Performance index always returns 100%

[ryanjclark]

I don't think this query makes sense. It always returns 100%, even on a stricter bucket like le="5".

I think a more appropriate query would be:

    sum(
      rate(
        strimzi_canary_records_consumed_latency_bucket{le="5"}[5m]
      )
    )
  /
    sum(
      rate(
        strimzi_canary_records_consumed_latency_bucket{le="+Inf"}[5m]
      )
    )
*
  100

[pantaoran]

I was also thinking about this recently.
We had a situation where the average latency of one broker was consistently >1000ms. This led to the performance index just omitting that broker's time series, because it resulted in NaN since the dividend was 0.
The line of the performance index was still a straight 100%, just with the values from the "bad" broker missing. That is pretty useless I feel.