K8sPCM: A Domain-Specific Language for Cloud-Native Performance Modeling Workflows

Kubernetes deployment decisions, such as replica counts and container resource limits, directly influence the performance and hardware cost of cloud-native systems. However, those decisions are normally validated only after actual deployment --- essentially, too late.
Moreover, the focus of existing tooling, which supports the definition and analysis of Kubernetes deployments is not analysis-ready and more focused on operational visualization. One existing solution here is to do predictive analysis using architecture-level performance models like the Palladio Component Model (PCM). But currently, PCM models can only be constructed from Kubernetes manifests, requiring a level of Palladio expertise beyond most practitioners. In this paper, we present k8sPCM, a ready-to-use Eclipse plugin built on an Xtext-based Domain-Specific Language (DSL). Our DSL mirrors Kubernetes manifest structure and one-click generates a scaffolded PCM instance while also providing content assist, construct templates, and on-the-fly validation. We evaluated the usability of k8sPCM with a user study, in which twelve participants of varying familiarity with Kubernetes and Palladio modeled a TeaStore subset. ... mehrOur results show that k8sPCM achieved a mean System Usability Scale score of 82.5 (i.e., nearly ``Excellent'' on the Bangor scale); further, we are able to demonstrate that higher Kubernetes familiarity correlated with lower reported mental effort.