Financial institutions are rapidly migrating critical workloads to cloud-native stacks, yet Kubernetes and IaC do not inherently provide the operational resilience required for regulated payment, treasury, and credit platforms. Incidents are frequently change-induced (unsafe configuration, missing disruption budgets, mis-sized resources) or correlated across failure domains (zone concentration, retry amplification, dependency latency), while observability remains descriptive rather than controlling. This paper introduces the Reliability-Aware Architecture Design Framework (RAADF), a lifecycle that translates business criticality into measurable Service Level Objectives (SLOs) and error budgets, synthesizes tier-specific Kubernetes resilience constraints, and enforces them through policy-as-code and GitOps-managed IaC. RAADF closes the loop by computing SLIs from production telemetry, gating progressive delivery on error-budget burn, and continuously validating assumptions via fault-injection tests. We specify a reference architecture, an implementation playbook, and a reproducible evaluation protocol. In controlled experiments on a payment-authorization workload deployed across three availability zones, we compare a baseline Kubernetes/IaC setup against RAADF under (i) zone loss, (ii) network partition and dependency-latency injection, and (iii) unsafe rollout scenarios. Results are reported using MTTR-to-SLO, availability and p95 latency SLIs, burn-rate dynamics, change failure rate, rollback time, and blast radius. RAADF enables resilience that is both operationally effective and auditable for regulated finance. By making reliability constraints explicit artifacts—SLO catalogues, policy rules, release gates, and test evidence—RAADF supports governance, incident learning, and third-party assurance. The approach is tool-agnostic: it can be implemented with Kubernetes distributions and telemetry backends without changing the lifecycle semantics.
Jayavardhan Reddy (Thu,) studied this question.