Thanks for the quick reply.
I completely agree that working with swap files in production is not a good idea.
Our usecase is completly development driven. We work with images / videostreams and Neuronal Networks on the Nvidia Jetson Nano. Since tensorflow is allocating a big chunk of the shared memory (gpu & cpu shared) our first naive approaches some times do not fit into the left memory of the jetson nano. Therefore the container gets killed by the balena engine. Since debugging and profiling our code is a bit hard, when it gets kill as soon as we start it, we would like to have an option to enable swap memory for debugging and profiling purposes.
But if you have any other ideas how to enable debugging and profiling, i am happy to try them out 