I spent a nice Friday on this open source conference so here are my impressions.
The first presentation worth mentioning is the one about the ELK stack given by Rastislav Rehak. He described the roles of “E”, “L” and “K” in the whole architecture and mentioned what’s new in the version 5.x. What was new to me was the concept of small agents called beats. It represents a small unit that is capable of sending events to Elasticsearch or to Logstash for further processing. There are also different kinds of the beat, such as: Metricbeat, Topbeat (basic utilization metrics based on the “top” command), Packetbeat for network, etc.
Probably the best talk was given by Jorge Hidalgo and its name was “Lightweight Java Architecture”. Although, the name sounds little bit oxymoronic, the topic was awesome. He was able to build three micro services running in Docker Swarm and communicating with each other. Spring Boot was used as a micro container, Hystrix for circuit-breaker, Eureka for the service discoverability and health checks and for configuration he used some git-based tool from Netflix (I forgot the name). Here is the demo he was showing.
I attended also a presentation about OpenCV on mobile platforms, because I was playing with it some time ago on Android. Guys from CSOB (Czech bank) were sharing their issues and best practices when working on a real time augmentation of the data from phone camera. At some point they were applying the OCR on the text, so the image had to be flat, therefore they had to do some linear transformation. OCR + the transformation + drawing stuff is expensive, so they put a classifier in front of the whole pipeline that recognized the well known invoices so that it was done only if looking at some invoice.
Presentation from Wojciech Furmankiewicz about Cloudforms was quite good for sys admins who work with some virtualization tool and want to automate things or just see the bigger picture. I didn’t know about some hidden features like the possibility to assign a VM to some user for some given time, with all the email magic happening behind the scenes. Also the ability of CF to suggest better resource utilization based on the historical data was new to me. For instance: hey, this machine has 8 cores, but it’s hardly ever fully utilized, while this machine has only 1 core, but there is a lot of processes running on it.
Another interesting topic was the Spark Streaming given by Martin Culen. Although he described the basic ideas, unfortunately, there was only a little about Spark itself. The majority of the presentation was about performance tuning of the whole system, including HDFS and Kafka nodes.
Last talk, I was able to attend, was called “Microservices, More Than Just a Small Codebase” and was presented by Philipp Fehre from Microsoft. He described their path when splitting the monolithic application into smaller micro services and all the problems they hit. This was really good to hear, because nowadays everyone want to use the microservices for everything without actually realizing that some parts shouldn’t be split apart.
In the end, I regretted, that I haven’t submitted any topic, so perhaps next year.