Обсуждение: Page type and version

As the upcoming release is breaking compatibility anyway:  what do you
think about placing a magic number and some format version info into
the page header?

One 32-bit-number per page should be enough to encode page type and
version.  We have just to decide, how we want it:

a) combine page type and version into a single 32-bit magic number
   HEAPPAGE73 = 0x63c490c9   HEAPPAGE74 = 0x540beeb3   ...   BTREE73    = 0x8cdc8edb   BTREE74    = 0xbb13f0a1

b) use n bits for the page type and the rest for a version number
   HEAPPAGE73 = 0x63c40703   HEAPPAGE74 = 0x63c40704   ...   BTREE73    = 0x8cdc0703   BTREE74    = 0x8cdc0704

The latter has the advantage, that the software could easily check for
a version range (e.g. if (PageGetVersion(page) <= 0x0703) ...).

One might argue, that one magic number *per file* should be
sufficient.  That would mean, that the first page of a file had to
have a different format.  Btree has such a meta page;  I don't know
about the other access methods.

With a magic number in every single page it could even be possible to
do a smooth upgrade:  "Just install Postgres 8.0 and continue to use
your PostgreSQL 7.4 databases" :-).  Whenever the backend reads an old
format page it uses alternative accessor routines.  New pages are
written in the new format.  Or the database can be run in
compatibility mode ...  I'm dreaming ...

Thoughts?

ServusManfred

Manfred Koizar wrote:
> As the upcoming release is breaking compatibility anyway:  what do you
> think about placing a magic number and some format version info into
> the page header?
> 
> One 32-bit-number per page should be enough to encode page type and
> version.  We have just to decide, how we want it:
> 
> a) combine page type and version into a single 32-bit magic number
> 
>     HEAPPAGE73 = 0x63c490c9
>     HEAPPAGE74 = 0x540beeb3
>     ...
>     BTREE73    = 0x8cdc8edb
>     BTREE74    = 0xbb13f0a1
> 
> b) use n bits for the page type and the rest for a version number
> 
>     HEAPPAGE73 = 0x63c40703
>     HEAPPAGE74 = 0x63c40704
>     ...
>     BTREE73    = 0x8cdc0703
>     BTREE74    = 0x8cdc0704
> 
> The latter has the advantage, that the software could easily check for
> a version range (e.g. if (PageGetVersion(page) <= 0x0703) ...).

Yea, b) sounds good.

> One might argue, that one magic number *per file* should be
> sufficient.  That would mean, that the first page of a file had to
> have a different format.  Btree has such a meta page;  I don't know
> about the other access methods.

Heap used to have a header page too but it was removed long ago.

We do have the TODO item:
* Add version file format stamp to heap and other table types

but I am now questioning why that is there.  btree had a version stamp,
so I thought heap should have one too, but because the PG_VERSION file
is in every directory, isn't that all that is needed for version
information.

My vote is just to remove the btree version.  If we decide to implement
multi-version reading in the backend, we can add it where appropriate.


> With a magic number in every single page it could even be possible to
> do a smooth upgrade:  "Just install Postgres 8.0 and continue to use
> your PostgreSQL 7.4 databases" :-).  Whenever the backend reads an old
> format page it uses alternative accessor routines.  New pages are
> written in the new format.  Or the database can be run in
> compatibility mode ...  I'm dreaming ...

Yes, and as I understand, it is pretty easy from a tuple to snoop to the
end of the block to see what version stamp is there.  Will we ever use
it?

--  Bruce Momjian                        |  http://candle.pha.pa.us pgman@candle.pha.pa.us               |  (610)
853-3000+  If your life is a hard drive,     |  830 Blythe Avenue +  Christ can be your backup.        |  Drexel Hill,
Pennsylvania19026