each failure leaves behind 1 stranded jconsole task



________________________________
From: bill lam <[email protected]>
To: Programming forum <[email protected]> 
Sent: Thursday, October 5, 2017 9:09 PM
Subject: Re: [Jprogramming] qrun - jcs - zmq



The mission of stress test is to make it fail and a large of task is
important, try on jconsole

qrun 99 99 1
or
2 qrun 99 99 2
and eventually
qrun each 500#<99 99 1

Any failure would mean it is unfit for serious production use.

I don't think the number of cores would affect its stability.

Did you check task manager for any stranded jconsole instances?


On Oct 6, 2017 8:43 AM, "'Pascal Jasmin' via Programming" <
[email protected]> wrote:

> with a separate program running on 6 cores,
>
> I can run in jqt without problem,
>
> qrun each 10 # < 99 5 3
>
>
> However, most (many at least) runs with more tasks, fail
>
> btw, your suggestions to use jconsole with ctrl-c apply just fine with jqt
> and jbreak.bat (and debug invoked at break)
>
> the logfile in ~temp, seems to just repeat the console output.
>
> There is a pattern to nearly all of the current failures:
>
> 1. It is hanging on terminating the last task "kill 98".  All runs always
> print "finished lastjob task", and hang on killing the task of the last
> finish. (not always the last job to finish last)
>
> there is no noticeable effect on success from adding an x parameter.
>
> ________________________________
> From: Eric Iverson <[email protected]>
> To: Programming forum <[email protected]>
> Sent: Thursday, October 5, 2017 4:43 PM
> Subject: [Jprogramming] qrun - jcs - zmq
>
>
>
> Pascal (and others interested in the qrun problem),
>
>
> I was happy when I was able to repeat the hang on my windows system. And
>
> then it went away. A race condition that depends on the weather?
>
>
> I have updated zmq/jcs addons with an improved qrun that logs more info.
>
>
> ctrl+c can be very useful in working with zmq. It is best to use jconsole
>
> in tracking down this problem. Jqt and JHS introduce unnecessary
>
> complications.
>
>
> Windows also complicates this as its support for ctrl+c has some problems
>
> vs zmq and sockets.
>
>
> In going over all the reports it seems that the problem is that one of the
>
> early tasks started never finishes its first request. The problem seems to
>
> be a race between starting the task and the first request to it.
>
>
> The new versions should help track this down.
>
>
> Please try the following and give back the results:
>
>
> 1. start jconsole
>
>    load'~addons/net/jcs/qrun.ijs'
>
>    qrun 99 99 1
>
>
> Poll now has a timeout. If you see poll line repeated every 5 seconds, you
>
> are likely hung waiting for something that isn't going to happen. The good
>
> news is that your session should respond to ctrl+c within 5 seconds.
>
>
> qrun now writes a logfile that might have some hints as to the problem.
>
> After qrun has hung, and you have done ctrl+c, take a look at: fread
>
> logfile_jcs_
>
>
> Please pass the contents of that file to me as it might hlep track this
>
> down.
>
>
> ***
>
> if it is a race between starting a task and sending it the 1st request, the
>
> problem might 'go away' if we add a sleep between starting all the tasks
>
> and starting any jobs. This is not a fix, but provides more info.
>
>
> If you can get the hang repeatedly, please see if you the following avoids
>
> the hang.
>
>
>    2 qrun 99 99 2 NB. sleep 2 seconds before starting requests
>
>
> ***
>
> Has anyone seen this problem on Linux? Can we say it is possibly a window
>
> only problem?
>
> ----------------------------------------------------------------------
>
> For information about J forums see http://www.jsoftware.com/forums.htm

> ----------------------------------------------------------------------
> For information about J forums see http://www.jsoftware.com/forums.htm
----------------------------------------------------------------------
For information about J forums see http://www.jsoftware.com/forums.htm
----------------------------------------------------------------------
For information about J forums see http://www.jsoftware.com/forums.htm

Reply via email to