In modern IT operations, the cost of incorrect repairs often exceeds that of inaction. Existing automated remediation systems are designed to generate actions rather than determine the necessity of intervention, leaving safety concerns as an afterthought requiring manual approval. This paper makes the following contributions:
-
It reformulates safe remediation as a risk-constrained intervention decision problem, casting it as a Constrained Markov Decision Process (CMDP) where the agent maximizes repair success while maintaining a bounded false remediation rate (FRR).
-
A three-dimensional risk decomposition is introduced, encompassing blast radius, reversibility, and epistemic uncertainty, providing operators with an interpretable safety interface for each action.
-
A context-adaptive human-in-the-loop (HITL) gate is designed, transforming escalation from a binary failsafe into a bandwidth-aware control layer that responds to on-call load and business criticality.
The full policy is learned offline from historical incident logs, enabling explicit control of expected FRR. Experiments on the Train Ticket microservice benchmark, using Chaos Mesh fault injection and an RCAEval-aligned fault taxonomy, demonstrate that our framework reduces FRR by 39% while improving repair success by 2.5 points over a strong runbook baseline, and decreases on-call escalation load by 17% compared to a fixed-threshold variant.
Blogger's Review: This paper provides a novel approach to safe remediation in microservice systems through the CMDP model. By introducing risk constraints and human-in-the-loop mechanisms, it significantly enhances repair efficiency and safety, showcasing substantial practical value.