https://bz.apache.org/bugzilla/show_bug.cgi?id=63614

--- Comment #35 from [email protected] ---
(In reply to Felix Schumacher from comment #34)
> Interesting would probably be to have the exact time, when JMeter thinks it
> ended the test -- and thus closes the jtl file -- and the last entries in
> the jtl file. Are there entries with timestamps older than the ending time.
> 
> Most probably it is useful to attach a complete log and jtl file as noted
> already.
> 
> It might even be helpful to know a bit more about the test you are trying to
> run. Why do you think that it is not ending at the scheduled time? Can you
> take threaddumps shortly after the test run should have ended, but did not?

Assume that the load test is set to run for 600 seconds. The threads continue
to send requests until 599th second. The Requests that started at 599th second
would continue to run until it gets a response. The threads do not abruptly end
those reauests. After 600 seconds there will be no fresh requests but the
already started requests continue to run. The server, to which the requests are
sent, could take much time (the server maybe taking few seconds or minutes
based on load, or based on nature of process, or based on the http timeouts).
So the Master Jmeter can not guess when the slaves will be done. The Master
JMeter has to wait for two things before it starts working for dashboard
report.. First is to receive the message of  "I am done with all requests" from
all slaves. Second is the JTL file should be closed.

I would think to obtain the exclusive lock on the file for reading & writing. I
guess the thread(T1) that is writing into JTL File and the thread(T2) that is
generating dashboard report are different threads. Both the threads may have to
obtain exclusive locks before reading or writing thus ensuring no more than one
thread deals with the JTL file at the same time. I am going to send the JTL
file & Jmeter.log files(with INFO level) to email.

-- 
You are receiving this mail because:
You are the assignee for the bug.

Reply via email to