Re: pageinspect patch, for showing tuple data

В письме от 3 августа 2015 15:35:23 пользователь Alvaro Herrera написал:
> Nikolay Shaplov wrote:
> > This patch adds several new functions, available from SQL queries. All
> > these functions are based on heap_page_items, but accept slightly
> > different arguments and has one additional column at the result set:
> >
> > heap_page_tuples - accepts relation name, and bulkno, and returns usual
> > heap_page_items set with additional column that contain snapshot of tuple
> > data area stored in bytea.
>
> I think the API around get_raw_page is more useful generally. You might
> have table blocks stored in a bytea column for instance, because you
> extracted from some remote server and inserted into a local table for
> examination. If you only accept relname/blkno, there's no way to
> examine data other than blocks directly in the database dir, which is
> limiting.
>
> > There is also one strange function: _heap_page_items it is useless for
> > practical purposes. As heap_page_items it accepts page data as bytea, but
> > also get a relation name. It tries to apply tuple descriptor of that
> > relation to that page data.
>
> For BRIN, I added something similar (brin_page_items), but it receives
> the index OID rather than name, and constructs a tuple descriptor to
> read the data. I think OID is better than name generally. (You can
> cast the relation name to regclass).
>
> It's easy to misuse, but these functions are superuser-only, so there
> should be no security issue at least. The possibility of a crash
> remains real, though, so maybe we should try to come up with a way to
> harden that.

I've done as you've said: Now patch adds two functions heap_page_item_attrs
and heap_page_item_detoast_attrs and these function takes raw_page and
relation OID as arguments. They also have third optional parameter that allows
to change error mode from ERROR to WARNING.

Is it ok now?

--
Nikolay Shaplov
Postgres Professional: http://www.postgrespro.com
Russian Postgres Company