Tomas Vondra <tomas.vondra@2ndquadrant.com> writes:
> FWIW this is where the view listing dictionaries loaded into shared
> memory would be helpful - you'd at least know there's a dictionary,
> wasting memory.
Well, that's only because we failed to make the implementation transparent
:-(. But it's not unlikely that an mmap-based implementation would be
simply incapable of supporting such a view: the knowledge of whether a
particular file is mapped in would be pretty much process-local, I think.
So I'd really rather we don't add that.
Also, while these dictionaries are indeed kind of large relative to our
traditional view of shared memory, if they're in DSM segments that the
kernel can swap out then I really suspect that nobody would much care
if a few such segments had been leaked. I find it hard to imagine a
use-case where DROP race conditions would lead us to leak so many that
it becomes a serious problem. Maybe I lack imagination.
regards, tom lane