[ 
https://issues.apache.org/activemq/browse/SM-1789?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=49437#action_49437
 ] 

Ron Gavlin commented on SM-1789:
--------------------------------

Chris, 

You might want to take a look at the throttling technique Guillaume is using in 
the new SMX4 cluster endpoint and follow suit when solving this problem. I 
believe he is using property maxPendingExchanges to capture the throttling 
high-water mark.

> smx-file async FilePollerEndpoint needs a throttling mechanism to avoid 
> creating excessive numbers of open exchanges and overloading the nmr
> --------------------------------------------------------------------------------------------------------------------------------------------
>
>                 Key: SM-1789
>                 URL: https://issues.apache.org/activemq/browse/SM-1789
>             Project: ServiceMix
>          Issue Type: Bug
>          Components: servicemix-file
>    Affects Versions: servicemix-file-2008.01
>            Reporter: Ron Gavlin
>            Assignee: Chris Custine
>            Priority: Critical
>
> now that the smx-file poller is asynchronous, i run into saturation problems 
> when several thousand files exist in the polled directory at startup. This is 
> similar to the problems I encounter in the async smx-jms consumer when I have 
> thousands of messages on the queue at startup. The jms issue is captured in 
> SM-1777. 
> I suggest that an abstract throttling mechanism be implemented so that config 
> options can be shared across components. I suspect the primary config option 
> would be maxOpenExchanges. When this value is exceeded for the smx-file 
> poller, it would stop polling until the maxOpenExchanges is reduced below 
> maxOpenExchange. For the smx-jms consumer or the cxf-bc jms consumer, setting 
> the Spring JMS concurrentConsumers to 0 would stop consumption.

-- 
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.

Reply via email to