Declarative approaches for custom cloud-edge serverless function scheduling

Serverless computing has evolved beyond cloud-centric models to include private edge and hybrid cloud-edge deployments, aiming to address latency and resource efficiency. However, existing serverless platforms typically rely on fixed, hardcoded scheduling policies that cannot meet the diverse functional, topological, and performance requirements of modern applications. In particular cloud-edge serverless applications, or deployments, spanning multiple regions introduce the need to govern the scheduling of functions to satisfy their functional constraints or avoid performance degradation to meet user-defined goals. This work addresses function scheduling by introducing Allocation Priority Policies (APP), a declarative language that allows developers to define application-specific scheduling policies. APP is implemented and evaluated in a hybrid cloud-edge scenario on top of Apache OpenWhisk. Building on APP, we develop tAPP, a topology-aware extension capable of enforcing placement constraints without requiring platform modifications. We demonstrate that tAPP supports deployment scenarios not feasible with unmodified OpenWhisk. We further extend the model with aAPP to express affinity-aware scheduling, where function allocation depends on the presence or absence of other functions. aAPP captures these constraints with minimal runtime overhead and improves performance where affinity is relevant. To reduce the manual effort in defining effective policies, we propose cAPP, which integrates static analysis. cAPP extracts cost models from function code and incorporates them into scheduling decisions, enabling more informed function placement. Finally, we present FunLess, a lightweight serverless platform for Cloud-Edge environments, using WebAssembly as its function runtime and featuring built-it support for APP. Moreover, we perform a comparative analysis of the energy consumption of FunLess, OpenWhisk and a container-based service architecture.