[Bug] pulsar-perf outputs values in nanos with values saying "ms"

[dlg99]

Search before asking

• I searched in the issues and found nothing similar.

Read release policy

• I understand that unsupported versions don't get bug fixes. I will attempt to reproduce the issue on a supported version of Pulsar client and Pulsar broker.

Version

2.10+, 3.x

Minimal reproduce step

run pulsar-perf consume

What did you expect to see?

latency values in milliseconds, as labels suggest

What did you see instead?

org.apache.pulsar.testclient.PerformanceConsumer - Throughput received:  332571 msg --- 3002.897  msg/s --- 165.144 Mbit/s  --- Latency: mean: 18740630.690 ms - med: 18640383 - 95pct: 20181503 - 99pct: 20482175 - 99.9pct: 20512767 - 99.99pct: 20512895 - Max: 20512895

value sin nanos with labels "ms"

Anything else?

I think here

pulsar/pulsar-testclient/src/main/java/org/apache/pulsar/testclient/PerformanceConsumer.java

Lines 464 to 472 in e5bd774

	log.info(
	"Throughput received: {} msg --- {} msg/s --- {} Mbit/s "
	+ "--- Latency: mean: {} ms - med: {} "
	+ "- 95pct: {} - 99pct: {} - 99.9pct: {} - 99.99pct: {} - Max: {}",
	intFormat.format(total),
	dec.format(rate), dec.format(throughput), dec.format(reportHistogram.getMean()),
	reportHistogram.getValueAtPercentile(50), reportHistogram.getValueAtPercentile(95),
	reportHistogram.getValueAtPercentile(99), reportHistogram.getValueAtPercentile(99.9),
	reportHistogram.getValueAtPercentile(99.99), reportHistogram.getMaxValue());

(and possibly in other places) histogram returns latency as it tracked, in nanoseconds. The perf tool should translate these into millis, as labeled.

This is not a blocker/not a priority, logging this as "gotcha" if anyone else wonders about the mismatch between values and labels.

Are you willing to submit a PR?

• I'm willing to submit a PR!