skippy-metrics

Shared telemetry naming conventions for staged runtime components.

Use this crate for stable attribute keys, metric names, and report vocabulary that must line up across skippy-server, metrics-server, benchmarks, and correctness tooling.

Architecture Role

skippy-metrics is the shared vocabulary for the staged request path. It does not collect or store telemetry itself; it keeps names stable while other crates emit or consume OTLP data.

flowchart LR
    Mesh["mesh-llm<br/>topology + lifecycle metadata"] -.-> M["metrics-server"]
    S0["stage servers<br/>request summaries"] -.-> M
    B["bench / diagnostic drivers<br/>run metadata"] -.-> M
    M --> D["metrics.sqlite"]
    M --> R["report.json<br/>SQLite-backed views"]

    V["skippy-metrics<br/>names + attributes"] -.-> S0
    V -.-> B
    V -.-> M

The hot inference path must not block on telemetry. Stage servers and runtime cache operations emit best-effort summaries, while metrics-server owns ingestion, SQLite storage, and report export.