> On May 13, 2015, 11:50 p.m., Aditya Auradkar wrote: > > clients/src/main/java/org/apache/kafka/common/metrics/stats/SampledStat.java, > > line 45 > > <https://reviews.apache.org/r/34170/diff/2/?file=958605#file958605line45> > > > > I think this is a good catch. > > > > Just so I understand, this is adding new samples for blank intervals of > > time when record was not called. As an optimization, I think you should > > only add "MetricConfig.samples" number of samples. In the rare case, there > > is no activity for 10 mins (say), this will add 10sec*6*10 = 600 samples > > which will be purged immediately on the next record call.
This is a really good point. It is totally possible for a metric to track activity on a topic that has no writes for a month, the first write would then cause you to cycle through a month of samples. The logic around correctly skipping these windows and calculating the correct window boundry needs to be carefully worked out. - Jay ----------------------------------------------------------- This is an automatically generated e-mail. To reply, visit: https://reviews.apache.org/r/34170/#review83687 ----------------------------------------------------------- On May 13, 2015, 10:32 p.m., Dong Lin wrote: > > ----------------------------------------------------------- > This is an automatically generated e-mail. To reply, visit: > https://reviews.apache.org/r/34170/ > ----------------------------------------------------------- > > (Updated May 13, 2015, 10:32 p.m.) > > > Review request for kafka. > > > Bugs: KAFKA-2191 > https://issues.apache.org/jira/browse/KAFKA-2191 > > > Repository: kafka > > > Description > ------- > > KAFKA-2191; Measured rate should not be infinite > > > Diffs > ----- > > clients/src/main/java/org/apache/kafka/common/metrics/stats/Rate.java > 98429da34418f7f1deba1b5e44e2e6025212edb3 > > clients/src/main/java/org/apache/kafka/common/metrics/stats/SampledStat.java > b341b7daaa10204906d78b812fb05fd27bc69373 > > Diff: https://reviews.apache.org/r/34170/diff/ > > > Testing > ------- > > > Thanks, > > Dong Lin > >