[jira] [Commented] (NIFI-1515) Large numbers of processors can cause stats to exhaust heap

["Puspendu Banerjee (JIRA)" <ji...@apache.org>]

    [ https://issues.apache.org/jira/browse/NIFI-1515?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15148143#comment-15148143 ] 

Puspendu Banerjee commented on NIFI-1515:
-----------------------------------------

A nohup can autostart NiFi in case of OOM, but there is a catch, you might end up hitting ulimits for fopen call.

> Large numbers of processors can cause stats to exhaust heap
> -----------------------------------------------------------
>
>                 Key: NIFI-1515
>                 URL: https://issues.apache.org/jira/browse/NIFI-1515
>             Project: Apache NiFi
>          Issue Type: Bug
>          Components: Core Framework
>    Affects Versions: 0.5.0
>            Reporter: Aldrin Piri
>            Assignee: Aldrin Piri
>
> There have been some OOME issues reported for very large graphs presumably tied to the stats held in a clustered environment on the NCM.
> An email exchange of the problem is detailed below:
> {quote}
> -----Original Message-----
> From: Corey Flowers [mailto:cflowers@onyxpoint.com]
> Sent: Monday, February 15, 2016 10:32 AM
> To: dev@nifi.apache.org
> Subject: EXTERNAL: Re: OutofMemory
> Hey Joseph,
>       I have a couple of clusters in the 600-1200 range and they have
> 16-32 GB jvm heap sizes respectfully. Really it depends on what processors you are using and your volumes. One thing that may help a little is to decrease your number of stored statics in the graph.
> There are two properties in the conf file, one is set to 1440 and the other is a time, which is every 1 min. I believe these stats are stored in the heap space. Devs correct me if I am wrong. You could lessen the time and amount to buy you a little space. I don't think this is a solution, really it is more of a band-aide.
> Good luck!
> Sent from my iPhone
> > On Feb 15, 2016, at 9:33 AM, Gresock, Joseph <jo...@lmco.com> wrote:
> >
> > Devs,
> >
> > We've been seeing some OutOfMemoryErrors on the NCM of our 10-node cluster recently.  The flow has ~600 processors, and the NCM runs on a VM with 8GB RAM.  We have 6G allocated to the Nifi JVM on this node.
> >
> > The specific log message we see is:
> >
> > WARN [Process NCM Request-6] org.apache.nifi.io.socket.SocketListener
> > Dispatching socket request encountered exception due to:
> > java.lang.OutOfMemoryError: Java heap space
> >
> > First, I'm hoping there's some advice on how to avoid this in the first place, but barring that, is there a way to configure Nifi to auto-restart the NCM when it gets this error?  I seem to remember seeing this in the past, but I couldn't find anything in bootstrap.conf or nifi.properties that looked related.
> >
> > Thanks,
> > Joe
> {quote}



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