I'd like to see the patch integrated into the main product when the
patch is ready -- that's been the goal all along. We've designed the
patch so that very little existing code needs to be changed.
But I'm still working on several problems in the patch:
1. When NT shuts down, standard Cygwin code catches a shutdown event
in each Cygwin process and generates a SIGHUP to itself. This
seems to be messing up clean shutdown of the PostgreSQL postmaster
and postgres processes. Since SIGHUP only causes the processes to
reread the PostgreSQL configuration file, I don't understand why
this is enough to screw up shutdown of the database, but it does.
I've tried a kludge to ignore SIGHUP in theses processes. This
results in clean database shutdown, but defeats the use of SIGHUP
to force rereading of the config file.
2. When the NT service for Postgres is stopped (individually) or
shutdown (along with all of NT), a new wrapper process sends
SIGINT to the postmaster process to trigger 'fast' shutdown. But
if there is a client connection still up, such as for psql, then
the postmaster and postgres processes continue running until the
client itself disconnects. This might be acceptable for a manual
stop, but not for shutdown where NT is going to kill the processes
after giving them a short time to shutdown gracefully. I plan to
change the shutdown case to send SIGQUIT for an 'immediate'
shutdown.
On Mon, May 07, 2001 at 04:25:59PM -0400, Bruce Momjian wrote:
> This sound pretty interesting. Any ideas if this should be integrated
> into the main tree?
--
Fred Yankowski fred@OntoSys.com tel: +1.630.879.1312
Principal Consultant www.OntoSys.com fax: +1.630.879.1370
OntoSys, Inc 38W242 Deerpath Rd, Batavia, IL 60510, USA
--ELM990197906-17697-1_--