Kubernetes ensures its scalability by relying on Fastly

The challenge

In just 10 years, Kubernetes has become the leading container orchestration technology for cloud environments. For Kubernetes maintainers, that growth is reflected in demands on the Kubernetes website and other important services like binary downloads. In 2023, the Kubernetes binaries site, dl.k8s.io, served more than 5 petabytes of binaries each month. To ensure speed, reliability, and global delivery at this scale, the Kubernetes infrastructure special interest group (K8s Infra SIG) decided to leverage a content delivery network (CDN).

The solution

The special interest group, including K8s Infra SIG co-chair and technical lead Arnaud Meukam, evaluated several CDN vendors before choosing Fastly. "We noticed that many open source organizations were moving to Fastly," Meukam said. "Fastly has a track record of supporting open source projects through its Fast Forward program, and it has a proven record of delivering fast and secure digital experiences."

Keeping Kubernetes sustainable as it scales

Kubernetes is well known for the strength of its community, but as with most open source projects, contributors' desire to help significantly exceeds the time they have available. Ensuring the sustainability of an open source project like Kubernetes requires making the most of limited resources.

What kind of scale is Kubernetes looking at? In 2022, Kubernetes was the primary container orchestration tool for 71% of Fortune 100 companies, and a Gartner report projected that by 2027, more than 90% of global organizations will run containerized applications. Here's how that looks on the Kubernetes website: In 2024, the site averaged 90 million requests per month, using 1 PB of bandwidth and generating 1.2 billion logs.

The Kubernetes team isn't worried. "We are confident that Fastly will help us maintain the sustainability of Kubernetes. We don't have to worry about scalability," Meukam said.

Gaining visibility to optimize delivery

Before making the switch to Fastly, Kubernetes infrastructure was maintained inside Google infrastructure and served via a Google Cloud Services (GCS) bucket. Because many Kubernetes maintainers don't work for Google, they lacked visibility into bandwidth or other metrics. Moving dl.k8s.io to Fastly gave the team much-needed observability into traffic on one of the most heavily used areas of the Kubernetes project.

The Kubernetes team now has the visibility to see who is pulling binaries, where traffic is coming from, and how the origin is behaving. As a result, they can ensure reliable delivery at any level of traffic and spot issues quickly to prevent user problems. "Origin Inspector helps us track our bandwidth between Fastly and GCS and ensure that our egress costs remain very low," Meukam said. "We recently began using Alerts, which helps us identify any issue coming from the origin and escalate an issue quickly."

Ensuring a seamless user experience globally

Cutovers can be a stressful process at massive scale, but fortunately for its many users, the Kubernetes project’s switch to Fastly was seamless. "The cutover to Fastly was a nonevent for our users, which was the goal," Meukam said. "We want to be able to make impactful changes to our infrastructure without users needing to know or care about it."

With the cutover complete, Kubernetes binaries are now closer to their users. Moving from one GCS bucket to multiple points of presence using Fastly’s network helps Kubernetes maintain service even for users in areas with lower quality internet connections. "Being closer to users around the world was a big factor for us in adopting a CDN, and Fastly's global network is a big factor in why we chose Fastly specifically," Meukam said.

Key takeaway

Moving to the Fastly CDN has given the Kubernetes project the scalability and reliability needed to sustain this vital piece of infrastructure. Using the full suite of Fastly observability tools helps the Kubernetes infrastructure team optimize traffic, keep egress costs low, and prevent interruptions for users.

"It's been amazing to work with Fastly," Meukam said. "Any issue we had was quickly escalated and solved, and if there was any confusion about documentation for the configuration, it was handled very nicely by the support team. We are really happy about our collaboration with Fastly."