jmeter-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Milamber <milam...@apache.org>
Subject Re: jmeter.properties cleanup
Date Thu, 09 May 2013 09:10:09 GMT

Le 09/05/2013 09:32, Philippe Mouawad a ecrit :
> On Thursday, May 9, 2013, Milamber wrote:
>
>>
>> Le 09/05/2013 08:16, Philippe Mouawad a ecrit :
>>
>>> On Thursday, May 9, 2013, Milamber wrote:
>>>
>>>   Le 08/05/2013 23:03, Philippe Mouawad a ecrit :
>>>>   Bad post, should have gone here
>>>>> ---------- Forwarded message ----------
>>>>> From: *Philippe Mouawad*
>>>>> Date: Wednesday, May 8, 2013
>>>>> Subject: jmeter.properties cleanup
>>>>> To: JMeter Users List <user@jmeter.apache.org>
>>>>>
>>>>>
>>>>> Hello,
>>>>>
>>>>> jmeter.properties has grown with a lot of properties that maybe are not
>>>>> that useful.
>>>>> I find it a good thing that lot of things are configurable in JMeter
but
>>>>> maybe it's too much and one of the issues is users may not find the
>>>>> really
>>>>> useful ones (recently for example with https.socket.protocols).
>>>>>
>>>>> I propose to remove the following:
>>>>>
>>>>>       - jmeter.loggerpanel.display=****false => It's so easy to just
>>>>> click it
>>>>>       - jmeter.errorscounter.display=****true => Why would someone
not
>>>>> want
>>>>> this
>>>>>       feature ?
>>>>>       - jmeter.toolbar.display=true => Why would someone not want
this
>>>>> cool
>>>>>       feature ?
>>>>>       - jmeter.toolbar => Will users really want to reorganize these
>>>>> icons ?
>>>>>       - jmeter.toolbar.icons => Same as before
>>>>>
>>>>>   If you are a JMeter plugins developer, you may want to re-organize
or
>>>> change the toolbar.
>>>>
>>>>        - onload.expandtree => Current default behaviour seems fine no
?
>>>>
>>>>>       - jmeter.save.saveservice.****autoflush => After some further
>>>>> thinking, why
>>>>>       would users not need this one ? If JMeter crashes and some data
is
>>>>> lost ,
>>>>>       then there are big chances that the test was not that fine before
>>>>> the
>>>>> crash.
>>>>>
>>>>>   No! I prefer (and I put) this property to the value "true" ! If you
>>>> make a
>>>> simple load test and we stop the test with a Ctrl-C, we lost a lot of
>>>> results (with some tests in my case, I've lost the *entire* results
>>>> (small
>>>> test of 5-10 min). Please don't touch this property, and I recommend to
>>>> put
>>>> to true by default. It's a very annoying behavior.
>>>>
>>>>
>>>> We could introduce a  shutdown hook to handle these Ctrl+C cases.
>>>>
>>> In my opinion it should be false as performances for high throughput tests
>>> are way better. And imho default settings should  be the most performing
>>> for a load testing tool  no ?
>>>
>> Not sure. A new JMeter user can be disoriented if he don't see his results.
>> I've prefer a reliable software that a more performance software which can
>> loose my results.
> With shutdown hook you won't loose results as quit signal will be trapped
> and we can flush the file, no ?
>
>> Perhaps, make this autoflush behavior more visible in JMeter UI. For
>> example, add a checkbox in Test Plan (or a checkbox-menu) to enable/disable
>> this option. (and add a warning message when the option is enabled : "you
>> can lost some results if you stop the test with Ctrl-C")
>>
>>
>> Do you think it s still needed with what  I described above ?
> What was the scenario that made you lost some resuts ?


A simple scenario :

Thread group with 1 / 1 / infinite loop
      |-- Java Sampler with 1000 ms delay


Launch JMeter in non-gui mode, with 30 sec and Ctrl-C :

milamber@ender:~/opt/apache-jmeter-2.10-SNAPSHOT/bin$ ./jmeter -n -t 
./lost-results.jmx -l myresults.csv
Creating summariser <summary>
Created the tree successfully using ./lost-results.jmx
Starting the test @ Thu May 09 10:05:23 WEST 2013 (1368090323095)
Waiting for possible shutdown message on port 4445
summary +      7 in     8s =    0.9/s Avg:  1076 Min:  1005 Max: 1162 
Err:     0 (0.00%) Active: 1 Started: 1 Finished: 0
summary +     27 in  30.2s =    0.9/s Avg:  1117 Min:  1023 Max: 1251 
Err:     0 (0.00%) Active: 1 Started: 1 Finished: 0
summary =     34 in    38s =    0.9/s Avg:  1108 Min:  1005 Max: 1251 
Err:     0 (0.00%)
^C
milamber@ender:~/opt/apache-jmeter-2.10-SNAPSHOT/bin$ wc -l myresults.csv
0 myresults.csv
milamber@ender:~/opt/apache-jmeter-2.10-SNAPSHOT/bin$ cat myresults.csv 
<=== empty file
milamber@ender:~/opt/apache-jmeter-2.10-SNAPSHOT/bin$

Not good in my opinion.

Milamber




>
>
>
>
>>
>>> Ok for the rest let's keep the statu quo if you think all are needed.
>>>
>>>   I have doubts about those ones:
>>>>> # Netscape HTTP Cookie file
>>>>> cookies=cookies => What does it do ?
>>>>>
>>>>> We could try to remove them and if users want them, we would have some
>>>>> bugzilla request to get them back.
>>>>>
>>>>>   If you remove these properties, you introduce a lot of incompatibilty
>>>> changes and (in my opinion) you remove some freedom of the user's
>>>> preferences. Please double check before remove.
>>>>
>>>> Milamber.
>>>>
>>>>
>>>>
>>>>
>>>>


Mime
View raw message