On Tue, Oct 01, 2019 at 05:02:28PM -0700, Andres Freund wrote:
> The reason for that is simply that at that point llvmjit.c's own
> shutdown hook has already shutdown parts of the JIT subsystem (which
> needs to flush profiling information to disk, for profiling to be
> useful).
Hmm. I missed the actual point. The current location for the session
end hook has been chosen because we are sure that any transaction has
been aborted properly, and we'd still be limited with a hook in
proc_exit_prepare() because of that same argument. I am just going to
revert the patch.
--
Michael