Local parallel cluster profile validation createJob takes a very long time to run

I am trying to validate my local parallel cluster profile on my system and when it gets to the step createJob and it runs for a very long time (> 30 minutes) (in MATLAB R2016b).

What might be some of the results why it seems to be taking so long?

When I type in

license('inuse')

It returns:

distrib_computing_toolbox
matlab

So it looks like that I am able to pull the license for the Distributing Computing Toolbox, but normally the createJob step in validating a local parallel cluster profile runs very quickly.

But now, it's either not running at all or it's taking > 30 minutes to run. (I've only been running it for half an hour so far, but there doesn't appear to be a "timeout" for that task where it will denote that task as failed to run.)

If anybody can provide some advice or guidance or maybe some suggestions for me to check, that would be greatly appreciated.

Thank you.

[Walter Roberson]

Did you post about this same thing earlier this week? I remember seeing a post with fairly similar circumstances.

[Ewen Chan]

Not that I'm aware of and/or not that I can remember.

I just looked at my questions and the answer is confirmed to be a 'no'.

[Walter Roberson]

https://www.mathworks.com/matlabcentral/answers/507238-parpool-does-not-open?s_tid=srchtitle is probably what I was thinking of

[Ewen Chan]

Yeah, I am not the author of that question.

Thank you.

Although I did try to start the parallel pool manually, and that failed, which is why I am trying something even more "basic" than that -- i.e. just running the validation test on the local parallel cluster profile and this appears to be failing as well.

Therefore; my working level assumption is that if I can't get the local parallel profile validated (and usually, it should validate very quickly), then pretty much anything and/or everything else after that is a bit of a moot point, if I can't even get through the validation step.

That's my current thought process.

[Raymond Norris]

Please contact Technical Support ([e-mail]) to resolve this.

[Ewen Chan]

Thank you.

I sent the email earlier this morning to support, and waiting for a response. (I understand that the normal turnaround time is a day, so that's ok. I understand that. Not a problem.)

Answer 1

[Raymond Norris]

Do you happen to have a local parallel pool already running, using all your cores, as you run your validation? I'm not sure when we added the check, but if you do, then MATLAB should have prompted you with the following dialog box

Although the parpool stage will fail, you could still run the validation and the createJob stage would hang.

What puzzles me is that you can call license while the validation is running, since validation blocks the MATLAB execution thread. Perhaps you ran that before validation, which in part lead me to think you might already have a parallel pool running.

By the way, validation for local should be unncessary. It's primarily for use with any other scheduler (e.g. MJS, Slurm) since there we're trying to connect through to firewalls, ports, etc If my observation above doesn't address your issue (not likely), I'd suggest contacting Technical Support ([e-mail]).

[Ewen Chan]

No, I don't have a local parallel pool running on the system.

I rebooted the computer, and started up my MATLAB again (after logging back into work's VPN - you know, because of the whole work-from-home thing), and started trying to validate local cluster profile once MATLAB was up.

parcluster passed.

I do have the Job Monitor running, which, my understanding is that, that will pull a distributed/parallel computing toolbox license.

But it has also been my experience (before) where that shouldn't affect my ability to validate the local cluster profile.

I am running this because I want to see if the problem is on my end or whether it's on our server/license end (which falls under our IT domain/realm).

Thank you.