Re: [HACKERS] Fwd: Joins and links

From: Hannu Krosing <hannu(at)trust(dot)ee>
To: Leon <leon(at)udmnet(dot)ru>
Cc: Vadim Mikheev <vadim(at)krs(dot)ru>, pgsql-hackers(at)postgreSQL(dot)org
Subject: Re: [HACKERS] Fwd: Joins and links
Date: 1999-07-06 23:18:10
Message-ID: 37828EB2.7DF44EAF@trust.ee
Views: Raw Message | Whole Thread | Download mbox | Resend email
Thread:
Lists: pgsql-hackers

Leon wrote:
>
> Hello Vadim,
>
> Tuesday, July 06, 1999 you wrote:
>
> >> These joins
> >> will never be too slow for an unknown reason. Think carefully.
> >> I hope you will make wise decision.
>
> V> Optimizer requires major rewrite in any case, even
> V> having links implemented.
>
> I am afraid that optimizer, even totally rewritten, can't choose
> the best method always. That is simply because it is such a
> complex animal :) Bacterium - simple links will always win
> in the field where they live :)

>From what I have read from earlier posts about the optimizer,
there can be situations where using links would actually be slower
than going through the optimiser, similar to the case where scanning
the whole table using an index can be orders of magnitude slower than
doing a direct scan.

That is of course if used unwisely ;)

Another thing that has remained unclear to me is the way to actually
insert or update the links - you can't just put another record there,
so that should be some kind of field (tid,oid,...) or some function
like last_touched('other_table_name').

So, what have you thought to put there ?

------
Hannu

In response to

Responses

Browse pgsql-hackers by date

  From Date Subject
Next Message Thomas Lockhart 1999-07-07 00:04:55 Re: A couple comments about datetime
Previous Message Peter Eisentraut 1999-07-06 22:24:57 Re: [HACKERS] Fwd: Joins and links