FOSDEM 2017 and the future or your infrastructure
A little less than couple of weeks ago I attended FOSDEM 2017 in Brussels, Belgium. Once again it was a great event packed with high quality content. I attended tracks about configuration management (Puppet, Ansible and friends), software defined networking and network function virtualization and containers/microservices.
Maybe it was only due to my selection of tracks (there are quite a few: https://fosdem.org/2017/schedule/tracks/) but it seemed that a lot of the talks in some way or other touched on either Kubernetes, OpenStack or both. This is quite understandable as Kubernetes is one the most hyped technologies in IT right now. OpenStack, on the other hand has been around for around 7 years now, whereas Kubernetes is only a couple of years old as an open source project. Both Kubernetes and OpenStack seem the projects that a lot of the innovation in open source infrastructure is happening right now. So, if I had to bet on a infrastructure technology for the next few years right now, I’d choose either of these depending a bit on the use case. Or you could do like the people in CERN do and use both (https://fosdem.org/2017/schedule/event/magnumcern/) and use OpenStack to enable deployment of Kubernetes/Mesos/Docker Swarm clusters.
If I were to pick a single talk from Saturday that I wanted to highlight, it would definetely be James Shubin’s talk on the next generation config management system mgmt that he has built. Unfortunately, the A/V system was a bit messed up during the Saturday morning in the configuration management devroom so there’s no recording available, but check out the recording of his talk (linked on the page) from last year’s DebConf. It’s mostly the same, although he had some new, cool demos to show in FOSDEM which were not present in the DebConf talk.
I also attended a few talks on the Software Defined Networking/Network Function Virtualization devroom in the afternoon. Some of the talks were pretty hard to follow as there’s a lot of new technology and components that just don’t exist in more traditional networking setups (see https://wiki.opnfv.org/display/fds/OpenStack-ODL-VPP+integration+design+and+architecture for example). I probably should’ve done a bit more background research than I did. Anyway, it is a field where a lot of innovation is happening and definitely something we should take a closer look when our own datacenter grows.
The containers and microservices devroom that I spent my Sunday in was mostly of fairly high quality talks. Some worth mentioning were Ray Tsang’s talk on gRPC, a RPC-framework made by Google and recently open sourced. It seemed pretty useful tool for interprocess communication if you want to do it without involving a message queue. Then there was a talk by CoreOS’s Josh Wood about an upcoming Kubernetes feature called “operators”. The idea of operators is to create a framework for creating so called “controllers” that know how to handle application-specific things when e.g. scaling out or handling failure. They are mostly needed for applications that are natively not that fit for running in containers as they handle some kind of data and internal state. So far, CoreOS has implemented operators for etcd key-value store and the Prometheus monitoring system.
There was also a nice talk from Chris Down from Facebook on how Facebook uses the Linux kernel feature cgroups (that is used by basically all the container technologies as well) instead of containers to manage applications in their infrastructure. Due their massive scale, they mostly run a single primary application on a single physical server. However, cgroups are used to manage augmenting services (configuration management, monitoring etc.) running on the same server so that they do not consume too much resources. Apparently, cgroups is such an important piece of technology for Facebook that they currently employ some of the core developers for the cgroups Linux subsystem. The talk was mostly about upcoming breaking change in the cgroups API (from cgroups v1 to v2).
I ended my FOSDEM by listening to the ending keynote about running Linux in Air Traffic Control systems. It was sadly lacking in technical detail (how’s the architecture like etc) but it was still interesting to hear that Linux is used to power such critical systems with high safety requirements.
I also attended a pre-FOSDEM event FOSDEM PgDay organised by the good folks at PostgreSQL Europe (https://www.postgresql.eu/). Unfortunately, there are no recordings of the talks but at least check the slides for Ilya Kosmodemiansky’s talks of PostgreSQL worst practices (http://www.slideshare.net/PostgreSQL-Consulting/postgresql-worst-practices-version-fosdem-pgday-2017-by-ilya-kosmodemiansky).
Overall, FOSDEM is a pretty great event with huge amount of content. As the event itself is free, it is also fairly cheap to attend. If you are interested in the latest and greatest in open source technology (and you should, you are working for a company that uses a lot of it), I suggest you to consider attending it next year. Talk to your manager. :)