Memory leak of parallel workers

I have a code using the local cluster for the execution of functions using parfeval. I was using this code on version 2018a and there was no problem. After I installed 2021a the same code (on the same machine) changed its behaviour.

Now it looks like there is a memory leak during parallel executions. After each evaluation, the used RAM of each worker grows and grows until it crashes. Speed of the growth suggests that large volumes of memory are being gathered in each worker. This leaking is not observed in 2018a version.

I tried to search for how to manage or monitor memory on the workers but I found nothing useable. Also, I never encountered a memory leak in MATLAB and I am unsure how to proceed.

Is there a documented change between versions 2018a-2021a that concerned memory management of parallel workers? Does someone have any experience with similar behaviour? What to do?

[Frantisek Gaspar]

As a follow-up, after the windows update the issue disappeared. I guess that the problem was in windows memory handling and not in MATLAB.