a bit more precise MaxOffsetNumber

Hi,

I've been digging in the sources, and I've noticed the MaxOffsetNumber
is defined (in storage/off.h) like this

(BLCKSZ / sizeof(ItemIdData))

I guess it might be made a bit more precise by subtracting the header
like this

(BLCKSZ - offsetof(PageHeaderData, pd_linp) / sizeof(ItemIdData))

although the difference is negligible (2048 vs 2042 for 8kB pages).

Tomas

2011/4/30 Tomas Vondra <tv(at)fuzzy(dot)cz>:
> I've been digging in the sources, and I've noticed the MaxOffsetNumber
> is defined (in storage/off.h) like this
>
>  (BLCKSZ / sizeof(ItemIdData))
>
> I guess it might be made a bit more precise by subtracting the header
> like this
>
>  (BLCKSZ - offsetof(PageHeaderData, pd_linp) / sizeof(ItemIdData))
>
> although the difference is negligible (2048 vs 2042 for 8kB pages).

I guess we could do that, but I'm not sure there's much point. It's
also not entirely clear that this would actually work out to a win,
because of the issues discussed in the "When can/should we prune or
defragment?" section of src/backend/access/heap/README.HOT

We could probably figure this out with some careful testing, but I'm
not sure it's worth the effort.

--
Robert Haas
EnterpriseDB: http://www.enterprisedb.com
The Enterprise PostgreSQL Company

Dne 7.5.2011 04:02, Robert Haas napsal(a):
> 2011/4/30 Tomas Vondra <tv(at)fuzzy(dot)cz>:
>> I've been digging in the sources, and I've noticed the MaxOffsetNumber
>> is defined (in storage/off.h) like this
>>
>> (BLCKSZ / sizeof(ItemIdData))
>>
>> I guess it might be made a bit more precise by subtracting the header
>> like this
>>
>> (BLCKSZ - offsetof(PageHeaderData, pd_linp) / sizeof(ItemIdData))
>>
>> although the difference is negligible (2048 vs 2042 for 8kB pages).
>
> I guess we could do that, but I'm not sure there's much point. It's
> also not entirely clear that this would actually work out to a win,
> because of the issues discussed in the "When can/should we prune or
> defragment?" section of src/backend/access/heap/README.HOT
>
> We could probably figure this out with some careful testing, but I'm
> not sure it's worth the effort.

No idea if it's worth the effort and if something can be broken by this
change. I've noticed this when trying to implement a more thorough check
of table contents vs. index. I was asking 'how many items can be stored
on a page' and I've noticed this.

Anyway I don't see how this could affect HOT updates? AFAIK this has
nothing to do with evaluation of free space on a page. It only affects
checking if the offset number may be valid (OffseNumberIsValid macro),
and code that needs to keep info about items on a page, as it usually
does something like this:

OffsetNumber unused[MaxOffsetNumber];

But yes, the amount of memory saved is negligible (12B) so the only
noticeable benefit might be catching some invalid offset numbers.

regards
Tomas