Re: [HACKERS] Possible explanation for Win32 stats regression

heh. I was just doing it the way Tom suggested - see attached. With a
little more trouble we could also keep track if the listened for events
and sometimes save ourselves a second call to WSAEventSelect, but I'm
not sure it's worth it.

cheers

andrew

korryd(at)enterprisedb(dot)com wrote:
>> Is anyone working on this?
>>
>> Tom Lane wrote:
>> > korry <korry(at)appx(dot)com <mailto:korry(at)appx(dot)com>> writes:
>> > > The problem is that, each time you go through
>> > > pgwin32_waitforsinglesocket(), you tie the *same* kernel object
>> > > (waitevent is static) to each socket.
>> >
>> > > The fix is pretty simple - just call WSAEventSelect( s, waitevent, 0 )
>> > > after WaitForMultipleObjectsEx() returns. That disassociates the socket
>> > > from the Event (it will get re-associated the next time
>> > > pgwin32_waitforsingleselect() is called.
>> >
>> > Hmm. Presumably we don't do this a whole lot (use multiple sockets) or
>> > we'd have noticed before. Perhaps better would be to keep an additional
>> > static variable saying which socket the event is currently associated
>> > to, and only issue the extra WSAEventSelect calls if we need to change
>> > it. Or is WSAEventSelect fast enough that it doesn't matter?
>> >
>>
>
> Here's a simple patch that fixes the problem (I haven't explored the
> performance of this patch compared to Tom's suggestion).
>
>