"Hiroshi Inoue" <Inoue@tpf.co.jp> writes:
> Now I realize that 7.1 already changed the handling of
> die interrupts fundamentally. For example we can't kill
> the backend which is in a trouble with an infinite loop.
> Was it an intended change ?
Doesn't bother me a whole lot; I don't think that's what the die
interrupt is for. In my mind the main reason die() exists is to
behave reasonably when the system is being shut down and init has
sent SIGTERM to all processes. Shared memory needs to be left in
a good state because the postmaster is going to try to run a
checkpoint. (Otherwise we'd just SIGQUIT all the backends.)
Therefore, it's more important to have a clean shutdown than to
have an instantaneous one.
regards, tom lane