Skip to content

Recipe: Detectar una regresión de query

Bonito builds a 7-day baseline per query fingerprint and flags regressions automatically — no thresholds to tune.

How it works

The store computes mean_ms and p95_ms per fingerprint over the retention window. On demand it compares the current average against the baseline and returns:

{
  "db_instance": "default",
  "fingerprint": "-6842865755026457642",
  "baseline": { "samples": 10, "mean_ms": 12.43, "p95_ms": 14.1 },
  "current_avg_ms": 42.5,
  "regression_pct": 241.9
}

regression_pct > 0 → the query got slower than its baseline.

Via the API

curl -H "X-API-Key: $BONITO_API_KEY" "localhost:8100/baseline/-6842865755026457642"
curl -H "X-API-Key: $BONITO_API_KEY" "localhost:8100/anomalies"    # all regressions

Via Prometheus

The store exposes anomaly metrics on :8000/metrics:

  • bonito_query_regression_pct{fingerprint="..."}
  • bonito_query_anomaly{fingerprint="..."} → 1 when in regression

The BonitoRegression alert rule (bonito_query_anomaly == 1) turns that into an alert → Remo (see alert-to-remo).

Via MCP

User: "is query -6842865755026457642 slower than its baseline?"
Agent: compare_to_baseline(fingerprint="-6842865755026457642")
       → verdict: "REGRESSION — 241.9% slower than 7-day baseline. Suggest CREATE INDEX."

Demo

In the live tenant, sim-regression-001 was injected with a 10ms baseline and 120ms current → regression_pct: 1100%, bonito_query_anomaly: 1, and the BonitoRegression alert fired to Remo.