MATLAB: “Warning: Unable to load Toolbox Path Cache” during parallelization

parallel computing

I am running multiple instances of C-code in parallel on Ubuntu. The C-code starts up matlab, calls a script, then closes. When running many at once (~8), I get the following error intermittently:

Warning: Unable to load Toolbox Path Cache
/home/chuck/.matlab/R2018a/toolbox_cache-9.4.0-2328427802-glnxa64.xml. The
cache file is not in the correct format. It will be regenerated upon quitting
MATLAB.

I have found that unchecking "Update Toolbox Path Cache" per this post will prevent the warning from occuring, but that it slows down each Matlab call.

Is there a way to prevent this warning other than not using the cache?

Best Answer

If the path cache is updated by multiple threads, it is likely, that one thread tries to read the file, while it is re-created by another thread. This is a standard problem of multi-threading. This ould be solved, if Matlab creates a file lock before one threads accesses the file, but as long as this is not implemented, disable the updates of the Toolbox Path Cache.

Related Solutions

MATLAB: Do I receive warning messages about the Toolbox Path Cache when starting MATLAB

The Toolbox Path Cache is a file that MATLAB uses to store information regarding the location of toolbox directories. It is used because loading the information from a single local file is quicker than accessing the file system, particularly when there are many toolboxes or the directories are being accessed across a network. The cache file will also be updated at the end of each MATLAB session. More information regarding the toolbox path cache is available here:

http://www.mathworks.com/help/matlab/matlab_env/toolbox-path-caching-in-the-matlab-program.html

Both error messages indicate that there is an issue with the current Toolbox Path Cache file. Generally both errors will be resolve by restarting MATLAB as it will generate a new Toolbox Path Cache. If the error does return there are two ways to reset the Toolbox Path Cache

1. Using the command:

>> rehash toolboxcache

2. From the MATLAB Desktop:

Open the Preferences GUI. For MATLAB 7.14 (R2012a) and earlier releases go to the File menu to Preferences. For MATLAB 7.14 (R2012b) and later releases go to the Home tab to Preferences. In the Preferences GUI, select the "General" tab. Click the "Update Toolbox Path Cache" button to reset the Toolbox Path Cache.

After the Toolbox Path Cache is reset, the error should not return on startup.

MATLAB: Concurrent runs of compiled standalone executable fail with “Could not access the MCR component cache”

Hi Costas,

usually simultaneous calls of compiled applications should not interfer with each other. Otherwise it would not work at all and crash the second call.

I must admit, I don't have a definite solution to your problem :(, your setup with many concurrent calls is probably less common ... ;-).

One thing that comes to my mind is the cache size: maybe increasing it could help? Call

winopen(mcrcachedir)

to open the cache folder. Edit the .max_size file and replace the 32 MByte (33554432 bytes) by some larger value, e.g. 64 MByte.

An alternative would be to not embed the .ctf file into the application (adding parameter -C). This way the ctf is extracted locally instead of somewhere in the user profile.

If you need a more sophisticated (and probably faster) way of doing such calls, an alternative would be to use the MATLAB Production Server as a server for executing MATLAB based computations ...

Titus

Best Answer

Related Solutions

MATLAB: Do I receive warning messages about the Toolbox Path Cache when starting MATLAB

MATLAB: Concurrent runs of compiled standalone executable fail with “Could not access the MCR component cache”

Related Question