Design of Dependency-Aware Package Orchestration Framework for Cloud-Native Network Functions

초록

Declarative deployment has emerged as a foundational approach for managing cloud-native systems, particularly cloud-native network functions (CNFs), where desired state specifications guide automated reconciliation. Modern telecom services such as network slicing and service function chaining rely on sets of interdependent functions whose correctness depends on coordinated initialization and readiness ordering. However, existing declarative package models treat each package as an isolated reconciliation unit and provide no abstraction for representing these group-level structures or their dependency relationships, resulting in misordered rollouts and unstable CNF initialization. To address this limitation, we propose a dependency-aware package orchestration framework that extends declarative management to support the coordinated rollout deployment of CNFs that must be deployed as a unified group. The framework introduces a DependencyGroup custom resource for declaratively defining the group composition and the dependency relations among its CNFs, and a Dependency Orchestrator that interprets these specifications to enforce readiness-aware and correctly ordered rollout while remaining fully compatible with GitOps workflows. A prototype implementation evaluated on a cloud-native 5G Core deployment shows that, compared with three baseline declarative strategies (FlatPackage_Sync, Init-GatedPackage_Sync, and ArgoSyncWave_Manual, the proposed approach eliminates rollout violations and reduces deployment latency by 19% compared to ArgoSyncWave_Manual and 31-54% compared to coordination-free approaches. These results demonstrate the effectiveness of integrating group-level dependency semantics directly into declarative orchestration for improving the reliability and efficiency of complex cloud-native network function deployments.

키워드