November 2014 Summaries

The Cassandra Cluster Manager (CCM) is a command line tool that enables the creation of Apache Cassandra clusters on localhost for development purposes. CCM 2.0, released in October, added support for DataStax Enterprise and running on Windows with Cassandra 2.1. It can be installed via PyPI or from source. The latest version provides a convenient Python API and allows users to set up Analytics datacenters, Cassandra datacenters, and Solr datacenters on their laptops. CCM is not intended for production use or performance testing; issues should be reported on GitHub.

DateTieredCompactionStrategy (DTCS) is a new compaction strategy introduced in Cassandra 2.0.11, designed for time series-like workloads. It aims to keep data written at the same time in the same SSTables to improve performance. DTCS was contributed by Björn Hegerfors at Spotify and is specifically tailored for handling time series data where data is mostly appended to existing partitions. The main purpose of compaction is to reclaim disk space used and to make sure that we can serve reads from as few SSTables as possible. DTCS groups SSTables in windows based on how old the data is in the SSTable, ensuring that new and old data are not mixed during compaction. The size of the compaction windows is configurable, with options like base_time_seconds and max_sstable_age_days to control the initial window size and when to stop compacting data, respectively. This strategy can greatly reduce the number of SSTables touched during a read for queries that request recent data, improving overall performance.

Hinted handoff (HH) is an integral part of Cassandra's write path that helps reduce inconsistency caused by temporary node unavailability periods. In earlier versions, hints were stored in a regular Cassandra table called system.hints. However, this approach had some drawbacks, including the use of queues and potential tombstone issues. To address these issues, starting with Cassandra 3.0, hints are stored in flat files, bypassing the storage engine altogether. This new implementation simplifies the replay process and reduces overall overhead surrounding hints.