Root Cause in One Question: Lustre Diagnostics with Harvest MCP- Nick Elvester | Net App

When a user reports a slow job, the answer usually spans two layers: the Lustre servers and the storage underneath them. Tracing it today means moving between Lustre dashboards, lfs and lctl, storage dashboards and the array management UI, then mapping OSTs to array volumes by hand. This session shows how NetApp Harvest, an open-source telemetry collector, and the Lustre exporter write to a single Prometheus database, and how the Harvest MCP server lets an AI client such as Copilot or Claude query both layers in plain English. We walk through how NetApp Harvest and an LLM can dramatically simplify troubleshooting at scale.