On 17/08/2009 4:00 PM, Michael Bell wrote:
> Oh, tried this with Windows firewall off, and AV off too. Here's the logging of the last few successful, and then the
unsuccessful....
It might be useful if you could:
- in pg_hba.conf, set 127.0.0.1 to "trust" or change the password
of the user you'll be testing with to something generic; then
- use Wireshark (http://wireshark.org/) to capture traffic on the
loopback interface during a failed login attempt that results in
the "incomplete startup packet" error.
- Change pg_hba.conf back / reset the password you changed
- Post the packet trace (libpcap format) or, if it's too big, put
it up somewhere people can download it and post a link to it.
Doing so may or may not help - but if the results look normal you at
least know the client is sending the right startup packet and it's a
server side problem.
"An operation was attempted on something that is not a socket" makes you
wonder if the socket was closed, the backend is getting FDs (handles)
mixed up, or something. I was about to suggest trying to get a log with
Process Monitor (from Sysinternals) but it doesn't look like Process
Monitor records TCP/IP socket activity, so it won't be much use. Sigh.
Maybe the wireshark trace will be informative...
--
Craig Ringer
> 2009-08-17 00:56:58 PDTLOCATION: BackendInitialize, .\src\backend\postmaster\postmaster.c:3205
> 2009-08-17 00:56:58 PDTLOG: XX000: could not receive data from client: An operation was attempted on something that
isnot a socket.
> 2009-08-17 00:56:58 PDTLOCATION: pq_recvbuf, .\src\backend\libpq\pqcomm.c:769
> 2009-08-17 00:56:59 PDTLOG: 08P01: incomplete startup packet
> 2009-08-17 00:56:59 PDTLOCATION: ProcessStartupPacket, .\src\backend\postmaster\postmaster.c:1469
>
>
>
>
>