[ https://issues.jboss.org/browse/WFLY-11718?page=com.atlassian.jira.plugin... ] Chao Wang commented on WFLY-11718: ---------------------------------- This can be produced 100% in slow invocation (break point or time waiting among invocations). I think this happens due to the smallrye-metrics [ExponentiallyDecayingReservoir|https://github.com/smallrye/smallrye-metri...] implementation as explained {code:java} /** * Creates a new {@link ExponentiallyDecayingReservoir} of 1028 elements, which offers a 99.9% * confidence level with a 5% margin of error assuming a normal distribution, and an alpha * factor of 0.015, which heavily biases the reservoir to the past 5 minutes of measurements. */ {code} which updates item weight based on time calculation {code:java} final double itemWeight = weight(timestamp - startTime); {code} So this will eventually affect weighted arithmetic mean result. [ExportersMetricScalingTest.histogram_json|https://github.com/smallrye/sma...] calls assertEquals with a delta value at the end as: {code:java} assertEquals(120.0, json.getJsonNumber("mean").doubleValue(), 0.001); {code} Last two teamcity failures are: {noformat} java.lang.AssertionError: Expected value of 'mean' is 3.0, but was 3.0089998312537976. java.lang.AssertionError: Expected value of 'mean' is 3.0, but was 3.007477427516212. {noformat} [~jmesnil] Can we add an acceptable delta as well for this? -- This message was sent by Atlassian Jira (v7.12.1#712002)