Parpool with multiple MATLAB sessions

I have one MATLAB session running ga() with parpool. When I do simple operations like loading files and plotting in another MATLAB session, sometimes (though not all the time) the session with ga() running will say "Analyzing and transferring files to the workers ...done." in the command line, and then it errors out.

The error says:

"An UndefinedFunction error was thrown on the workers for 'errorMetric'. This might be because the file containing 'GA_errorMetric' is not accessible on the workers. Use addAttachedFiles(pool, files) to specify the required files to be attached."

The errorMetric is the fitness function used in ga, and specifying the file where the function is defined using addAttachedFiles does not solve the issue.

Answer 1

[TED MOSBY]

Hi,

The error message you're having can mean that the workers in your parallel pool cannot access the errorMetric function file needed for the genetic algorithm (GA) to evaluate the fitness function. This might happen if the file is not properly attached to the workers or if there are issues with file paths or synchronization between sessions. As you have not provided with the code files and other details I can suggest some general ways which by which you can validate and verify if you are using the parallel pool in the right way. Below are some steps you can try to resolve this issue:

1. Ensure Proper File Attachment

It's important to ensure that the file path is correct and that the file is accessible to all workers.

% Assume 'errorMetric.m' is the file containing your fitness function 
pool = gcp(); % Get current parallel pool 
addAttachedFiles(pool, {'path/to/errorMetric.m'}); 

2. Use Absolute Paths

Ensure that you are using absolute paths when specifying the file to avoid any ambiguity about the file's location.

3. Verify File Accessibility

Check that the file errorMetric.m is not open or locked by another process, which could prevent it from being accessed by the parallel workers.

4. Check MATLAB Path

Ensure that the directory containing errorMetric.m is included in the MATLAB path. You can add it using:

addpath('path/to/your/functions'); 

5. Synchronize File Changes

If you make changes to errorMetric.m while the parallel pool is running, you'll have to update the workers with the latest version.

• Restarting the parallel pool after changes:

        delete(gcp('nocreate')); % Close current pool 
        parpool; % Start a new pool 

• Re-attaching the file after changes:

        pool = gcp(); 
        addAttachedFiles(pool, {'path/to/errorMetric.m'}); 

6. Debugging

To diagnose the issue, try these steps:

• Check Worker Logs: Use spmd blocks to debug and check if the file is accessible on each worker

        spmd 
            if exist('errorMetric.m', 'file') == 2 
                disp('File is accessible on worker'); 
            else 
                disp('File is not accessible on worker'); 
            end 
        end 

• Use which Command: Verify the file path using the which command to ensure MATLAB is accessing the correct file.

        which errorMetric 

Let me know if this works for you!