Re: TOAST performance (was Re: [GENERAL] Delete Performance) - Mailing list pgsql-hackers

From Bruce Momjian
Subject Re: TOAST performance (was Re: [GENERAL] Delete Performance)
Date
Msg-id 200111170201.fAH21q115282@candle.pha.pa.us
Whole thread Raw
In response to TOAST performance (was Re: [GENERAL] Delete Performance)  (Tom Lane <tgl@sss.pgh.pa.us>)
List pgsql-hackers
> The "delete" took about 2 seconds.  I then did it over with the
> 'value' being a 5K chunk of text, which according to octet_length
> got compressed to 3900 bytes.  (This'd require two rows in the TOAST
> table.)  This time the delete took 127 seconds.  I was expecting
> about a 3X penalty since we needed to delete three rows not one,
> but what I got was a 60X penalty.

Wow.  Can someone remind me why we take page images on delete?  We
aren't really writing anything special to the page except a transction
id.

> I've worried for some time that the decision to XLOG page images was
> costing us a lot more performance than could be justified...

Is it because we take a snapshot of the page before we write it in case
we only write part of the page?

--  Bruce Momjian                        |  http://candle.pha.pa.us pgman@candle.pha.pa.us               |  (610)
853-3000+  If your life is a hard drive,     |  830 Blythe Avenue +  Christ can be your backup.        |  Drexel Hill,
Pennsylvania19026
 


pgsql-hackers by date:

Previous
From: Tom Lane
Date:
Subject: Re: TOAST performance (was Re: [GENERAL] Delete Performance)
Next
From: Bruce Momjian
Date:
Subject: Re: TOAST performance (was Re: [GENERAL] Delete Performance)