Let's discuss some realistic scenarios for an in-image low-space detection and handling mechanism. I think that, in many situations, users just expect some magic to the resolve a low-space issue. In reality, however, it can be rather tricky to recover and keep on working within such a too-big-to-handle image.

Is it a rogue workspace do-it that needs to be aborted and GC'ed?

Is it a server process that needs to cut off some erroneous client-request processing?

Do we have to save precious data to an external, out-of-image storage?

Do we expect some handling to then keep on working within the image?

...

When thinking about the interactive tooling that is required to handle a low-space situation, I find it hard to estimate the kind of memory we actually need for such interactivity...

As long as the VM does not grow further, the host environment is fine, I suppose. This is the status quo. The main issue seem to be our expectations of how the image behaves in such a situation...

We might even need a 75% warning besides a critical low-space detection...

Best,

Marcel