Dong Lin created KAFKA-2280:
-------------------------------
Summary: Measurement of bytesInRate and bytesOutRate in
BrokerTopicMetrics should be consistent
Key: KAFKA-2280
URL: https://issues.apache.org/jira/browse/KAFKA-2280
Project: Kafka
Issue Type: Bug
Reporter: Dong Lin
There is inconsistency in the way bytesInRate and bytesOutRate is measured in
Kafka -- when the follower fetches bytes from the leader, the leader's
bytesOutRate is increased but follower's bytesInRate is not.
Ideally we should disambiguate between the two (i.e., replication traffic and
client traffic). But doing that changes the API in the sense that other
applications (e.g. iGraph) will need to read from different metric names. For
example, we will need to remove bytesInRate and bytesOutRate, and add
clientBytesInRate, clientBytesOutRate, replicationBytesInRate, and
replicationBytesOutRate.
The easiest solution is to include replication traffic in
BrokerTopicMetrics.bytesInRate. It is probably consistent with what users will
expect from bytesInRate.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
