Everyone:
I figured since others have been reporting issues with
XP/Cygwin/PostgreSQL, I'd chime in with my own observations. (XP Pro
SP1, Cygwin 1.3.18-1, PostgreSQL 7.2.3-2, cygipc 1.13-2), and make sure
they (and any comments others add) get into the archives.
(1) I didn't have the problem some have reported with with ipc-daemon
--install-as-service; it worked fine for me.
(2) I have been having intermittent failures. It took hours to even
figure out how to replicate them, but I believe I have the pattern:
In XP Pro, more than one user can be logged in interactively (a user
can, rather than logging off, "Switch User", which leaves the first
user's processes running).
I have a privileged user (root) and an unprivileged user (me), in
addition to the postgres user. AFAICT, the second user to log on to
Windows cannot connect using psql (or seemingly by any other method).
So if a user logs into the system, then another user logs in, the
second can't connect.
This is irrespective of RUNAS, etc. -- if I'm logged in as root on the
GUI, I can connect as root, postgres, and/or me from that, uh, session
(I don't know the Windows term for this). If I then switch users
(without logging out of root), I can't connect as any user from the
other GUI session.
The message in the log under these circumstances is:
postmaster: StreamConnection: accept: Connection aborted
I also have an application which connects via JDBC running as a service
on this box, and it appears to display similar behavior, though I
haven't focused on using that application as a test case (I've been
using psql).
It's possible that I'll be able to come up with even more specific
detail on what triggers the problem, but these are my conclusions after
about 20 minutes of "confirmation" testing (logging in and out
continuously is tedious work, and I had of course considered and
discarded about 20 theories before even considering the possibility that
the user ID of the session might matter in any way).
Whether this is at the PostgreSQL level or the Cygwin level (or the
cygipc level), I have no idea (not much of a Windows/UNIX hacker
myself). This isn't a huge showstopper for me now that I know the
rules, though it will lead to minor inconvenience now and again.
If anyone wants me to run any other tests or check anything else, I'd be
glad to do so (I don't want to bore you with cygcheck or anything unless
there's some promising avenue down which someone wants to venture).
-- David.