Re: [HACKERS] Dbsize backend integration

Bruce Momjian wrote:

> Dave Page wrote:

>> pg_relation_size(text)   - Get relation size by name/schema.name
>> pg_relation_size(oid)    - Get relation size by OID
>> pg_tablespace_size(name) - Get tablespace size by name
>> pg_tablespace_size(oid)  - Get tablespace size by OID
>> pg_database_size(name)   - Get database size by name
>> pg_database_size(oid)    - Get database size by OID
>> pg_table_size(text)   - Get table size (including all indexes and
>> toast tables) by name/schema.name
>> pg_table_size(oid)    - Get table size (including all indexes and
>> toast tables) by OID
>> pg_size_pretty(int8)     - Pretty print (and round) the byte size
>> specified (eg, 123456 = 121KB)

> OK, so you went with relation as heap/index/toast only, and table as the
> total of them.  I am not sure that makes sense because we usually equate
> relation with table, and an index isn't a relation, really.
>
> Do we have to use pg_object_size?  Is there a better name?  Are
> indexes/toasts even objects?

Relation is not an ideal names, but I heard people talk about heap relation
and index relation. Indexes and tables (and sequences) are treated in a
similar way quite often. Think of ALTER TABLE example_index RENAME TO
another_index. This is even less obvious.  Of course in relational theory,
an index would not be a relation, because an index is just implementation
detail.

I don't like object_size any better, since that makes me rather think of
large objects or rows as objects (object id...).

Perhaps pg_table_size should be split into pg_table_size and
pg_indexes_size, where pg_indexes_size is the aggregate of all indexes on a
table und pg_table_size is just table+toast+toast-index.

If noone has a better idea for pg_relation_size, I would rather keep it for
consistency with the contrib module, and because it's not too far off.

Best Regards,
Michael Paesold