Life in the Valley
The creator of the network monitoring system Riemann, Kyle Kingsbury, has put together a comprehensive series of blog posts, on the fault-tolerance, high-availability, and general correctness of number of database and storage technologies. Of the technologies discussed I am most familiar with — elasticsearch and Apache Kafka — I found the posts to be a great read.
If you haven’t read them yet, you should check them out on his site.
Java is the predominant language of Big Data technologies. HBase, Lucene, elasticsearch, Cassandra – all are written in Java and, of course, run inside a Java Virtual Machine (JVM). There are some other important Big Data technologies, while not written in Java, also run inside a JVM. Examples include Apache Storm, which is written in Clojure, and Apache Kafka, which is written in Scala. This makes basic knowledge of the JVM quite important when it comes to deploying and operating Big Data technologies.
In my last blog post I explained why writing design documents is such a powerful approach to building well-engineered systems. But what should one document? When it comes to software, if one documents too much, the content of the documentation can become inaccurate very quickly, and inaccurate documentation is quickly ignored.
Many software engineers never write design documents. Design documentation takes time, and implementations often proceed so far without any documentation that if it happens, it’s an act of recording what has been done — a tedious task at the best times. Many software engineers argue “the code exists, it’s running, it’s working, let’s move on and build the next thing.”
My father worked for many years in QA at Beckman, an American medical instruments firm. His job was to ensure that newly-manufactured centrifuge rotors would hold up when spun at thousands of RPMs. He used to tell me that the Beckman philosophy could be summarised in one sentence — “There is no substitute for quality”.
I came across a very readable paper on distributed systems — Distributed systems for fun and profit. I recommend it for anyone interested in learning more about distributed systems, and the challenges involved with designing, building, and operating distributed systems.
Packt recently asked me to review their new publication Mastering ElasticSearch by Rafał Kuć and Marek Rogoziński. Since most of my experience with elasticsearch has been from a systems points of view — index management, cluster maintenance, indexing performance — I paid most attention to the chapters about those parts of elasticsearch.
AWS have posted the video online of Jim Nisbet’s and my talk at AWS:reinvent 2013. In it, Jim and I describe the system we built at Loggly, which uses Apache Kafka, Twitter Storm, and elasticseach, to build a high-performance log aggregation and analytics SaaS solution, running on AWS EC2.
This past week I had the opportunity to speak, with my colleague Jim Nisbet, at AWS re:Invent 2013. Titled “Unmeltable Infrastructure at Scale: Using Apache Kafka, Twitter Storm, and Elastic Search on AWS“, Jim and I described the architecture of Loggly’s next-generation log aggregation and analytics Infrastructure, which went live 3 months ago, and runs on AWS EC2.
Loggly recently held an elasticsearch meetup, which was a great success. One question that was repeatedly asked was how to ensure elasticsearch does not suffer a partition — known as a split-brain. This can be a particular problem in AWS EC2, where the network is subject to interruptions. It can also happen if the elasticsearch master node performs long garbage collection cycles.
One configuration that is very effective at preventing this problem is described in this post.
After 14 months of hard work, the next generation of Loggly has been released. It’s been a great time to be part of the Software Infrastructure team at Loggly and we have put together a superb log aggregation & real-time analytics platform.
As technical lead at Loggly, responsibility for a well-engineered infrastructure ends with me. And one way to ensure the system is designed and implemented well is to stay as close as possible to the code, ensuring that the team and I write quality software.
But it can be difficult to complete the design and implementation of the features I am responsible for, ensure that what the team produces is well-implemented, and understand every line of code — there is only so much time in the day.
When running a large real-time processing system, monitoring is critical. But it does more than allow you to keep an eye on your system. During development it allows you test hypotheses about how it works, how it performs when certain parameters are changed, and takes the guessing out of working with dynamic systems.