[jira] [Commented] (NIFI-362) Funnels appear to consume a lot of CPU in some environments

["Mark Payne (JIRA)" <ji...@apache.org>]

    [ https://issues.apache.org/jira/browse/NIFI-362?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14332242#comment-14332242 ] 

Mark Payne commented on NIFI-362:
---------------------------------

The issue appears to be related to the fact that when a component has no work to do we continually run, checking for work to do. This patch backs that off a bit so that if a component has no work to do, it will be automatically "yielded" for 10 miliseconds. Additionally, when a component is yielded, rather than continually checking if the yield duration has ended, we cancel the task and re-submit it to start with an initial delay equal to the yield duration

> Funnels appear to consume a lot of CPU in some environments
> -----------------------------------------------------------
>
>                 Key: NIFI-362
>                 URL: https://issues.apache.org/jira/browse/NIFI-362
>             Project: Apache NiFi
>          Issue Type: Bug
>          Components: Core Framework
>            Reporter: Mark Payne
>            Assignee: Mark Payne
>         Attachments: 0001-NIFI-362-Avoid-continually-scheduling-components-to-.patch, 0002-NIFI-362-Ensure-that-we-synchronize-on-ScheduleState.patch
>
>
> At this point, a few different people have reported seeing high CPU usage with funnels. Personally, I have never been able to duplicate it on a windows machine (I have tried on a few different boxes). Looking at some Linux VM's (running on top of Linux hardware) I have a cluster of 3 nodes. When adding funnels, two of the nodes exhibit the behavior but one does not. The nodes are (in theory) identical.
> A coworker running Mac OSX has verified that he sees CPU usage of about 35% when adding a funnel to an otherwise-stopped flow.



--
This message was sent by Atlassian JIRA
(v6.3.4#6332)