[ https://issues.apache.org/jira/browse/KAFKA-1930?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14369623#comment-14369623 ]
Jun Rao commented on KAFKA-1930: -------------------------------- Another thing that we need to think through is the histogram support in the new metrics package. So far, we haven't been using histogram in the clients because (1) it needs to know the range of the value and (2) it needs to pick a reasonable BinScheme. For example, in the broker side, we use a histogram to measure the request time. The range of the time can be large since it depends on the request timeout set by different clients. Typically, those values will be small, say just a few milli secs. However, sometimes those values can be really large, say 10s of seconds. So, it's not clear if the current histogram support in the new metrics can measure this effectively. The way that histogram in Coda Hale works is to use a variant of reservoir sampling. It works reasonably well with data of different ranges and distributions. The downside is that it can potentially use more memory. > Move server over to new metrics library > --------------------------------------- > > Key: KAFKA-1930 > URL: https://issues.apache.org/jira/browse/KAFKA-1930 > Project: Kafka > Issue Type: Improvement > Reporter: Jay Kreps > Assignee: Aditya Auradkar > > We are using org.apache.kafka.common.metrics on the clients, but using Coda > Hale metrics on the server. We should move the server over to the new metrics > package as well. This will help to make all our metrics self-documenting. -- This message was sent by Atlassian JIRA (v6.3.4#6332)