Support Data Pipeline Design: Separate Metrics for Alerts and Logs for Rollback Evidence

A technical article advises separating telemetry types for customer support data pipelines. It argues dashboards should provide low-cardinality metrics for alerting on pipeline health. Structured logs are needed to provide the detailed, searchable evidence required for safe rollback operations. This separation prevents operational errors and ensures the business database remains the authoritative state. The design ensures detection and reversal of issues are handled by appropriate systems.