jmeter-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From sebb <seb...@gmail.com>
Subject Re: Bug 62388 - Invalid SampleResult breaks Dashboard graphs that depend on Time
Date Sun, 27 May 2018 10:22:53 GMT
On 27 May 2018 at 10:08, Philippe Mouawad <philippe.mouawad@gmail.com> wrote:
> Hello,
> Should we do something for this one:
> I am personally mitigated :
>
>    - It would be a plugin issue here
>    - but users might think it's a jmeter bug
>
> Possible options:
>
>    - Ignore row ? : Not clean
>    - Fail on wrong row indicating the line number, user would decide if he
>    removes it and regenerates
>    - Other ideas ?

SampleResult#sampleEnd already logs an error if called twice.
It could also log an error if startTime is 0.

Could extend the checks to report other incorrect call sequences.

Maybe even have a simple validate method that is called just before
storing the sample result.

I think this would make it clear that the problem is with the sampler.

>
> ------------------------------------------------------------------------------------------------
>
> If for some reason a plugin generates a SampleResult where sampleStart
> was not called, timeStamp column will be 0 in CSV.
>
> This will break the report leading to weird graphs.
>
> Shouldn't there be a protection in JMeter against this by ignoring
> sampleresult with warning/error ?
>
> In the case we face we have 4 samples in this case for 200000 correct
> sampleResults, as a consequence the graph is corrupt.
>
> ------------------------------------------------------------------------------------------------

It's not possible in general for JMeter to fix the value after the event.
Whereas a human may be able to adjust the records based on surrounding samples.
So I think logging is the best that can be done.

The graph code could do the same, and log invalid samples.
If SampleResult has an isValid() method, this would be easy to do.
Or optionally ignore them.

Given that this is a bug, I don't think we *have* to do anything.
But it might be sensible to improve the error checking as noted above
to log when samples are created incorrectly.

Maybe also add comments to the documentation to note that graphs will
misbehave if given bad input (giving the example of missing start
time)

> Thanks
> Regards

Mime
View raw message