Re: [HACKERS] include/config.h FOLLOWUP - Mailing list pgsql-hackers

From Bruce Momjian
Subject Re: [HACKERS] include/config.h FOLLOWUP
Date
Msg-id 481b3e60e0421279e9f34469f26f51bc
Whole thread Raw
List pgsql-hackers
I believe it is 8k to match the base size for the filesystem block size,
for performance.

>
> I just wrote about issues regarding upping the max blocksize.  There is
> stuff in storage/itemid.h that would need to be modified (also in
> bufmgr.h, but that is trivial).  Th idea is that:
>
> typedef struct ItemIdData
> {
>                                                   /* line pointers */
>         unsigned        lp_off:13,              /* offset to find tup */
>         /* can be reduced by 2 if necc. */
>                                 lp_flags:6,             /* flags on tuple */
>                                 lp_len:13;              /* length of
> tuple */
> } ItemIdData;
>
> the  ItemIdData.lp_len is maxed at 2^13, or 8192.  My query then is:
> which of the items is it saying 'can be reduced by 2 if necc'?  For, if
> that is accurate, I'd like to e.g. lp_off:11 and lp_len:15, allowing for
> tuples of 32768 or something.  And also not breaking the nice 32 bit
> length of the struct.
>
> Lemme know...
>
> ps. why not allow for larger tuples in general?  Do we take a speed hit?
>
>


- --
Bruce Momjian
maillist@candle.pha.pa.us

------------------------------

pgsql-hackers by date:

Previous
From: Bruce Momjian
Date:
Subject: Re: [QUESTIONS] tuple is too big?
Next
From: Bruce Momjian
Date:
Subject: Aggregate fix