Markus Schaber wrote:
> Additionally, in most UPDATE cases, the new row version will fit into
> the same page as the old version. In this case, the index does not have
> to be changed, which is an additional speed improvement.
Actually, when the UPDATE puts a new row version in the same heap page,
the index must be updated anyway. All the rest of what you said is
There is another reason not to put visibility info in the index, which
is that it would be extremely complex to update all indexes to contain
the right visibility (and maybe impossible without risking deadlocks).
Updating only the heap is very simple.
Alvaro Herrera http://www.CommandPrompt.com/
The PostgreSQL Company - Command Prompt, Inc.
In response to
pgsql-performance by date
|Next:||From: Worky Workerson||Date: 2006-10-23 15:10:19|
|Subject: Re: Best COPY Performance|
|Previous:||From: Markus Schaber||Date: 2006-10-23 13:25:49|
|Subject: Re: Index on two columns not used|