Improve PrometheusReporter performance #93
Description
PrometheusReporter#reportHistogramDurationSamples and PrometheusReporter#reportHistogramValueSamples scored pretty bad in terms of throughput (0.693 ± 0.102 ops/ms and 0.777 ± 0.120 ops/ms, respectively) comparing to other StatsReporter implementations (e.g. M3Reporter: 188.096 ± 7.091 ops/ms and 127.358 ± 27.740 ops/ms).
The poor performance caused by usage of Histogram#observe method in a cycle, which executes another for-cycle over all the histogram's buckets (see discussion in the PR for more details).
Metadata
Metadata
Assignees
Labels
No labels