mirror of
https://github.com/Memo-2023/mana-monorepo.git
synced 2026-05-20 01:21:24 +02:00
c94ab01c69
Replaces the stub /metrics endpoint with a real prom-client registry
(mana_mcp_ prefix, {service="mana-mcp"} default label). Default
process metrics come along for free.
Policy-gate telemetry is the whole point — without it we can't soak
POLICY_MODE=log-only safely or decide when to flip to enforce. New
counter mana_mcp_policy_decisions_total{decision, reason, mode} buckets
every evaluatePolicy() call:
decision ∈ {allow, deny, flagged}
reason ∈ {admin-scope-not-invokable, destructive-not-allowed,
rate-limit-exceeded, injection-marker, clean, unknown}
mode ∈ {log-only, enforce}
So the rate of "would have been denied" during soak is visible directly
as policy_decisions_total{decision="deny", mode="log-only"}.
Also:
- mana_mcp_tool_invocations_total{tool, outcome} — success |
handler-error | input-invalid. Policy denies are NOT counted here
(they're in policy_decisions_total above); this counter only counts
calls that actually reached the handler or tripped zod validation.
- mana_mcp_tool_duration_seconds histogram per tool/outcome.
Dep: prom-client ^15.1.3 (same version mana-ai pins).
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
60 lines
2.5 KiB
TypeScript
60 lines
2.5 KiB
TypeScript
/**
|
|
* Prometheus metrics — exported on GET /metrics.
|
|
*
|
|
* Mirrors the shape of `services/mana-ai/src/metrics.ts` so Grafana and
|
|
* status.mana.how recognise this service without special-casing. Metric
|
|
* names use the `mana_mcp_*` prefix; labels stay low-cardinality on
|
|
* purpose (tool name is high-cardinality but still a fixed registry, so
|
|
* it's acceptable — we have ~20 tools today).
|
|
*/
|
|
|
|
import { Counter, Histogram, Registry, collectDefaultMetrics } from 'prom-client';
|
|
|
|
export const register = new Registry();
|
|
register.setDefaultLabels({ service: 'mana-mcp' });
|
|
collectDefaultMetrics({ register, prefix: 'mana_mcp_' });
|
|
|
|
// ── Policy gate ──────────────────────────────────────────────
|
|
|
|
/**
|
|
* One sample per `evaluatePolicy()` call.
|
|
*
|
|
* Labels:
|
|
* - `decision`: `allow` | `deny` | `flagged` (flagged = allow with a
|
|
* reminder, e.g. freetext injection marker hit)
|
|
* - `reason`: `admin-scope-not-invokable` | `destructive-not-allowed`
|
|
* | `rate-limit-exceeded` | `injection-marker`
|
|
* | `clean` (no reason applied; for dashboards)
|
|
* - `mode`: `log-only` | `enforce` — lets us diff how many
|
|
* decisions WOULD block vs. actually blocked during soak
|
|
*/
|
|
export const policyDecisionsTotal = new Counter({
|
|
name: 'mana_mcp_policy_decisions_total',
|
|
help: 'Tool-policy gate decisions, bucketed by outcome and reason.',
|
|
labelNames: ['decision', 'reason', 'mode'] as const,
|
|
registers: [register],
|
|
});
|
|
|
|
// ── Tool invocations ─────────────────────────────────────────
|
|
|
|
/**
|
|
* Every tool that makes it past the policy gate lands here. `outcome`
|
|
* is `success` | `handler-error` | `input-invalid` so dashboards can
|
|
* differentiate "tool ran but failed" from "LLM sent malformed args".
|
|
* Policy-denied calls are NOT counted here — they never reach the
|
|
* handler — and are visible under `policyDecisionsTotal{decision='deny'}`.
|
|
*/
|
|
export const toolInvocationsTotal = new Counter({
|
|
name: 'mana_mcp_tool_invocations_total',
|
|
help: 'Tool handler invocations (after policy gate).',
|
|
labelNames: ['tool', 'outcome'] as const,
|
|
registers: [register],
|
|
});
|
|
|
|
export const toolDuration = new Histogram({
|
|
name: 'mana_mcp_tool_duration_seconds',
|
|
help: 'Handler wall-clock latency per tool.',
|
|
labelNames: ['tool', 'outcome'] as const,
|
|
buckets: [0.005, 0.01, 0.025, 0.05, 0.1, 0.25, 0.5, 1, 2.5, 5, 10],
|
|
registers: [register],
|
|
});
|