For the first time CloudFire is crossing the ocean to participate in KubeCon North America. To make you part of it, as at the previous event in Valencia, I'm going to share with you my impressions about the highlights of the most important conference for the open source and cloud native community that I attended in late October.
Cloud Native Foundation puts its focal point on Cloud Native and chooses the city of Detroit, which is in the midst of an economic revival after a less-than-rosy period.
As per usual, the organisation was impeccable. It allowed everyone both to follow the extremely valuable technical talks and to engage with engineers from companies of different sizes from all over the world.
In addition to the canonical talks related to the development of the projects that make up the CNCF landscape, the main topics during this three-day event were Network, Stateful and Storage, Platform Engineering, Multi-Tenancy and optimisations related to the Kubernetes world and Cloud Native projects.
Network and Kubernetes
As many experts in the field know, the substrate on which Cloud Native projects are developed cannot be separated from the network. From this perspective, the advancement made by the implementation of eBPF by the major projects has been remarkable. There were also many talks regarding the usage of eBPF to track connections from and to PODs, granular policy on network access and usage, and more.
One of the most interesting talks was undoubtedly Five Ways With a CNI: Understanding Kubernetes Networking For Performance-Intensive Workloads, given by Stig Telfer of StackHPC Ltd and Erez Cohen of Nvidia. The focus of this talk was an overview of the current state of major CNIs from a performance perspective.
The point I'd like to focus on concerns the technologies that are currently being used and that appear to be mature enough to allow HPC processing to be performed without perceiving the network as a system bottleneck. The functionalities provided by the different CNIs make it in fact possible to meet even the most stiff security specifications, whether they are configured directly by the CNI or make use of integrative designs.
Stateful and Storage
Another particulary relevant topic during KubeCon NA was the persistency of data on Kubernetes. Such data is exposed using Statefulset and related. While the vast majority of applications running on Kubernetes are currently ephemeral (meaning the data they contain doesn't need to be maintained) the need to use the reconciliation engine for those PODs whose data is vital is beginning to feel crucial like never before.
Despite projects within CNCF Landscape are currently growing in number, the answer from the community isn't clear yet.
Regarding the storage, the discussion is still open and all to be explored. We shall see what the next steps are going to be.
Platform Engineering
Another popular topic at the Detroit event was the evolution of the concept of DevOps.
In fact, during KubeCon, many admitted how much the DevOps approach has been the focal point of development in recent years. Implementing it, however, shows some limitations. More specifically, the ability to find comprehensive personnel with cross-functional skills, prepared not only to work on the entire infrastructural stack, but also able to develop and work with code, has ultimately been proven to be rather difficult. Having a vision that is as broad as it is required is not easy at all. Having hyperspecialised figures also requires as much effort as the latter with regard to the communicational part.
For this reason, the figure of the platform engineer, as well as more serverless approach are emerging as alternative and complementary avenues to the DevOps figure.
In fact, this new proposal allows developers to specialise and to focus on everything related to the Dev world. This leaves what's related to operations and infrastructure to the Platform Engineer. With a serverless approach, the figure of the developer is transformed: they no longer need to be aware of the operations that ensure that their application works. Instead, they can simply launch it and get the cloud resources it needs dynamically.
Multi Tenancy
By simply wondering around booths of KubeCon NA 2022 and seeing how many talks about it are scheduled, we can safely say that Multi tenancy is this edition's focal point.
CloudFire is working with Clastix's team on this very issue. It is clear how much the entire community is looking for a solution that allows for centralised management of multiple Kubernetes clusters with the lowest possible effort.
In this sense, the Capsule and Kamaji projects are certainly the most interesting ones and have attracted the curiosity of many industry bigwigs. Now the question must be: how many cluster kubernetes should one make? One cluster? One per department? And in the case of cloud providers such as Cloudfire, how can they provide a managed Kubernetes service? The technology is now mature enough to allow the management of an incredible number of clusters with automation and in self-service mode.
Conclusions
As per usual, at the end of the event we feel overwhelmed with ideas and with a wealth of knowledge expanded, as well as enriched with numerous new contacts with whom I have had the pleasure of talking and exchanging views.
The event definitely earned a 10/10, both in terms of organisation and attendees. It allowed CloudFire to get an even clearer view of what the future of the Cloud is going to be: Cloud Native!
See you in 2023 with more events like this! ☁️💙
