KubeCon North America – Seattle
I made the trip up to Seattle for KubeCon North America at the end of 2018 along with a bunch of us from Sumo Logic. KubeCon is a conference that specializes in all things Kubernetes and focuses on updating the world on the state of the Kubernetes ecosystem. This year’s event was massive with 8,000 attendees, and talks given by representatives from Amazon Web Services (AWS), Google Cloud Platform (GCP), and Microsoft Azure to name a few big wigs that were there. Every day the keynote presentations showed us how much Kubernetes had grown in the last year, and the numbers seemed to back it up.
While there was a plethora of topics discussed, I walked away from the conference surprised at the importance of Prometheus for monitoring a Kubernetes stack. Some of the other trends that stood out at KubeCon focused on various ways that users monitor their Kubernetes stack, including end-to-end observability, data ingest volume and the rise of tracing. I’ll share a brief recap on these areas in this article.
How Do Users Monitor Their Kubernetes Stack?
There was a consistent theme across all the talks given on observability at KubeCon: We need better observability of our Kubernetes stacks! Kubernetes is complex. It’s has a level of complexity that makes the day two challenge extremely difficult, especially without the proper observability tooling. The founder of LightStep, Ben Sigelman, gave a really impactful talk on the “Three Pillars of Observability,” which are metrics, logs and tracing, and how each of these are individually flawed. With Kubernetes, you now need a solution that can actually handle the weaknesses of the individual data streams, bring out the benefits, and stitch the benefits together to create a comprehensive story. This is something that no observability market vendor has been able to do successfully yet.
Solving the Data Cost Challenge
Let’s say you’ve been able to ingest the data. Then there is the question of cost. With more data streams, and more ephemeral data streams, you have a huge rise in data costs. Managing the cost of monitoring Kubernetes becomes almost as important as being able to capture all the data in the first place. All the observability talks helped highlight this unfulfilled challenge from existing vendors.
Tracing is the Present, and the Future
Furthermore, while we’ve been talking about metrics and logs for a while, the distributed complex control plane of Kubernetes has really magnified the need for distributed tracing as a way of stitching together transactional understanding across the stack. When paired with service meshes, tracing really yielded some impressive observability into the stack. For example, Lyft shared in a presentation how they paired Envoy with LightStep’s tracing setup to get an understanding of each transaction through their edge nodes through to the actual application-layer microservice to microservice interactions. It helped them understand where their latency was coming from and better manage the inter-microservice network communication. If KubeCon is any indication, tracing will become ever more important for Kubernetes.
Prometheus is King
Finally, what was 100 percent apparent from KubeCon was that Prometheus is the biggest winner in the monitoring market for Kubernetes. Not only was Prometheus the most used monitoring tool through the observability talks, it was also the most cross-integrated product. From WeaveWorks to Grafana Labs building integrations on top of Prometheus, to service meshes like Istio, Tetrate.io, and Envoy piping data directly into Prometheus by default, Prometheus is natively available, natively connected and natively scalable on Kubernetes. This raises the bar for what SaaS vendors need to do to really compete in the Kubernetes monitoring space.
KubeCon North America 2018 really highlighted the growth of Kubernetes in capturing the mindshare of site reliability engineers (SREs) worldwide. When it comes to monitoring Kubernetes, users’ needs are quickly becoming far more complex, and cross-cutting than ever before. Monitoring tools need to compete with an OSS solution in Prometheus that has taken the lion’s share of Kubernetes users.
In 2019, SaaS offerings hoping to challenge Prometheus for the monitoring Kubernetes throne will need to find ways to make it easier to integrate multiple streams of data (metrics, logs and traces), and cheaper to manage that data. Given how quickly Kubernetes has grown, I’m sure 2019 will be an exciting year of innovation on monitoring, operationalizing, and developing on top of Kubernetes.