Thread: Enum on-disk format

Enum on-disk format

From
Scott Bailey
Date:
I'm trying to better understand the internals of Postgres, and I'm
looking at the enum type. The docs say that an enum value is stored on
disk as 4 bytes. But enum_send() returns a bytea representing the actual
text of the value and not the index of that value.  So what step am I
missing here?

Also, is there a way to see the raw data for the tuple on a page? I was
using pageinspect to try to figure out what was happening on the disk.
The get_raw_page function returns the entire page and heap_page_items
will allow me to find the substring that represents a given tuple. But
the rows were much wider than I thought they would be, 28 bytes + 2 byte
spacer to store 4 bytes of data. Is there any way to see which bytes of
an item pointer actually map to columns in a table? And where can I find
more info on how Postgres stores tuples?

Scott Bailey

Re: Enum on-disk format

From
Tom Lane
Date:
Scott Bailey <artacus@comcast.net> writes:
> I'm trying to better understand the internals of Postgres, and I'm
> looking at the enum type. The docs say that an enum value is stored on
> disk as 4 bytes. But enum_send() returns a bytea representing the actual
> text of the value and not the index of that value.  So what step am I
> missing here?

The wire format isn't necessarily the on-disk format.  In this case
we concluded that the internal OID value wouldn't be of any use to
clients.

> Also, is there a way to see the raw data for the tuple on a page?

Try contrib/pageinspect, and read
http://developer.postgresql.org/pgdocs/postgres/storage-page-layout.html

            regards, tom lane

Re: Enum on-disk format

From
Scott Bailey
Date:
Tom Lane wrote:
> Scott Bailey <artacus@comcast.net> writes:
>> I'm trying to better understand the internals of Postgres, and I'm
>> looking at the enum type. The docs say that an enum value is stored on
>> disk as 4 bytes. But enum_send() returns a bytea representing the actual
>> text of the value and not the index of that value.  So what step am I
>> missing here?
>
> The wire format isn't necessarily the on-disk format.  In this case
> we concluded that the internal OID value wouldn't be of any use to
> clients.
>
>> Also, is there a way to see the raw data for the tuple on a page?
>
> Try contrib/pageinspect, and read
> http://developer.postgresql.org/pgdocs/postgres/storage-page-layout.html
>
>             regards, tom lane
>

Thanks Tom that did the trick. The only I/O functions I'm aware of are
send, recv, in and out. What controls converting from/to wire and
on-disk formats? And why is wire format little endian and disk big endian?

And for posterity, here's how to get to the raw tuple data.

SELECT substring(page, lp_off + t_hoff + 1, lp_len - t_hoff) AS tuple_data,
   sub.*
FROM (
     SELECT (heap_page_items(page)).*, page
     FROM (
         SELECT get_raw_page('test', 0) page
     ) s
) sub



Re: Enum on-disk format

From
Peter Eisentraut
Date:
On ons, 2009-11-18 at 22:33 -0800, Scott Bailey wrote:
> The only I/O functions I'm aware of are
> send, recv, in and out. What controls converting from/to wire and
> on-disk formats?

send and recv

>  And why is wire format little endian and disk big endian?

The wire format is network order (which is big endian), the disk format
is whatever your CPU uses.