Adaptive microservice management becomes the key to SLO in cloud-native environments. The analysis shows how system dynamics affect control and architecture.
Microservice systems amplify dynamics that often remained hidden in monoliths. The load becomes non-stationary, the call graph changes with each release, and the network and neighboring workloads introduce noise. As a result, management turns into a task with partial observability and changing conditions. This directly links autoscaling, placement, routing, and isolation into a single control loop. The problem does not manifest immediately — until local optimizations begin to conflict and break the SLO.
The authors consider adaptive microservice management as a closed feedback loop: telemetry → decision → actuation. The foundation consists of cloud-native primitives: Kubernetes (HPA, VPA, cluster autoscaling), service mesh (Istio, Envoy), and the observability stack (Prometheus, Jaeger, OpenTelemetry). The key idea is to classify systems along four axes: where control is applied, what dynamics are modeled, how decisions are made, and how realistic the assessment is. This separation removes confusion between mechanisms and the real conditions in which they operate.
The synthesis of 84 systems shows a consistent pattern: dynamics in production are partially modeled. For example, workload is considered more often than the evolution of the call graph or network variability. At the same time, decisions at different levels — from node to service mesh — interact and can create oscillations or delays in convergence. A separate insight is that the quality of results heavily depends on the fidelity of the assessment. Synthetic loads and simplified simulations often do not reproduce the coupling between orchestration and the network, which distorts conclusions.
The practical takeaway for architects is that transitioning to dynamics-aware management requires coordination between layers. Isolated controllers (for example, an autoscaler without considering routing) provide local gains but worsen end-to-end latency. Abstractions are needed that connect telemetry and control decisions, as well as more realistic evaluation pipelines (for example, Kubernetes-in-the-loop). Without this, even correct algorithms exhibit unstable behavior in production. In the industry, this looks like an evolutionary improvement: from reactive controllers to coherent systems that account for the full dynamics.
Information source
arXiv is the largest open preprint repository (since 1991, under the auspices of Cornell), where researchers quickly post working versions of papers; the materials are publicly accessible but do not undergo full peer review, so results should be considered preliminary and, where possible, checked against updated versions or peer‑reviewed journals. arxiv.org