Diagnostics

Diagnostics warn about patterns no single query would show you. They count within a scope, usually one web request, and cost a little per statement.

Setting it up

db, err := ant.Open[AppDB](ctx, postgres.Dialect, dsn,
	ant.WithDiagnostics(ant.Diagnostics{
		RepeatLimit: 10,
		Budget:      500 * time.Millisecond,
		Alert: func(ctx context.Context, a ant.Alert) {
			slog.Warn("database", "kind", a.Kind, "message", a.Message, "sql", a.SQL)
		},
	}),
)

The default Alert logs through slog.

A scope per request

ant.Watchfunc Watch(ctx context.Context) context.ContextWatch returns a context that counts statements for Diagnostics. starts counting. Without a scope, repeats and budgets don't count.

func middleware(next http.Handler) http.Handler {
	return http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
		next.ServeHTTP(w, r.WithContext(ant.Watch(r.Context())))
	})
}

What it finds

Kind	Means
repeated	One statement ran more than RepeatLimit times in a scope, which is what N+1 looks like
budget	Statements spent longer than Budget in total, which a slow-query log can't see if each is fast
cartesian	A query joined two to-many relations, which multiplies rows. This warns with or without a scope

See it tested: tests/features/developer_tooling/diagnostics_test.go