shelf

Check and compare Trino replica performance before and after Shelf (read-cache) integration. Use this skill whenever the user asks to "check perf", "check today", "how is Shelf doing", "compare replica performance", "check replica1", or any variation of monitoring Trino cluster performance with or without Shelf cache. Also trigger when the user asks about read throughput, cache hit rates, query latency trends, or failure rate analysis on Trino replicas. This skill contains all validated SQL queries, methodology, and interpretation guidance for the Shelf performance analysis workflow.

Install Shelf (the Iceberg-native read cache for Trino) into a user's Kubernetes cluster end-to-end. Detects the Trino deployment, installs the Helm chart with sensible defaults, flips Trino's `s3.endpoint` to point at Shelf, validates the cutover with a byte-identity smoke query, and rolls back on signal. Use when the user says "install shelf", "make my Trino faster with shelf", "set up shelf for me", or "deploy shelf-project on my cluster". Designed for users with **no prior Trino / Helm / K8s expertise** — the agent does the work, the user only confirms cluster-mutating steps.