Macoy Jones created KAFKA-6206:
----------------------------------
Summary: Kafka processes ignore read/write performance and
reliability
Key: KAFKA-6206
URL: https://issues.apache.org/jira/browse/KAFKA-6206
Project: Kafka
Issue Type: Improvement
Components: replication
Affects Versions: 0.9.0.0
Environment: RHEL 6.8
Reporter: Macoy Jones
Priority: Minor
According to the documentation, for a leader to be considered alive, it must be
able to maintain its session with ZooKeeper via ZooKeeper's heartbeat
mechanism.
As this criteria stands, a leader will continue to attempt to fulfill the
read/write requests for in-sync replicas without taking into account resource
availablity, performance factors, or hardware/application health. This
introduces the risk of unchecked degradation to the performance and/or
reliability of data read/write operations to a subset of data sources in the
event of a server problem, and the risk grows with partition leader density
A viable solution to this may be some type of logic that allows a leader to be
blacklisted as such if it fails to read/write beyond a configurable baseline.
--
This message was sent by Atlassian JIRA
(v6.4.14#64029)
