https://bz.apache.org/bugzilla/show_bug.cgi?id=64631
Bug ID: 64631
Summary: Bug: JMeter master in CLI mode writes results to JTL
file twice (and the second time not all fields)
Product: JMeter
Version: 5.3
Hardware: PC
Status: NEW
Severity: major
Priority: P2
Component: Main
Assignee: [email protected]
Reporter: [email protected]
Target Milestone: JMETER_5.3.1
Created attachment 37375
--> https://bz.apache.org/bugzilla/attachment.cgi?id=37375&action=edit
JMX file
1: If running JMeter master in CLI (non-GUI) mode running a test with a JMeter
slave leads to the master writing all data twice to the results.csv/jlt file
(which introduces severe errors in the result), of course ... or it would be,
if there wouldn't be another sub-error to this:
2: In this case the first half of the results file contains correct data as it
seems with all fields filled. The second half (which shouldn't exist; see 1) of
the results file contains a copy of the first half, but with 4 fields less.
Missing fields (in my example): Encoding, SampleCount, ErrorCount, Hostname),
so you get a non-well formed CSV file (1st half with 21 fields, 2nd with 17
fields).
So the data is wrong and the file is corrupt.
Due to the fact that it only happens in CLI mode, I assume that the JMeter
slaves send correct data via RMI in both cases and that the JMeter master is
just writing it wrong, if started in CLI mode.
My setup:
OS: Windows10 Version 1809 Build 17763.1339
Java: Oracle JDK 1.8.0_261
JMeter 5.3
JMeter-relevant environment setup:
JMeter master (jmeter.bat), 1 JMeter slave (jmeter-server.bat), 1 apache httpd
webserver with a very simple test website ("WebTours" you might know it from
LoadRunner), 1 perfmon server agent; all on the same machine.
JMeter master is called from a batch file like this:
call %JMETER_HOME%\bin\jmeter.bat -n -t %JMX_FILENAME% -p user.properties -l
%OUTPUT_DIR%\results.csv -j %OUTPUT_DIR%\jmeter.log -r -X -H %PROXY_HOSTNAME%
-P %PROXY_PORT% -N %PROXY_NOPROXY%
All computer references used are a fully qualified hostname (because some
things don't work with localhost / 127.0.0.1; which is okay, of course).
The test has 10 threads with 2 iterations. My system runs this easily without
any kind of low ressources issues in a few minutes in the configuration
sketched above, so that's fine.
The issue happened every time I tested it (100% reproducible).
Workarounds (either a) or b); I will likely go with a) as long as this problem
exists) as long as this bug exists:
a) run JMeter master in GUI mode (as long as it's on a separate machine (which
it should be anyway) it shouldn't really influence your results, although the
JMeter User Manual recommends to use CLI)
b) manually have a close look at the middle of the results file and cut it in
half carefully at the right position and delete the 2nd part (and make sure
that all report/graph/... generation happens after that)
Attachment: JMX file used (hostname replaced with abc.def.ghi.jkl.mno ; the
rest is unchanged)
If you need additional information, not already in the ticket or attachments,
feel free to ask.
--
You are receiving this mail because:
You are the assignee for the bug.
