Kubernetes use has surged in recent years. While there has been much consolidation around Kubernetes for container orchestration (with some outliers), companies still adopt many different technologies to address Kubernetes multi-cluster management.
The Cloud Native Computing Foundation (CNCF) has released a Technology Radar report that explains the prevailing attitudes regarding Kubernetes multi-cluster management tools and strategies. Below, we’ll highlight results from the report and use these insights to gauge the overall state of multi-cluster management habits.
Cluster Deployment Tech
CNCF End User Technology Radar is “an opinionated guide to a set of emerging technologies.” To evaluate new trends, CNCF polls 140 companies within its community on their technology preferences. This particular study asked respondents to evaluate Kubernetes multi-cluster tools on a scale that gauges their state from ‘assess’ to ‘trial’ to ‘adopt.’
As you can see, in terms of cluster deployment, there is not one single toolset to rule them all. Instead, there appears to be equal adoption of custom in-house tools, private cloud managed K8s, public cloud managed K8s and HashiCorp Terraform—all of these strategies landed in the ‘adopt’ category.
“Compared with other Radars, we had more tools than ever in the ‘adopt’ category,” says Gabe Jackson, cloud platform tech lead at Mattermost. “This actually demonstrates how fractured the cluster management space currently is.”
When it comes to multi-cluster management, there is no silver bullet. Though the report recognized HashiCorp’s Terraform and custom-built in-house tools as popular choices for deployment, few comprehensive solutions are available. Specific company requirements often contribute to a diverse toolset, which usually ends up requiring more tools to function than other facets of the container ecosystem, the report suggests.
The size of the organizations, application development team and the number of clusters will also dictate tooling preferences. For example, small deployments find kOps more desirable, whereas managed K8s is more common within larger deployments.
Kubernetes is quite flexible, allowing for individual configurations for security, networking and storage. Due to its extensibility, multi-cloud conditions and specific organization requirements, Kubernetes cluster management tends to involve additional custom tooling.
As a result, core services and add-ons include Flux, Customize, Operators, Argo, Helm and custom in-house tools—all of which are placed in the ‘Adopt’ category. Helm is commonly adopted with GitOps. Also, out of these, Operators stood out on top with 24 ‘adopt’ votes, likely due to their simple design, problem-solving capabilities and ability to tailor use to various projects, suggests the report. Many open source projects lie Kassandra, ElasticSearch and others now use operators.
The Radar also found that many developers are eagerly following the development of the Cluster API, a Kubernetes sub-project maintained by a Kubernetes Special Interest Group (SIG). A declarative Cluster API could significantly simplify the process of bootstrapping a working cluster, and therefore ease life cycle management (provisioning, upgrading and operating) for multiple Kubernetes clusters.
Multi-cluster management is still maturing; organizations have adopted various methods to manage their clusters, often involving in-house creations. Though several multi-cluster tooling options permeate the space, there is no clear choice as to which is the correct strategy.
“Everyone starts with one Kubernetes cluster and expands and grows their environments from there, but currently there is no clear path for organizations starting this journey,” says Federico Hernandez, principal engineer at Meltwater.
Nevertheless, it seems like organizations are making do. And hopefully, these findings come as a reassurance that others are implementing cluster management and provisioning in a similar way. Alternatively, it could influence movement toward new tooling decisions.
Tools like Helm and Operators are clearly popular for cluster management. The Cluster API also seems promising, though it’s still in the ‘assess’ stage at this time—companies are still evaluating its use. Some respondents felt it’s not adequately mature yet. However, Cluster API use may increase after more production evidence, especially if new features mean it accomplishes all kOps does and more.