jmeter-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Philippe Mouawad <philippe.moua...@gmail.com>
Subject Re: IF Controller degrading throughput
Date Fri, 09 Dec 2011 06:38:00 GMT
Then your issue CAN come from soapXml/rpc sampler.
Because my tests of the 2 propositions i made don't show any issue anymore,
Max perm gen reached is 110 m with javascript and 40 mo with expression.

Maybe you should open a Bugzilla issue attaching your test plan.
Regards
Philippe
Http://www.Ubik-Ingenierie.com

On Thursday, December 8, 2011, Stanton W Derry <derry@cpt.com> wrote:
> Hello,
>
> I have already exceeded your recommendation, my setting is at 256m.  I
came to this value running with SoapXML/RPC samplers.  If I change to just
HTTP samplers I don't see the same growth in Permgen.  Will try the
Variable expression setting.
>
> Stan
>
> On 12/8/11 4:49 PM, Philippe Mouawad wrote:
>
> Hello,
> After further analysis, this is not a memory leak but a size
misconfiguration:
> Fix is to add this in jmeter file to HEAP variable.:
> -XX:MaxPermSize=128m
>
>
> Another simpler option is to check:
> "Interpret condition as Variable expression"
>
> Please let me know which one of the two works.
>
> Regards
> Philippe.
> http://www.ubik-ingenierie.com
>
> On Thu, Dec 8, 2011 at 10:04 PM, Philippe Mouawad <
philippe.mouawad@gmail.com> wrote:
>
> Hello,
> I confirm the issue is due to a memory leak but not in Heap zone but in
Non Heap zone. I reproduced the issue.
> See :
>
> https://issues.apache.org/bugzilla/show_bug.cgi?id=52305
>
> Issue is due to either a leak in rhino or in our JMeter usage of rhino.
>
> Workaround is to use expression in your IfController
>
> ${__jexl(${JMeterThread.last_sample_ok} == true)}
>
>
>
>
>
>
> Regards
> Philippe
> On Thu, Dec 8, 2011 at 9:58 PM, Stanton W Derry <derry@cpt.com> wrote:
>
> Hi Phillippe,
>
> Memory isn't an issue as there are no heap dumps for any of my runs. Also
my heap settings are -Xms2048m -Xmx2048m.  I noticed the issue with my
performance runs and started eliminating the HTTP/Soap Samplers in my test
plan.  When I disabled the IF Controller the throughput increased.  I then
created a three Sampler plan using the Java Request Sampler and an If
Controller to verify my findings.  The Java Request Sampler settings are:
>
> Sleep_Time 20
> Sleep_Mask 0x00
>
> The only listener used is the Summary Listener.
>
> Easily recreated.
>
> Stan
>
>
> On 12/8/11 3:07 PM, Philippe Mouawad wrote:
>
> Hello,
> Can you make your test plan available on a public URL (or maybe open a
Bugzilla issue attaching it).
> What kind of sampler do you use ?
> What Listeners do you have  except "Generate Summary Results" ?
> Can you open jmeter and search for:
> DUMP="-XX:+HeapDumpOnOutOfMemoryError"
>
> And add this to it
> -XX:HeapDumpPath=<change path here>
>
> Run you test and check no file is generated in the path, if you find
something put the hprof file on external website.
>
>
> Regards
> Philippe
> http://www.ubik-ingenierie.com
>
> On Thu, Dec 8, 2011 at 7:18 PM, Stanton W Derry <derry@cpt.com> wrote:
>
> I'm using JMeter 2.5.1 r1177103 and noticed when an If Controller is used
it degrades the throughput till a full gc occurs at which time the
throughput is back up. I used two different test scripts to verify this
anomaly.
>
> The test plans used are:
>
> Thread Group 1 - Threads 40 - Forever - Duration -3600
>    Sampler
>    Sampler
>    Sampler
>
> and
>
> Thread Group  2 - Threads 40 - Forever - Duration -3600
>    Sampler
>    IF Controller - ${JMeterThread.last_sample_ok}
>            Sampler
>    Sampler
>
> The samplers response times are ~20ms and in the Summary Listener for
Thread Group 1 see a total throughput of 1136.6/sec.  For Thread Group2 the
max reached is 714/sec, declines to about 600/sec, gc executed, and then
climbs back to ~670 starting the cycle over.
>
> Has anyone encountered a similar issue when using the IF Controller?
>
> Thanks
> Stan
>
> ---------------------------------------------------------------------
> To unsubscribe, e-mail: user-unsubscribe@jmeter.apache.org
> For additional commands, e-mail:

-- 
Cordialement.
Philippe Mouawad.

Mime
  • Unnamed multipart/alternative (inline, None, 0 bytes)
View raw message