[ 
https://issues.apache.org/jira/browse/DISPATCH-58?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14009678#comment-14009678
 ] 

Ted Ross commented on DISPATCH-58:
----------------------------------

Another possibility:  Modify the router so that it doesn't replenish credit 
when a message is undeliverable, thus applying link backpressure until there is 
a destination for the address.


> Router/Broker interaction - Unroutable/released messages are continually 
> resent
> -------------------------------------------------------------------------------
>
>                 Key: DISPATCH-58
>                 URL: https://issues.apache.org/jira/browse/DISPATCH-58
>             Project: Qpid Dispatch
>          Issue Type: Bug
>          Components: Routing Engine
>    Affects Versions: 0.2
>            Reporter: Ted Ross
>
> There's an issue with the fact that the router releases unroutable messages.  
> When the message originates from a broker, that broker will continually 
> re-send the message, burning network bandwidth, until the message becomes 
> routable.
> Here are a few ideas as to how to address this situation:
>   - Have the ingress router hold the message for a time if it is not routable 
> to avoid rapidly releasing.
>   - Introduce a notion of not-routable-now vs. not-routable-ever.  This would 
> allow the router to hold-for-retry or immediately reject.
> Are there cases where the current behavior is desirable?  If the producer 
> node has multiple outgoing links, a quick release would allow it to use an 
> alternate link.



--
This message was sent by Atlassian JIRA
(v6.2#6252)

---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]

Reply via email to