Kubernetes (K8s) is a fascinating piece of technology. It can run a variety of different workloads. However, Kubernetes is not the only option.
A closer look at Cloud Foundry (CF) will reveal that for several use cases, a Cloud Foundry environment may be preferred over Kubernetes. This article highlights Cloud Foundry for VMs (BOSH), in contrast to CF-4-K8s which is worth a dedicated comparison.
Table of Contents
While the idea of Kubernetes was to be a platform to build platforms, the concept of Cloud Foundry is to provide a turnkey application developer platform solution.
With Cloud Foundry, it is possible to set up a platform environment on any major infrastructure, whether public or on-premises. Especially when its life cycle is managed using BOSH, Cloud Foundry shows its strengths: running thousands of applications for hundreds of tenants in a single, large environment.
In such a Cloud Foundry environment, the update of operating systems, as it has been necessary after the Shellshock exploit, is as simple as changing a few lines of YAML code. The update will then be executed as a wave of reconstructing virtual machines across the environment when using BOSH-based data service automation. Often these environments comprise thousands of virtual machines hosting apps from different tenants. Cloud Foundry tenants, also called “organizations,” are regularly used to represent organizational units such as internal business units. Because of Cloud Foundry’s ability to isolate tenants, large environments are possible while maintaining a centralized platform, unfolding a significant economy-of-scale effect.
A Cloud Foundry platform environment initially comes with a relatively large infrastructure footprint. Generally speaking, the overhead of a Cloud Foundry environment depends on the level of redundancy and thus the desired availability of the resulting platform.
Once a Cloud Foundry environment has been set up, it provides several key services required by any application development platform, including user management and authentication services comprising the ability to create organizations and “spaces,” a container orchestrator to schedule workloads across the cluster, the ability to deploy code using buildpacks or container images as well as a central service marketplace.
The central marketplace in Cloud Foundry allows the integration of Open Service Broker API-compatible service brokers. The marketplace provides a list of available services and will enable users to manage data service instances in an on-demand self-service fashion.
Assuming there is a list of service brokers available, application developers can instantly deploy applications from local code, create database instances such as a PostgreSQL streaming cluster and bind apps to the database(s). The developer’s user experience is based on well-defined APIs, and workloads can be described using the CLI or by applying YAML specifications.
Simply speaking, Cloud Foundry provides enterprise-grade hosting, enabling a high degree of automation and the ability to enforce organizational policies.
The usage of buildpacks is an excellent example of how an organization can decide whether to allow local decisions in engineering teams or instead enforce the use of quality-assured provisioning automation by either allowing arbitrary buildpacks or restricting the choice to a list of well tested, security checked buildpacks.
The CF marketplace is another example of central governance. The selection of service brokers for data service automation and offering other services provides fine-grained control over how data is treated. Engineers do not need to select, install and life-cycle manage operators. All they have to do is select a data service, choose a service plan and create a service instance. It is the job of the service broker to allow comprehensive life-cycle management, including configuration changes, version upgrades, scalability, handling of logs and metrics, to name a few tasks.
Vanilla Kubernetes provides namespaces, a concept used to separate workloads within a Kubernetes cluster. However, namespaces do not isolate as well as Cloud Foundry organizations, nor does Kubernetes provide a user-management and authentication service out of the box.
If used as an application developer platform, Kubernetes environments tend to have a different structure than Cloud Foundry environments. In particular, the number of Kubernetes clusters is much higher than the number of Cloud Foundry instances. While a Cloud Foundry environment (the staging or production environment, usually contains a single Cloud Foundry installation), a Kubernetes environment often consists of multiple Kubernetes clusters. While a single Kubernetes cluster might be easier to operate than a single Cloud Foundry, managing 100 Kubernetes clusters is several orders of magnitude harder than managing a single Cloud Foundry. This scenario might appear extreme, but it shows the underlying problem.
Managing a large number of Kubernetes clusters creates complexity that is not necessary with Cloud Foundry. Admittedly, there is room for a dialectical discussion. Workloads must be compatible with Cloud Foundry, which is more demanding toward 12-factor compliance than Kubernetes. Kubernetes can be customized much more than Cloud Foundry and therefore accept workloads that Cloud Foundry couldn’t. Though this argument also makes a point for Cloud Foundry, which is a crucial statement to be pointed out: If an organization runs large swaths of Cloud Foundry-compatible workloads, Cloud Foundry might represent a better operational economy.
If workloads are more heterogeneous, Kubernetes wins, and increased operational complexity is to be tolerated.
Article previously published at The New Stack digital magazine.