Multiplexing SUGUSR1

I've been looking at the signal handling part of the synchronous
replication patch. It looks OK, but one thing makes me worry.

To set or clear the flag from PGPROC, to send or handle a signal, we
have to acquire ProcArrayLock. Is that safe to do in a signal handler?
And is the performance impact of that acceptable?

Another observation is that the patch introduces a new function called
SendProcSignal. Nothing wrong with that, except that there's an existing
function called ProcSendSignal, just above SendProcSignal, so there's
some potential for confusion. The old ProcSendSignal function uses the
per-backend semaphore to wake up a backend. It's only used to wait for
the cleanup lock in bufmgr.c. I'm tempted to remove that altogether, and
use the new signal multiplexing for that too, but OTOH if it works,
maybe I shouldn't touch it.

Attached is a patch with some minor changes I've made. Mostly cosmetic,
but I did modify the sinval code so that ProcState has PGPROC pointer
instead of backend pid, so that we don't need to search the ProcArray to
find the PGPROC struct of the backend we're signaling.

--
Heikki Linnakangas
EnterpriseDB http://www.enterprisedb.com