You can also identify where top performance hitters are and ensure that a
ControlRate or otherwise throttled amount of data and/or threads are
leveraged at once.  This allows you to effectively control how much effort
to put on any single point of the flow at once.  This is necessary when you
want to avoid a hotspot which consumes to many resources and you are
willing to trade off latency for ensuring the system never runs 'too hot'.

Thanks

On Fri, Jun 7, 2019 at 11:54 AM Woodcock, Michael W., M.B.A. <
[email protected]> wrote:

> We had this issue when NiFi flows were blocked and lots of content and
> flowfile data was on disk and in RAM.  We got around it by temporarily
> setting NiFi  JVMs to more RAM to allow the glut of content to pass and
> then lowering the JVM RAM back to normal so that garbage collection can
> occur.
>
> In Ambari
>
> Initial memory allocation
>
> Max memory allocation
>
>
>
> Mike Woodcock
>                                                      Mayo Clinic Division
> of Information Technology
> Big Data Team, VH South, 2-205                                200 First St
> SW, Rochester, MN USA 55905
> 507. 293.8496
>
>
>
>

Reply via email to