Kafka is a distributed publish-subscribe system for processing large amounts of streaming data.
Kafka provides an extremely high throughput distributed publish/subscribe messaging system. Additionally, it supports relatively long term persistence of messages to support a wide variety of consumers, partitioning of the message stream across servers and consumers, and functionality for loading data into Apache Hadoop for offline, batch processing.
Kafka was developed at LinkedIn to process the large amounts of events generated by that company's website and provide a common repository for many types of consumers to access and process those events. Kafka has been used in production at LinkedIn scale to handle dozens of types of events including page views, searches and social network activity. Kafka clusters at LinkedIn currently process more than two billion events per day.
Kafka fills the gap between messaging systems such as Apache ActiveMQ, which provide low latency message delivery but don't focus on throughput, and log processing systems such as Scribe and Flume, which do not provide adequate latency for our diverse set of consumers. Kafka can also be inserted into traditional log-processing systems, acting as an intermediate step before further processing. Kafka focuses relentlessly on performance and throughput by not introspecting into message content, nor indexing them on the broker. We also achieve high performance by depending on Java's sendFile/transferTo capabilities to minimize intermediate buffer copies and relying on the OS's pagecache to efficiently serve up message contents to consumers. Kafka is also designed to be scalable and it depends on Apache ZooKeeper for coordination amongst its producers, brokers and consumers.
Kafka is written in Scala. It was developed internally at LinkedIn to meet our particular use cases, but will be useful to many organizations facing a similar need to reliably process large amounts of streaming data. Therefore, we would like to share it the ASF and begin developing a community of developers and users within Apache.
Many organizations can benefit from a reliable stream processing system such as Kafka. While our use case of processing events from a very large website like LinkedIn has driven the design of Kafka, its uses are varied and we expect many new use cases to emerge. Kafka provides a natural bridge between near real-time event processing and offline batch processing and will appeal to many users.
Our intent with this incubator proposal is to start building a diverse developer community around Kafka following the Apache meritocracy model. Since Kafka was open sourced we have solicited contributions via the website and presentations given to user groups and technical audiences. We have had positive responses to these and have received several contributions and clients for other languages. We plan to continue this support for new contributors and work with those who contribute significantly to the project to make them committers.
Kafka is currently being used by developed by engineers within LinkedIn and used in production in that company. Additionally, we have active users in or have received contributions from a diverse set of companies including MediaSift, SocialTwist, Clearspring and Urban Airship. Recent public presentations of Kafka and its goals garnered much interest from potential contributors. We hope to extend our contributor base significantly and invite all those who are interested in building high-throughput distributed systems to participate. We have begun receiving contributions from outside of LinkedIn, including clients for several languages including Ruby, PHP, Clojure, .NET and Python.
To further this goal, we use GitHub issue tracking and branching facilities, as well as maintaining a public mailing list via Google Groups.
Kafka is currently being developed by four engineers at LinkedIn: Neha Narkhede, Jun Rao, Jakob Homan and Jay Kreps. Jun has experience within Apache as a Cassandra committer and PMC member. Neha has been an active contributor to several projects LinkedIn has open sourced, including Bobo, Sensei and Zoie. Jay has experience with open source software as the originator of the Project Voldemort project, as well as being active within the Hadoop ecosystem community. Jakob is an Apache Hadoop committer and PMC and previous Apache ZooKeeper contributor.
The ASF is the natural choice to host the Kafka project as its goal of encouraging community-driven open-source projects fits with our vision for Kafka. Additionally, many other projects with which we are familiar with and expect Kafka to integrate with, such as Apache Hadoop, Pig, ZooKeeper and log4j are hosted by the ASF and we will benefit and provide benefit by close proximity to them.
The core developers plan to work full time on the project. There is very little risk of Kafka being abandoned as it is a critical part of LinkedIn's internal infrastructure and is in production use.
All of the core developers have experience with open source development. LinkedIn open sourced Kafka several months ago and has been receiving contributions since. Jun is an Apache Cassandra committer and PMC member. Jay and Neha have been involved with several open source projects released by LinkedIn. Jakob has been actively involved with the ASF as a full-time Hadoop committer and PMC member.
The current core developers are all from LinkedIn. However, we hope to establish a developer community that includes contributors from several corporations and we actively encouraging new contributors via the mailing lists and public presentations of Kafka.
Currently, the developers are paid to do work on Kafka. However, once the project has a community built around it, we expect to get committers, developers and community from outside the current core developers. However, because LinkedIn relies on Kafka internally, the reliance on salaried developers is unlikely to change.
Kafka is deeply integrated with Apache products. Kafka uses Apache ZooKeeper to coordinate its state amongst the brokers, consumers, and soon, the producers. Kafka provides input formats to allow Hadoop MapReduce to load data directly from Kafka. Kafka provides an appender to allow consuming data directly from Apache log4j.
While we respect the reputation of the Apache brand and have no doubts that it will attract contributors and users, our interest is primarily to give Kafka a solid home as an open source project following an established development model. We have also given reasons in the Rationale and Alignment sections.
Information about Kafka can be found at \[http://sna-projects.com/kafka/\] The following links provide more information about the project:
Kafka roadmap and goals: \[http://sna-projects.com/kafka/projects.php\]
The [GitHub] site: \[https://github.com/kafka-dev/kafka\]
Kafka overview from Jay Kreps: \[http://www.slideshare.net/ydn/hug-january-2011-kafka-presentation\]
Kafka overview from Jakob Homan: \[http://bit.ly/fLmoZz\]
Kafka paper at NetDB 2011: \[http://research.microsoft.com/en-us/um/people/srikanth/netdb11/netdb11papers/netdb11-final12.pdf\] \\
Kafka has been under development at [LinkedIn] since November 2009. It was open sourced by [LinkedIn] in January 2011. It is currently hosted on github under the Apache license at \[https://github.com/kafka-dev/kafka\]
Kafka is mainly written in Scala with some performance testing code in Java. Several clients have been contributed in other languages, including Ruby, PHP, Clojure, .NET and Python. Its source tree is entirely self contained and relies of simple build tool (sbt) as its build system and dependency resolution mechanism.
The dependencies all have Apache compatible licenses.
JIRA Kafka (KAFKA)
The existing code already has unit tests, so we would like a Hudson instance to run them whenever a new patch is submitted. This can be added after project creation.
Chris Douglas (Apache Member)
We are requesting the Incubator to sponsor this project.