Guest blog by Pankaj Gupta from Citrix
KubeCon + CloudNativeCon San Diego 2019 was the place to take stock and ponder the progress being made with Kubernetes. It was also the place to learn the art of possibilities with Kubernetes – the new paradigm of building applications for faster release cycles, modularity, and portability. Unsurprisingly, it was being driven by the many Kubernetes Knights – who, clad in the armor plating of their service meshes, are charging forward with great speed and agility, lancing their way through everything we thought possible, scaling the towers and breaking down the walls of the castle as they press forward on their microservices crusade.
How far Knights have taken Kubernetes can be summed up by the following key data points from KubeCon + CloudNativeCon NA ’19.
160,000 pods scale in production @Ebay
Ebay now has a staggering 160,000 pods in production spread across more than 60 clusters. This is an enormous number of microservice instances. And they are not alone. Several enterprises report tens of thousands of nodes in their environments – Nordstrom (60k); Yahoo (36K); Lyft (25k). This sort of hyperscale deployment was once only the purview of the largest public cloud providers and now the enterprise is adopting it to deliver consumer services.
What a long way we’ve come in 5 short years of Kubernetes. It is truly inspirational to see that the promise of microservices as a highly scalable architecture is here today and growing strong. Imagine what scale will be possible next?
1 million new containers a day @Uber
One million new containers in batches are launched every day at Uber – sometimes at the rate of 1,000 per second. This is a mind-boggling number of new containers. To manage and run this number of containers is amazing but to start them en masse is breathtaking and perfectly illustrates the extreme ephemeral nature of microservices at scale.
To launch this many containers on demand shows that the Kubernetes has matured and delivers the agility and velocity that customers require for large production environments. Moreover, while the sheer scale of launching 1 million pods is amazing, doing it repetitively is priceless and doing it every day without breaking anything is truly a state of nirvana.
450 new pods in the blink of an eye @Uber
30 seconds is the time taken for Uber to launch 40,000 pods across 8,000 nodes with their optimization. That is ultrafast at scale. In the time it takes the human eye to blink (approx. 300 ms) it is possible to bring up 450 microservice instances.
Once again, this stretches the boundaries of what was thought possible only a short while ago for autoscale and highlights the real-world velocity to respond to change. Of course, when you are dealing with the ephemeral nature of microservices at this ultrascale you have to make sure that the rest of your infrastructure can cope with this sort of change. Can yours?
100,000 sidecars for service mesh @Lyft
Lyft has 100,000 sidecars in its service meshes. This is huge and shows service mesh has arrived at scale. Offering great observability, granular traffic management, and enhanced security, it is not surprising that service mesh has emerged as the most sought after microservices architecture over the last 18 months. This level of deployment shows that service mesh has moved out of the realm of the aspirational into reality. It is not just real, but real at scale running production grade services.
11,000 Walmart stores – Kubernetes moving to edge
This is the number of stores that Walmart will take Kubernetes to in the near future. The world’s largest retailer is shifting their microservices and service mesh to the edge, closer to the point of sale, to improve the customer experience. This bold shift to the edge is the biggest indicator that Kubernetes is mature enough to leave the data center and becoming hugely distributed. Next time you visit Walmart, look for containers in aisle K8s.
F-16 fighter jet soaring high with Kubernetes
The US Department of Defense has deployed microservices into the F-16 Fighter, one of the most advanced war planes in human history. Kubernetes has helped them to evolve their legacy, manual software processes and enabled them to deploy software faster and more reliably to bring new capabilities to the pilots quickly. Perhaps F-16s with Kubernetes should be called K-16s?
These data points highlight the power of Kubernetes and the velocity of innovation it can bring. The outstanding scalability that Kubernetes offers has to be matched by the accompanying infrastructure. Inspired by these scales, Citrix has recently tested that it’s possible to keep pace with the change events associated with a 50,000 pods creation across 1,000 nodes with a single instance of Citrix ADC VPX proxy to deliver applications. The tests illustrated that the Citrix ADC comfortably outpaced the speed of deployment and had a lot of spare capacity for growth. In reality, customers will use multiple instances of proxies like Citrix ADC to manage and scale their workloads to the extents discussed above, but Citrix just wanted to push limits to the extreme with a single proxy instance.
The Kubernetes Knights have shown us what’s possible today and, in the spirit of chivalry are generously giving back to the community as they open source many of their advances. But, how far can Kubernetes go? Where will the next Kubernetes deployment be – an interplanetary spacecraft? Nobody really knows, but wait for KubeCon 2020.