Insights from KubeCon+CloudNativeCon 2017: Kubernetes as the Linux of the Cloud

On December 6, despite an unusually cold Austin, the Austin Convention Center was bustling with the third KubeCon+CloudNativeCon, attracting more than 4,000 participants. Liulishuo’s infra team, having run Kubernetes in production for nearly a year and using many CNCF projects, shares their perspective on why this event is a must‑attend for every cloud‑era engineer.

Kubernetes is the Linux of the Cloud

CNCF President Dan Kohn quoted Jim Zemlin to illustrate Kubernetes' rapid growth in code size, community activity, and search popularity, noting that Kubernetes now even has a certification program. The expanding number of CNCF projects, members, and conference attendees further confirms this trend.

For engineers, this raises the question: if Kubernetes truly becomes the Linux of the cloud, does every cloud developer need to master it, and will development become simpler and more efficient as Kubernetes becomes more powerful and user‑friendly?

Many in the community have pondered this. In the first day’s keynote, Brendan Burns suggested that building distributed systems is still too complex and proposed treating Kubernetes as a runtime with containers as objects, experimenting with the Metaparticle project to express containers directly in code. The following day, Tim Hockin and Michael Rubin discussed how the Kubernetes project could evolve similarly to the Linux kernel and distro split, aiming for stability amid rapid updates and exploring whether Kubernetes can be modularized to improve user experience.

Is Multi‑Cloud the Future?

Kubernetes has undeniably made multi‑cloud deployments easier and reduced migration costs. Cloud providers are unlikely to monopolize users, so many offer managed Kubernetes services to retain customers, while companies build their own enterprise‑grade K8s solutions. However, achieving true multi‑cloud requires handling differing networks, storage, and APIs, which depends on a variety of tools and ongoing exploration.

Deploying Kubernetes across different clouds remains challenging. Although managed services exist, self‑managed clusters are sometimes unavoidable. Kris Nova and Robert Bailey presented the emerging Cluster Management API, which lets users declaratively define clusters as Kubernetes resources, allowing the platform to provision the underlying infrastructure automatically.

CoreOS’s Brian Redbeard emphasized operational best practices beyond tooling, such as avoiding overlapping network ranges, recommended logging and monitoring solutions, and strategies for robust Kubernetes deployments.

Service Mesh

The hottest topic at the conference was Service Mesh. With Kubernetes already the standard for container orchestration, Service Mesh addresses the inherent complexity of distributed systems, especially in diagnostics and recovery. It abstracts language constraints, offering advanced routing, circuit breaking, flow control, distributed tracing, and fine‑grained security, making the development and debugging of stateless distributed services as straightforward as single‑machine applications.

Data Storage

Storage‑related sessions were also popular. Running databases on Kubernetes (DB on K8s) and reliable storage orchestration remain challenging, especially for cross‑cluster, cross‑zone disaster recovery.

eBay shared their geo‑distributed database solution, combining local disks for I/O performance with a service layer for sharding, and using local and network volumes for backup. Their architecture includes local and global coordinators for multi‑zone coordination. Oracle presented “Running MySQL on Kubernetes,” discussing various deployment options and their yet‑unreleased MySQL Operator.

Google’s session on Local Ephemeral Storage covered request, allocation, and eviction mechanisms at pod, node, and namespace levels for use cases like logs and runtime writes. A presenter from Hitachi introduced Raw Block Volume, a bare‑metal‑style volume for scenarios requiring direct block access, with detailed slides and implementation insights.

Other storage sessions highlighted the anticipation of faster AWS EBS mounting and broader MySQL‑on‑K8s adoption.

China’s Growing Influence

Even before the conference, several Chinese sponsors were visible, and the event featured many “China elements.” Chinese companies, including Alibaba Cloud, played significant roles as CNCF members and certified partners. The next KubeCon+CloudNativeCon is slated for Shanghai, Liulishuo’s headquarters, underscoring China’s rising prominence.

Editor’s Note

As 2017 ends, Liulishuo thanks its readers for their support and hopes the shared articles have been inspiring. The team has moved to a new headquarters with extensive amenities and has opened its 2018 hiring plan, inviting engineers of all specialties to apply via [email protected].

We look forward to collaborating with you in 2018!