June 2015 Summaries

Telegraf is an open source metrics collector that writes data to InfluxDB, with support for various services such as system, Docker, Redis, MySQL, and PostgreSQL, and plans to add more features and plugins in the future.

InfluxDB version 0.9.0 has been released with developer and production support, marking a significant milestone in its development. The release is built on over seven months of effort from the development team and hundreds of community feedback, promising a stable and incremental improvement process without breaking API changes or data storage modifications. Stability improvements have been made, including clustering implementation that's considered an alpha release with limitations, but future point releases will lift these constraints. Developer support plans are now available, offering quick answers to questions during the development phase for $350, while production support tiers offer different SLAs. Future releases will follow a regular cadence of three weeks each, with drop-in replacements and no breaking changes to the API or data storage. The goal is to reach maturity on features and testing before moving to version 1.0.

InfluxDB v0.9.0 release update 3` The development team has reworked the clustering and write path of InfluxDB, marking this as a major step forward in the 0.9.0 release, with significant improvements to stability and performance. This release introduces a new line protocol for writing data, which is vastly more performant than the previous JSON protocol. The changes will require users to blow away old data and config files, but future updates can be made without requiring migrations. A single server version of 0.9.0 with clustering marked as alpha will be released, followed by a stable release with regular point releases every three weeks.

The InfluxDB clustering design combines elements of both CP (Consistency) and AP (Availability) systems to achieve a highly available and eventually consistent system for handling time series data. The primary write path uses an AP system, favoring throughput over strong consistency, while the cluster metadata is stored in a CP system using Raft consensus. Writes are handled by leveraging the fact that time series data is almost always new immutable data, allowing for a simpler system that achieves goals for scalability and maximum throughput by relaxing constraints not necessary for the use case of time series data. The design also incorporates mechanisms such as hinted handoff, anti-entropy repair, and conflict resolution to ensure eventual consistency across the cluster.