My parallel worker fall back to 3 workers when running at 4 workers

I was running a my code using parpool(4) which using 4 workers. However, I noticed after the execution of my code(having each worker writing to their own files), the workers falls back to 3. What could be the reason behind this?

[Walter Roberson]

What are you examining that leads you to suspect that it is falling back to 3 workers?

Is work being done on the 4th worker for part of the time?

[Yi Xien Yap]

Yes! The work actually done on the 4th worker for part of the time, and then it is falling back to 3 worker. I tried to shutdown the pool and recreate the pool again, and same thing happen again.

[Walter Roberson]

Okay, what are you observing that leads you to say that the 4th worker stops being used?

[Yi Xien Yap]

I first created 4 files, and parpool(4) and I used getCurrentTask().ID to get the worker ID in the parfor loop to store the file in their unique files. I then open the file that each of the workers saved, I realize one worker is not saving to its unique files, and the workers has turn into 3.

[Walter Roberson]

Possibly one of the workers encountered an error ?

[Yi Xien Yap]

How do you check if it encountered an error or not?

[Yi Xien Yap]

I created a parpool(4), that working fine until I run it, then fall back to 3, it happen occasionally.

Then I check, it doesnt show any error

[Walter Roberson]

@Raymond Norris have you observed this in the past, number of workers dropping even though no tasks have errors and no tasks are finished ?

[Raymond Norris]

It looks familar. It is possible for a pool to drop a worker and still stay open, but MATLAB would have notified you

An unexpected error occurred during PARFOR: Error in remote execution of parfor: Java is shutting down

Sounds like there's some frailty in the parpool if no user code is explicitly erroring out. I would suggest contacting Support ([e-mail]).

Answer 1

[Farouk]

I've had this issue before. Its a memory problem! Are you processing large dataset?