[
https://issues.apache.org/jira/browse/IGNITE-10496?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16710730#comment-16710730
]
Peter Karich edited comment on IGNITE-10496 at 12/5/18 11:12 PM:
-----------------------------------------------------------------
I observed the following behaviour: when two nodes "put" objects with the same
frequency into a mostly empty queue then the "take" frequency on both nodes
should be similar. But one node is heavily preferred, forcing several
unnecessary hops to it. This should be more efficient.
was (Author: peathal):
I observed the following behaviour: when two nodes "put" objects with the same
frequency into a mostly empty queue then the "take" frequency on both nodes
should be similar. But what I observe is that one node is heavily preferred,
forcing several unnecessary hops to it. This should be more efficient.
> Unfair queue polling for mostly empty queue
> -------------------------------------------
>
> Key: IGNITE-10496
> URL: https://issues.apache.org/jira/browse/IGNITE-10496
> Project: Ignite
> Issue Type: Bug
> Affects Versions: 2.6
> Reporter: Peter Karich
> Priority: Minor
>
> I created a simple consumer+producer scenario:
> [https://github.com/karussell/igniteexample/blob/master/src/main/java/test/IgniteTest.java]
>
> This leads to the bug that mostly one machine gets all items from the queue,
> IMO it should work like round robin nature and make it possible for the other
> machine to poll. If this is not a bug it should be a least written somewhere
> in the documentation.
>
> Also please note, if you increase the sleep in the example above the load is
> equal like in the original example code here:
>
> http://apache-ignite-users.70518.x6.nabble.com/Clearing-a-distributed-queue-hangs-after-taking-down-one-node-td7353.html
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
