Confidential Computing Pipelines for Kubernetes Observability Data: An Attested Architecture for Logs, Metrics, Traces, and Audit Events
Abstract
Kubernetes observability pipelines collect logs, metrics, traces, and audit events that are essential for reliability engineering, incident response, compliance, and capacity planning. However, these same signals frequently contain sensitive application payloads, tenant identifiers, access tokens, internal topology, operational metadata, and personally identifiable information. Conventional protections--transport encryption, storage encryption, role-based access control, and backend retention policies--do not protect telemetry while it is being parsed, enriched, sampled, correlated, redacted, routed, or queried. This creates a security gap: the observability pipeline itself becomes a privileged data-processing system. This paper proposes ConfObs-TEE, a confidential computing architecture for Kubernetes observability data. This design-science study presents a reference architecture and a prospective prototype-evaluation plan; it does not report measured implementation results. The architecture places the highest-risk telemetry operations inside hardware-backed, remotely attested Trusted Execution Environments (TEEs), while preserving compatibility with OpenTelemetry Collector patterns and Kubernetes-native deployment models. ConfObs-TEE uses attested collector agents, confidential gateway collectors, policy-bound key release, signed pipeline manifests, field-level tokenization, envelope encryption, and verifiable lineage metadata to reduce trust in cluster nodes, cloud administrators, telemetry operators, and observability backends. The design is grounded in current cloud-native mechanisms, including OpenTelemetry Collector receivers, processors, and exporters; Kubernetes metadata enrichment; Confidential Containers; Key Broker Service-style attestation; and the IETF RATS attestation model. The paper contributes a threat model for observability data in Kubernetes, a reference architecture, a policy and key-management design, a security analysis, and an evaluation methodology for future prototype validation. The central design conclusion is that confidential computing should not be applied indiscriminately to the entire observability stack, but selectively to the raw-data, enrichment, correlation, redaction, and key-handling stages where plaintext exposure and operator privilege are most consequential.