That definitely is an architectural issue. Unfortunately it is essentially a cache management matter (cache invalidation considers when something becomes useless/wasteful, as well as when it becomes incorrect) which is a hard problem to solve.
There are legitimate circumstances where a user causes a long running prices to start, that it is wasteful to continue if incomplete when they vanish (because the result isn't going to be useful to another user, or even that same user if they return).
While you can't catch everything (the user's machine losing power or all connectivity, etc.) you can catch a deliberate exit. And in the case of a long running process, if the user's exit was accidental it is more likely that they'll be back and having continued the process is useful.
That definitely is an architectural issue. Unfortunately it is essentially a cache management matter (cache invalidation considers when something becomes useless/wasteful, as well as when it becomes incorrect) which is a hard problem to solve.
There are legitimate circumstances where a user causes a long running prices to start, that it is wasteful to continue if incomplete when they vanish (because the result isn't going to be useful to another user, or even that same user if they return).
While you can't catch everything (the user's machine losing power or all connectivity, etc.) you can catch a deliberate exit. And in the case of a long running process, if the user's exit was accidental it is more likely that they'll be back and having continued the process is useful.