R2017a parfor error when starting parpool inside function body

Hello,

I'm testing the work of the distributed cluster MATLAB R2017a (in practice I use R2015b). MDCE components runs on Linux machines (Ubintu and Mint), client is Linux machine too. I have some problems in the new version, which did not arise in the previous one:

If I run the cluster (named MJSP17 in example) inside the body of a simple function " main.m" that uses a table (with three columns) as input arguments, I get an error in accessing variables:

Warning: While loading an object of class 'table':
Invalid default value for property 'propertyNames' in class 'table':
Error: File: metaDim.m Line: 246 Column: 20
Arguments to IMPORT must either end with ".*" or else specify a fully qualified class name: "matlab.internal.tabular.private.tabularDimension.checkReservedNames_impl" fails this test.
> In parallel.internal.pool.deserialize (line 9)
  In parallel.internal.pool.deserializeFunction (line 12)
  In remoteParallelFunction (line 33)
Error using main (line 28)
Struct contents reference from a non-struct array object.

1. Script that calls main.m:

% Generate filelist
gcpfilelist = {[cd filesep 'main.m']};

% Make an example of "data" table: | ID | PeriodID | Value |
numid     = 1000;                    % Number of unique ID
maxlength = 74;                     % Maximum number of Period for every ID
varlength = randi(maxlength, numid, 1) + 1;
data      = table(repelem((1:numid), varlength)', ...
            cell2mat(arrayfun(@(x) (1:x)', varlength, 'UniformOutput', false)),...
            randi(10^6, sum(varlength), 1), 'VariableName', {'FcID' 'PeriodID' 'Value'});

% Call main finction
main( data, gcpfilelist );

2. main.m example:

function [ batchoutput ] = main( data, gcpfilelist )

unqid       = unique(data{:,'FcID'}); 
n           = numel(unqid);
batchoutput = cell(n, 1);

%%Make main processing (using functions from project folders)

parpool('MJSP17', 'SpmdEnabled', false); 

% addAttachedFiles(gcp, gcpfilelist);
% fprintf(1, '============== ATTACHED GCP FILES =================\n');
% curgcp = gcp; curgcp.AttachedFiles
% fprintf(1, '===================================================\n');

parfor j = 1:n %numel(metascenariofc.FcID)

      curid           = unqid(j);
      curFcIDdata     = data(data.FcID == curid, :);
      batchoutput{j}  = curFcIDdata;

end

delete(gcp('nocreate'));

% Completeness check
any(cellfun(@(x) isempty(x), batchoutput))

end

The problem is solved by the transfer list of project's working files ( gcpfilelist) to the gcp.

3. Uncomment in main.m:

addAttachedFiles(gcp, gcpfilelist);
fprintf(1, '============== ATTACHED GCP FILES =================\n');
curgcp = gcp; curgcp.AttachedFiles
fprintf(1, '===================================================\n');

But, this is extremely inconvenient, and besides, this error did not arise in R2015b.

Tell me, please, how to avoid the occurrence of this error without sending the structure of project files to all nodes of the cluster?

Thank you!

[Walter Roberson]

possibly adding the jar to the static class path might work; https://www.mathworks.com/matlabcentral/answers/101261-why-do-i-get-java-class-import-errors-when-compiling-a-deployed-application-using-matlab-compiler-4

[Edric Ellis]

The error you're getting appears to be coming from MATLAB's built-in type table - which seems strange to me. Could you check that you're using precisely the same version of MATLAB for both client and workers?

[Jonas]

I call:

pctRunOnAll ver MATLAB

Results on all workers are the same:

MATLAB Version: 9.2.0.... (R2017a)