Re: DWH varchar2(4000)

Space is not an issue; a VARCHAR2 uses only as much storage as needed for the data. The precision of 4000 is only a constraint for stored values and a hint to display programs; it is not like the right-padded CHAR datatype.

What is the issue is datatype conversion. Besides implicit sorting semantics for numeric and date/datetime datatypes which will be eliminated, the Oracle optimizer in particular relies on the semantics of the datatype to understand data cardinality. For example, two dates stored as DATE datatypes with values of 28-FEB-2014 and 01-MAR-2014 are known to be "adjacent" values, while values of 28-FEB-2016 and 01-MAR-2016 are known not to be adjacent. This may not seem like a big deal, but it really is, particularly for DWH databases which need to make the best decisions possible for an execution plan for long-running SQL statements.

Oracle is not a NOSQL datastore and not using the appropriate datatype is tantamount to crippling the Oracle optimizer, which is crucial for query performance whether they think it is or not.

If the developers/data modelers are pre-provisioning "spare columns" because they are afraid of running ALTER TABLE ... ADD COLUMN, then the alternative they have chosen has absolutely zero advantages over using ALTER TABLE ... ADD COLUMN. Every possible objection they can dream up is magnified by the course they have chosen. If they simply don't want to worry about type-conversion, then they simply need to grow up and realize that data typing is important on this platform, even if it isn't on other platforms.

Bottom-line: use the correct datatype. If they can't decide, then they're doing it wrong.

Best of luck!

-Tim

On 12/22/14 8:01, Ricard Martinez wrote:
> Hi list,
>
> We have a DWH database on 11.2.0.4.
> The developers/data modelers are creating all the tables with
> varchar2(4000) as standard by default. They don't care if they just
> insert 10 bytes, 200 or nothing.
> When asked about why this standard is in place, the answer is just:
> "Just in case we need it".
>
> For my point of view this is not a good practice, as can cause memory
> waste on client applications, as well as data garbage or errors like
> ORA-01450: maximum key length (3218) exceeded.
>
> But as i'm not a DWH expert, maybe im missing information that proves
> that this is really a good practice, and not just a laziness way of
> working.
>
> I will be glad to know your opinion, or point me to any doc/test that
> proves me wrong, in order to learn about this.
>
> Thanks

--
http://www.freelists.org/webpage/oracle-l