Report

When the metrics-api scaler aggregates metrics from multiple Kubernetes service endpoints with aggregationType: average (the default), and every endpoint request fails, aggregateMetricsFromMultipleEndpoints still divides the aggregated value by expectedNbMetrics after it has been decremented to 0.

Expected Behavior

When no endpoint returns a successful metric:

• The scaler should not perform average division with a zero divisor.
• It should return a clear error (e.g. could not fetch from any endpoint).
• The reported metric value should remain finite (e.g. 0) and must not be +Inf / NaN.

Actual Behavior

When all parallel endpoint fetches fail:

1. Each failure decrements expectedNbMetrics.
2. expectedNbMetrics reaches 0.
3. The code still runs aggregation /= float64(expectedNbMetrics) for aggregationType: average.
4. The function can return +Inf (or NaN) as the metric value.

Steps to Reproduce the Problem

.

Logs from KEDA operator

example

KEDA Version

2.19.0

Kubernetes Version

None

Platform

None

Scaler Details

metrics_api_scaler

Would you be open to contributing a fix?

Yes

Anything else?

No response