Bug 57545 - Failed samples and aggregate report
Summary: Failed samples and aggregate report
Status: NEEDINFO
Alias: None
Product: JMeter
Classification: Unclassified
Component: Main (show other bugs)
Version: 2.12
Hardware: All All
: P2 enhancement (vote)
Target Milestone: ---
Assignee: JMeter issues mailing list
URL:
Keywords:
Depends on:
Blocks:
 
Reported: 2015-02-06 20:46 UTC by Vladimir Sitnikov
Modified: 2015-04-18 12:33 UTC (History)
1 user (show)



Attachments

Note You need to log in before you can comment on or make changes to this bug.
Description Vladimir Sitnikov 2015-02-06 20:46:05 UTC
I believe JMeter support of tracking failed samples can be
significantly improved.
Each and every manual suggests using "Aggregate Report" ([1]) or
friends, however it silently returns wrong data without any warning.

For instance: (sample, response time, status)
Sample1, 9 sec, OK
Sample1, 0 sec, ERR
Sample1, 0 sec, ERR

JMeter would show "average response time" as 3 seconds, and same for
percentiles (median should be 9 seconds, not 0).
However, it is common for failed requests to run much faster.
It does not matter very much how fast you can crash, but it does
matter how fast the successful responses are.

I see two problems here:
1) Default configuration averages/computes quantiles for both OK and
ERR responses. So users get wrong values in the report.
2) There is no easy way to track OK and ERR separately. Well, one can
add _two_ copies of AggregateReport (one for success, another one for
fails), however is that ever suggested in the documentation? Is that a
good user experience? One would have to switch back and forth from one
to another.

The easiest solution seems to change statistics key from "sampleLabel" to "sampleLabel,success", so successful and failure cases are computed in different buckets.
Comment 1 Philippe Mouawad 2015-02-10 22:51:34 UTC
Thanks for report.
Wouldn't it be better to ignore samplers in error for metrics related to response time ?
Although this could exclude the ones that timeouted which is a couter example of the reported bug.

Separating ok/ko as you propose, how would you compute error rate ? or did I misunderstand ?

Thanks