January 2013 Summaries

Apache Cassandra uses the Yammer Metrics library for tracking client request metrics since version 1.1. The library offers tools like Counters, Gauges, and Timers with an easy-to-use API. Metrics can be exported to files or other systems periodically using official "metrics reporters" or community-developed ones. In this article, the author demonstrates how to configure Cassandra to report metrics to Graphite using a java agent. New metrics provide more information than previous versions and are easily integrated with other systems.

CCM (Cassandra Cluster Manager) is a tool that creates multi-node Cassandra clusters on the local machine. It's useful for setting up clusters quickly for development and testing, and serves as the foundation for cassandra distributed tests (dtests). To install CCM, you need to have the cql and PyYAML PyPI packages installed. Once installed, you can create a cluster by specifying the Cassandra version, number of nodes, and the name of the cluster. CCM will download the specified version of Cassandra and customize .yaml files for each node. Additional features include adding new nodes to a cluster, switching between clusters, displaying nodetool ring info or showing cassandra logs for individual nodes, and upgrading nodes in a rolling fashion. The full documentation on CCM can be found by running "ccm" to see a list of commands, and "ccm <command> -h" for information on an individual command.

The text discusses a comparison of distributed databases, specifically focusing on Cassandra's performance advantages over other NoSQL systems like MongoDB, Riak, and HBase. It highlights Cassandra's adoption of several architectural best practices that contribute to its superior scalability and throughput. These include being fully distributed with every machine handling a proportionate share of system activities, using a log-structured storage engine for efficient I/O operations on both hard disks (HDD) and solid-state drives (SSD), and having tight integration with its storage engine to support advanced concurrent structures and row-level isolation without locking. The text also mentions the importance of implementation details in performance optimization, such as Cassandra's ability to append updated data without rewriting or re-reading existing data, which ensures that updates to a Cassandra row or partition remain fast as the dataset grows.

To get assistance for issues related to DataStax products like DSE, the first step is to search Google using specific keywords or phrases related to the problem. Stack Overflow is another useful resource where you can find answers by using tags such as Cassandra. Planet Cassandra provides information and help specifically for Apache Cassandra. The official DataStax website offers documentation, tutorials, and support forums monitored by DataStax employees who are ready to answer questions. Community meetups and conferences provide opportunities to connect with like-minded individuals and domain experts. IRC channels such as #cassandra and #cassandra-dev can also be helpful for general discussions or development-related inquiries. Paid services by DataStax, including support and consulting, are available for more comprehensive assistance. When asking questions on various platforms, providing relevant information about the issue will help others provide better solutions.

The text announces the release of Cassandra 1.2.0, an update with key improvements that will make it a good year to be a Cassandra user. Additionally, DataStax documentation for this version is now available.

The text discusses the process of upgrading an existing single token per node architecture to vnodes without any downtime in a Cassandra cluster. It explains that enabling vnodes involves setting the num_tokens parameter and restarting the node, which splits the current range into the specified number of tokens while maintaining contiguousness. To realize the benefits of vnodes, the 'cassandra-shuffle' utility is used to randomly redistribute the vnodes amongst the cluster. The upgrade procedure can be tested on a per-datacenter basis and may require adjustments in workflow, such as using nodetool status instead of nodetool ring and removehost instead of removetoken for node removal. Despite some variance in the 'owns' percentage, this is normal and expected with vnodes.