Metrics Reference¶

Tool Metrics¶

Each tool call records metrics exposed by the health tool under tool_metrics.

Schema¶

{
  "tool_metrics": {
    "search": {
      "calls": 1250,
      "errors": 3,
      "error_rate_bps": 24,
      "total_ms": 45230,
      "avg_ms": 36,
      "p95_ms": 500,
      "p99_ms": 892,
      "max_ms": 892,
      "buckets": {
        "<=5ms": 234,
        "<=10ms": 456,
        "<=50ms": 389,
        "<=100ms": 98,
        "<=500ms": 67,
        "<=1000ms": 5,
        ">1000ms": 1
      }
    }
  }
}

Fields¶

Field	Type	Description
`calls`	integer	Total calls
`errors`	integer	Total failed calls
`error_rate_bps`	integer	Error rate in basis points (`10000 * errors / calls`)
`total_ms`	integer	Cumulative latency
`avg_ms`	integer	Mean latency
`p95_ms`	integer/null	Approximate p95 latency
`p99_ms`	integer/null	Approximate p99 latency
`max_ms`	integer	Maximum single-call latency
`buckets`	object	Histogram buckets

Search Concurrency¶

health also reports search saturation under search_concurrency:

{
  "search_concurrency": {
    "enabled": true,
    "max_concurrent": 4,
    "available_slots": 2,
    "in_flight": 2,
    "saturated": false,
    "max_queue": 8,
    "available_queue": 8,
    "queued": 0,
    "queue_saturated": false
  }
}

Use this to detect queue pressure and tune:

DBT_NOVA_SEARCH_MAX_CONCURRENT
DBT_NOVA_SEARCH_MAX_QUEUE