and use cases from companies like IBM and Hulu, as well as an opportunity to meet developers and users of ScyllaDB.

ScyllaDB 1.3 introduces native support for the Thrift API to accommodate legacy systems still reliant on this original Cassandra protocol, despite the shift towards the more widely used CQL query language. Thrift, initially developed at Facebook and now an Apache project, remains integral to several key integrations such as KairosDB and Presto, although newer integrations like Apache Spark exclusively use CQL. ScyllaDB's implementation allows Thrift operations to be conducted on CQL tables, ensuring backward compatibility without requiring users to update their applications. The performance of Thrift is comparable to CQL, with benchmark tests showing similar rates and latencies across different workloads. However, there are limitations, such as the lack of support for super columns and mixed column families. The upcoming ScyllaDB Summit will further explore Thrift and other ScyllaDB advancements, offering insights into how companies like IBM and Hulu leverage the database for enhanced performance.

ScyllaDB version 1.2.2, released in July 2016, is a bugfix update for the 1.2 stable branch, maintaining backward compatibility and supporting rolling upgrades. The release addresses several issues, including a problem with nested types that led to invalid types when queried, a race condition in Seastar causing crashes, memory over-allocation errors, and the creation of unnecessary table directories upon restart. The update also fixes a regression that slowed down performance on multi-core machines and resolves high read backlog issues leading to out-of-memory errors. Contributions from Avi Kivity and others include various bug fixes and performance improvements, such as better handling of user-defined type metadata and more conservative memory allocation for queued reads.