PCT : cannot cancel a running job without a PCT chrash

Hello

I think i have an infinite loop running in a worker. Each time i try to cancel it, PCT crashes. When i restart parallel pool with :

if isempty(gcp('nocreate'))
    p = parpool(1);
else
    p = gcp('nocreate');
end

and then i ask for jobs with :

p.Cluster.Jobs

which gives :

ans = 

 Job

    Properties: 

                   ID: 1
                 Type: concurrent
             Username: blafa
                State: running
       SubmitDateTime: 02-Apr-2019 14:49:24
        StartDateTime: 02-Apr-2019 14:49:33
     Running Duration: 0 days 0h 14m 47s
      NumWorkersRange: [1 1]

      AutoAttachFiles: true
  Auto Attached Files: List files
    AutoAddClientPath: true
        AttachedFiles: {}
      AdditionalPaths: 9 paths

    Associated Tasks: 

       Number Pending: 0
       Number Running: 1
      Number Finished: 0
    Task ID of Errors: []
  Task ID of Warnings: []
  

when i try to cancel it, PCT crashes :

 p.Cluster.Jobs.cancel
The client lost connection to worker 1. This might be due to network problems, or the interactive communicating job might have
errored.

Answer 1

[Edric Ellis]

When you run a parallel pool, PCT uses a parallel.Job behind the scenes to launch and co-ordinate the workers. By directly cancelling that Job, you're asking the PCT Cluster object to forcibly terminate all the worker processes. This causes the parallel pool session to abort, because the workers have been shut down. This is precisely what you're seeing here.

Could I ask: where is the actual problem you're encountering?

[Mikaël LE GRAND]

Hello Edric, thanks for your quicly answer.

I understand your comments. But the Job we talk about reappears each time i launch parallel pool, even after a reboot of my computer ! Like i said above, when i try to cancel it, parallel pool crashes. So, when i restart parallel pool and i refresh the Job Monitor, this damned job is still there, and so on.

I found no way to destroy it for ever... It continues to run, i think, because, shame on me, there is an infinite loop in the code of the worker. So, it never vanishes...

[Edric Ellis]

The job reappears because it is being used behind the scenes by the parallel pool. This is entirely normal - you should leave it running, and when the parallel pool is deleted, the job will be deleted.