Re: (ITS#4605) backs-sql: speedup when using views

[reiffert@Student.Physik.Uni-Mainz.DE]

On Sat, 1 Jul 2006, Pierangelo Masarati wrote:

> Your hack appears a misuse of back-sql that breaks its generality.  The
> fact that you get poor performance out of back-sql, especially when
> using views(, especially if your views are not optimized by adding
> appropriate indices to keys, but this is out of scope here)

Yes, thats beyond the scope here. Whenever mysql-5 can handle indices
on views ... I thank god and use it!

> is intrinsic
> in the design of back-sql.  If you need performance, use back-bdb or
> back-hdb.

Unfourtunatly I'm in the situation where my company left back-bdb and
put all adresses in a mysql-database, so I'm very sad about the new
situation but I can't change it.

> If you put everything in one view that hijacks the name of a
> table that is required internally by back-sql to work appropriately,
> and, as such, should not be altered, then you cannot complain about it
> breaking basic implicit join and aliasing rules.

Dear Pierangelo, I'm not complaining about broken stuff, all I was
finding out is, that when I hijack the system like I did, it works
pretty much faster for me. Magnitudes.

>
> Your issue is the result of a misuse of the software;

I came here to say "hello, here is my workaround, let's make
it better, so that the software gets better".


> I suggest your
> patch be rejected as it would break back-sql destroying its generality

Acknowledged, as this patch definitly is evil. How can it be done
better?

Implement a slapd.conf variable "hijacker true" with the default
"hijacker false"?
Can it be done in other ways? Or is rejecting the patch everything
that can be done?

>
> and versatility.  Feel free to use your patched version as a custom
> backend that suits your addressbook-only needs, though.

Kind regards
Thomas Reifferscheid