[Tsung] Total number of requests on a test seems weird

Sebastian Cohnen sebastian.cohnen at gmail.com
Thu Aug 27 10:50:26 CEST 2015


Hey,

> On 27.08.2015, at 10:14, Emmanuel Peroumalnaïk <peroumalnaik.emmanuel at gmail.com> wrote:
> 
> Hello Sebastian, 
> 
> I do have errors in my tsung.log. If I take the last sample from the logs, I have the following : 
> 
> stats: error_unknown 0 49

That happens from time to time, should not be a problem in these low numbers.

> stats: error_connect_nxdomain 0 829

That's interesting :)

> stats: error_unknown_msg 0 15231
> stats: error_connect_eaddrinuse 0 28183

This is a problem. You could try to do several tweaks on the network layer, like increase the port range of ephemeral ports ("net.ipv4.ip_local_port_range = 10240 65535"), or enable ("net.ipv4.tcp_tw_reuse = 1"). That should help with that issue.

> stats: error_connect_econnrefused 0 25
> stats: error_abort_max_conn_retries 0 17538
> stats: error_abort 1 1
> 
> I noticed before that I had lots of 'eaddrinuse' errors, which is due to the fact that I had lots of clients making lots of requests, which means that at some point it will exhaust the ports available on the clients. 
> 
> But I thought that the 'eaddrinuse' error was per request, meaning that if a client gets an 'eaddrinuse' error, it won't crash and the client will continue to make requests. Maybe I was wrong here ? 

There are certain errors, where a client will abort it's session. So I would try to address those issues first and observe "finish_user_count". "finish_user_count" should be near to zero, since you are using a loop and want each client to stay until the tests is aborted after 30 minutes.



Best

Sebastian


> 
> On Thu, Aug 27, 2015 at 10:04 AM, Sebastian Cohnen <sebastian.cohnen at gmail.com <mailto:sebastian.cohnen at gmail.com>> wrote:
> Hey Emmanuel,
> 
> "finish_user_count" refers to users that have finished their session (in the last 10 sec and overall). This can be duo to the fact that the 
> 
> user successfully finished their session, but also if a user session is terminated because of errors. Did you check your tsung.log for errors?
> 
> 
> Best
> 
> Sebastian
> 
> 
>> On 27.08.2015, at 09:51, Emmanuel Peroumalnaïk <peroumalnaik.emmanuel at gmail.com <mailto:peroumalnaik.emmanuel at gmail.com>> wrote:
>> 
>> Thank you for the answer. 
>> 
>> I'm fully aware that the test was limited to 30 mins, but I thought that more users had finished their sessions since the finish_user_count was indicating  32_818 in the reports, which is almost the number of users created through the test. 
>> 
>> So this leads me to my second question : what does 'finish_user_count' means in this case ? 
>> 
>> On Wed, Aug 26, 2015 at 11:32 PM, Nicolas Niclausse <nicolas at niclux.org <mailto:nicolas at niclux.org>> wrote:
>> Emmanuel Peroumalnaïk a écrit le 26/08/2015 17:54 :
>> > Hello,
>> >
>> > I have a hard time figuring out the number of requests a test I ran should
>> > generate, and I was hoping you could help me out.
>> 
>> Hello
>> 
>> 
>> >   <load duration="30" unit="minute">
>> 
>> Here you are asking tsung to stop after 30mn, even if a lot of users have
>> not yet finished their sessions. So remove this.
>> 
>> cf.
>> http://tsung.erlang-projects.org/user_manual/conf-load.html#duration-of-the-load-test <http://tsung.erlang-projects.org/user_manual/conf-load.html#duration-of-the-load-test>
>> 
>> > The thing is that, with such configuration, I expected to have around
>> > 33_600 users x 2M requests, which should be about 67.2 billion requests.
>> 
>> That's quite a lot of requests. Even at 800K req/s, that should take a day
>> to finish. So 30mn is way too short.
>> 
>> > In the end, when I look at the reports, I have "only" 3_261_628 requests in
>> > the request count, which is way below what I expected. On top of that, my
>> > finish_user_count indicates 32_818, which means that almost all my users
>> > ran during the test.
>> >
>> 
>> 
>> --
>> Nicolas
>> 
>> 
>> 
>> -- 
>> Peroumalnaik Emmanuel
>> _______________________________________________
>> Tsung-users mailing list
>> Tsung-users at lists.process-one.net <mailto:Tsung-users at lists.process-one.net>
>> https://lists.process-one.net/mailman/listinfo/tsung-users <https://lists.process-one.net/mailman/listinfo/tsung-users>
> 
> 
> 
> 
> -- 
> Peroumalnaik Emmanuel

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.process-one.net/pipermail/tsung-users/attachments/20150827/095a6354/attachment.html>


More information about the Tsung-users mailing list