On 16/04/10 16:23, A. Kretschmer wrote:
> In response to Raymond O'Donnell :
>> On a related note, what happens when you do something like this? -
>>
>> select count(*) ....
>>
>> Does any data actually get read?
>
> No, it check's only the visibility for each record -> seq-scan.
... though in practice with OS and disk readahead this probably means
all the data actually gets read from disk, though PostgreSQL doesn't
have to process all of it.
I sometimes wonder if being able to store visibility info externally to
a tuple in a separate file - in condensed fixed-width form - would be
useful for performance, especially where the table has quite wide tuples
with types that are big-ish but not TOASTable. Sure, it'd be more disk
seeking but OTOH it'd be more likely to stick around in cache, could
even be put on other storage, etc.
I suspect that even testing the notion out would involve ripping out and
rewriting half of Pg's guts, though, so it's pretty much hot air anyway.
--
Craig Ringer