MATLAB: Is it possible to control the MATLAB Distributed Computing Engine from different user accounts on a network for Distributed Computing Toolbox

differentMATLAB Parallel ServerParallel Computing Toolboxschedulerusers

I have an MDCE setup by one user (user 1) on a network, and would like to be able to write output files to another user's shared directory (user 2) on the same network.

Best Answer

The ability to send output files from the MDCE user's account to another user's shared directory is available for Distributed Computing Toolbox 2.0 (R14SP3+) or later with the help of schedulers.

For previous versions, restrict the output directory's read access rights while maintaining write and list access rights for user 2, and maintaining full read and write permissions for user 1 as a workaround.

Related Solutions

MATLAB: Do I receive an error when attempting to start a worker for the MATLAB Parallel Server 1.0.1 (R14SP2)

This enhancement has been incorporated in Release 2007a (R2007a). For previous product releases, read below for any possible workarounds:

Note: There errors have changed as of MATLAB Distributed Computing Engine 2.0 (R14SP3+). See the Related Solutions listed below for more information.

These errors usually occur when one of the following is true:

1. The worker failed to checkout the license from the license server

2. Using ssh to connect to a worker node

3. The MATLAB Parallel Server was never installed on worker node

4. MDCE service was never started

Read below for possible causes and the solutions to above problems:

When you start a worker on any worker node, it sends a request to the network license manager to checkout the license. If a worker is unable to find the license manager or the license manager does not have the correct license for the worker, the Engine will fail to start a worker and will generate errors.

1. Check the worker log file located in /var/log/mdce or C:\TEMP\MDCE\log

- The log file will usually display any license manager errors or MATLAB startup error messages. These will need to be resolved before the worker will start. Look at the solutions related to the license manager error given in the worker log.

2. Check the mdce-service.log file located in /var/log/mdce or C:\TEMP\MDCE\log.

- In the event the worker log was not created, the service log should display some information as to the unsuccessful start of the worker.

3. Check the lmlog (license manager log) on the network license manager. This log will have information about the license usage. look for the error message related to the Distributed Computing engine.

4. Check the connectivity between the worker node and the license manager. Make sure that the worker can ping the license manager.

5. Start a worker on a headnode to verify that the license server is up and running and your license file is all correct.

6. After resolving issues with the license file, follow the steps below:

- Stop MDCE Service

- Delete all the log files from /var/log/mdce or C:\TEMP\MDCE\log

- Start MDCE Service

- Start the Job Manager

- Start the worker

If you are using ssh session to start a worker on Unix/Linux machine, read below:

To resolve this issue, you should instead log into the machine and perform an "su" command to login to the user account to start the mdce service.

This will also prevent the service from shutting down when you exit the ssh session.

1. Login to the machine:

ssh hostname

ssh user@hostname

2. "su" into the account to start mdce (usually root):

su root

3. Start the mdce service and the worker or job manager from this login session.

You should now be able to exit this session if needed, and the service will continue to run successfully.

If you are using ssh session to start a worker on MAC, Please read below:

There is a bug in the MATLAB Parallel Server 1.0.1 (R14SP2) when trying to start workers on a Macintosh via a remote ssh session. To work around this issue, try using an rsh session to run your remote startup script.

If you have installed only the Distributed Computing Toolbox and are attempting to start a worker, you will receive this error because the MATLAB Parallel Server is not installed. This must be installed first before calling these commands.

For more information about the Distributed Computing Toolbox and the MATLAB Parallel Server, please visit the Documentation page at the following link:

https://www.mathworks.com/help/parallel-computing/index.html

The worker will not start if the MDCE service has not been started before starting a worker. If this is the case, then follow the instruction below to start MDCE and jobmanager on the headnode and MDCE service on the worker node.

-To Check the status of MDCE:

Windows - Task Manager (mdced.exe) or Administrative Tools > Services

UNIX - $MATLAB/toolbox/distcomp/bin/mdce status

-To start the mdce service if it is not running:

Windows - From the MATLABROOT/toolbox/distcomp/bin/ directory

Type: mdce start

UNIX - $MATLAB/toolbox/distcomp/bin/mdce start

-To start a JobManager:

WINDOWS - $MATLAB\toolbox\distcomp\bin\win32\startjobmanager -name MyJobManager

UNIX - $MATLAB/toolbox/distcomp/bin/startjobmanager.sh -name MyJobManager

Note that if you have tried all of the above and are still not able to start the worker, then try to start MATLAB on the worker node to see if there is any problem with MATLAB startup (this will help us identify the cause of the problem). Remember, you need to have a MATLAB license to start MATLAB on a worker node.

NOTE: Starting in R2019a the following name changes occurred:

MATLAB Distributed Computing Server was renamed to MATLAB Parallel Server
mdce_def was renamed to mjs_def
mdce binary was renamed to mjs

MATLAB: How to configure a MATLAB worker in a cluster to have the security setting of a user who submits the job

By default, on Windows the workers run as a local system user, and on Linux and Mac OS X as root users. These users might have limited access to certain directories and network mounted drives. To change this setting you will need to modify the MDCE_USER property in mdce_def.bat file. You will need to set this property to a user who has read and write permissions to the group folder.

You will need to restart the MDCE daemon for the changes to take effect by executing the following commands:

 mdce stop
 mdce uninstall
 mdce install
 mdce start

Additionally, starting with MATLAB R2010b MATLAB Distributed Computing Server allows for a finer level of job security control, where each parallel job can be run as the user who submitted it to the cluster. To enable this, set the SECURITY_LEVEL to 3 in the mdce_def file. For more information on setting the job security level in MATLAB Distributed Computing Server please refer to the following documentation:

https://www.mathworks.com/help/releases/R2020a/matlab-parallel-server/set-matlab-job-scheduler-cluster-security.html

Best Answer

Related Solutions

MATLAB: Do I receive an error when attempting to start a worker for the MATLAB Parallel Server 1.0.1 (R14SP2)

MATLAB: How to configure a MATLAB worker in a cluster to have the security setting of a user who submits the job

Related Question