Re: 7.4RC1 tag'd, branched and bundled ...

Jan Wieck wrote:
> Gaetano Mendola wrote:
> 
>> Tom Lane wrote:
>>
>>> "Marc G. Fournier" <scrappy@postgresql.org> writes:
>>>
>>>> On Tue, 4 Nov 2003, Gaetano Mendola wrote:
>>>>
>>>>> Is in the last Tom's patch about Vacuum sleep between pages  ?
>>>
>>>
>>>
>>>> that won't be in v7.4, to the best of my knowledge ...
>>>
>>>
>>>
>>> Definitely not.  It's a very experimental patch.
>>
>>
>> I not agree, is an experimental patch that introduce just
>> a delay, you now it better than me, and this delay can be
>> shipped with a default value 0.
>> Alias an experimental feature that can be disabled.
> 
> 
> Something that knowingly introduces portability issues and platform 
> dependant behaviour is absolutely inacceptable this late in the release 
> cycle where we already have a significant number of platform reports. 
> Configurable or not doesn't matter.

I see your point, anyway we'll see on the road how much people that know 
the existence of that patch will have in production a 7.4 + the patch.
May be an User Survey will be usefull ?
I spoke for my reality where with postgres we manage a service that
must be up and running 24/24 h 7/7 d.
( http://www.myopensky.com/os/OSwork.html )
I'm really stressed by the peoples that manage the service about
to do an Oracle migration, and the argumentation is always about the 
vacuum procedure that push down the db performances.

>>
>> I think that we are going to see a lot of 7.4 installation
>> with that patch applied.
> 
> 
> We are using usleep() and other equally risky functionality here to 
> quickly get something hooked together that confirms one theory or 
> another. We explicitly discourage people from attempting to squeeze this 
> sort of theory evaluation code into production installations.

I agree in general with you for these "general" arguments, but here we 
are talking about to introduce a sleep ( removable by guc ) or not! What 
about the hash refactoring introduced with 7.4? Are you going to
discourage people to use the hash?



Regards
Gaetano Mendola