[
https://issues.apache.org/jira/browse/KAFKA-3014?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Guozhang Wang resolved KAFKA-3014.
----------------------------------
Resolution: Fixed
Fix Version/s: 0.9.0.1
Issue resolved by pull request 696
[https://github.com/apache/kafka/pull/696]
> Integer overflow causes incorrect node iteration in leastLoadedNode()
> ----------------------------------------------------------------------
>
> Key: KAFKA-3014
> URL: https://issues.apache.org/jira/browse/KAFKA-3014
> Project: Kafka
> Issue Type: Bug
> Reporter: Jason Gustafson
> Assignee: Jason Gustafson
> Fix For: 0.9.0.1
>
>
> The leastLoadedNode() implementation iterates over all the known nodes to
> find a suitable candidate for sending metadata. The loop looks like this:
> {code}
> for (int i = 0; i < nodes.size(); i++) {
> int idx = Utils.abs((this.nodeIndexOffset + i) % nodes.size());
> Node node = nodes.get(idx);
> ...
> }
> {code}
> Unfortunately, this doesn't handle integer overflow correctly, which can
> result in some nodes in the list being passed over. For example, if the size
> of the node list is 5 and the random offset is Integer.MAX_VALUE, then the
> loop will iterate over the following indices: 2, 3, 2, 1, 0.
> In pathological cases, this can prevent the client from being able to connect
> to good nodes in order to refresh metadata.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
