Re: Broken Pipe Exception - failed reporting to console

Lei <[email protected]> Tue, 26 Aug 2014 12:30:28 -0700
Newsgroups gmane.comp.java.grinder.user
Message-ID <CAD+vERsVHKv2=SO07=hMahtS5Z0MncXUvspxp-+p72Gqv7b78Q@mail.gmail.com>
On Mon, Aug 25, 2014 at 7:18 AM, Gary Mulder <[email protected]>
wrote:

> On 25 August 2014 08:34, Joan Picanyol i Puig <[email protected]>
> wrote:
>
>> * Lei <[email protected]> [20140823 16:20]:
>
>  > However once the agent start testing, it failed reporting to console
>> with
>> > below Broken Pipe exception.
>>
>> My bet is that you are trying to create processes and/or threads
>> too fast and overloading your agent. I suggest trying to diagnose wether
>> it is a maximum agent load issue or a rate issue (i.e.: wether you can
>> create the intended load over a longer timeframe).
>>
>
> To confirm whether Joan's theory is correct run some commands on the agent
> to measure load. On Linux run:
>
>  $ vmstat 3
> procs -----------memory---------- ---swap-- -----io---- -system--
> ----cpu----
>  r  b   swpd   free   buff  cache   si   so    bi    bo   in   cs us sy id
> wa
>  5  1      0 26217732 149992 2266752    0    0     0     3    8   10  0  0
> 100  0
>  7  0      0 26217116 149992 2266780    0    0     0     1 76425 91969 13
> 14 73  0
>
> The r column shows how many threads want to run on average. lsof -i is
> another useful command to confirm you're not running out of TCP/IP sockets
> (maximum 32K allowed).
>
> Of course if you're running Windows it might be harder to see if you're
> hitting resource limits...
>
> Regards.
> Gary
>
>
Thank you pica and Gary!

I didn't see the heavy load on the console. Looks like once the agent gets
some load (more than 100 TCP connections), the connection between agent and
console becomes unstable.

I have tested different configuration from grinder properties. Here are
some observations:

Reduce the number of below two properties to some small number, the problem
goes a way. But it defeat the purpose of load testing.

 grinder.processes
 grinder.threads

Reducing grinder.processIncrement and grinder.initialProcesses and
increasing grinder.processIncrementInterval can mitigate or delay this
problem sometimes.

Since I didn't find any better way to resolve this issue, I might just go
with agent without console, and use "grinder to graphite" (
https://bitbucket.org/travis_bear/grinder_to_graphite) for collecting
testing result.

Thanks,
Lei

------------------------------------------------------------------------------
Slashdot TV.  
Video for Nerds.  Stuff that matters.
http://tv.slashdot.org/

_______________________________________________
grinder-use mailing list
[email protected]
https://lists.sourceforge.net/lists/listinfo/grinder-use