On 12/09/2011 11:02 PM, Jerome Schulteis wrote:
>
> For bonus points, verify each Microsoft DLL as having a valid signature
> by double-clicking on it in the list, and report any DLLs that fail
> verification.
> All Microsoft DLLs verified. The following are all of the non-Microsoft non-.nls DLLs:
>
> iconv.dll LGPLed libiconv for Windows NT/2000/XP and Windows 95/98/ME Free Software Foundation 1.9.0.0
> libeay32.dll OpenSSL Shared Library The OpenSSL Project, http://www.openssl.org/ 0.9.8.12
> libiconv-2.dll LGPLed libiconv for Windows NT/2000/XP/Vista and Windows 95/98/ME Free Software Foundation
1.12.0.0
> libintl-8.dll LGPLed libintl for Windows NT/2000/XP/Vista and Windows 95/98/ME Free Software Foundation
0.17.0.0
> libxml2.dll
> postgres.exe PostgreSQL Server PostgreSQL Global Development Group 9.0.4.11104
> ssleay32.dll OpenSSL Shared Library The OpenSSL Project, http://www.openssl.org/ 0.9.8.12
> zlib1.dll zlib data compression library 1.2.3.0
Damn, scratch that theory. It's not a buggy hook DLL with some kind of race.
Maybe this exists in the standard Pg distribution and people don't
usually hit it because there aren't usually user logon/logoff events on
most database servers while they're doing work? Or people who've noticed
it have never bothered to track it down and isolate it until you did?
It sounds like you've tested this on a clean Windows install in a VM, yes?
The next step would be to figure out which DLL's InitInstance() or
ExitInstance() is failing, and how to get the postmaster to print the
formatted version of that error so we can see the substitutions in the
message string. I'm not able to take that on right now as I'm behind on
my main work and moving house as well.
Is anyone willing to volunteer to dig into this one?
--
Craig Ringer
