Gregory Stark <stark@enterprisedb.com> writes:
> Is there any way to tell, perhaps from the command string, that the process is
> about to start exiting? What stage of exiting is it that we think the kernel
> goes to lunch?
I haven't really done any detailed investigation, but I would think that
a simple process exit (when there's not temp tables to drop or anything
like that) should complete within one scheduler timeslice. That would
mean that when this problem occurs, it's usually because the kernel hasn't
scheduled the backend at all since the disconnect message was sent;
which in turn means there is no way at all to know that the backend is
going to exit when it does get a chance to run.
regards, tom lane