Re: pgstat wait timeout (by Robert Schnabel)

On 16-12-2009 16:28, Alvaro Herrera wrote:
> Alexey Luchko wrote:
>
>> We've tried to restart the service. It was looking like the service
>> stopped successfully, but two postgres.exe processes left running.
>> One of them was utilizing one core. It had no tcp connections in any
>> state. It was
>> "C:/Program Files/PostgreSQL/8.4/bin/postgres.exe" "--forkcol" "1228"
>>
>> sysinternals' procexp.exe showes the following stack for this process:
>> """
>> ntkrnlpa.exe+0x8dace
>> ntkrnlpa.exe+0x29a62
>> ntkrnlpa.exe+0x33178
>> hal.dll+0x6199
>> hal.dll+0x63d9
>> hal.dll+0x6577
>> hal.dll+0x3902
>> ntkrnlpa.exe+0x89721
>> ntdll.dll!KiFastSystemCallRet
>> mswsock.dll!ServiceMain+0x2949
>> mswsock.dll!NSPStartup+0x1893
>> WS2_32.dll!WSAUnhookBlockingHook+0x486
>> WS2_32.dll!WSAUnhookBlockingHook+0x660
>> WS2_32.dll!WSAEventSelect+0x2f
>> postgres.exe!mdpostckpt+0x181b38
>> postgres.exe!mdpostckpt+0x18bbdb
>> """
>
> This stack trace seems bogus.  The stats collector should not be calling
> mdpostckpt at all.

It seems that the process left and the stack trace is from the process
waiting for the stats collector.

--
Alexey