To: tyagel@yahoo.com
Cc: pgsql-general@postgresql.org
Subject: Re: Vacuum related question 
In-reply-to: <EOEFLPAIAEDLPGGLIIGNEEEACIAA.tyagel@yahoo.com> 
References: <EOEFLPAIAEDLPGGLIIGNEEEACIAA.tyagel@yahoo.com>
Comments: In-reply-to "G. Thomas Yagel, Jr." <tyagel@yahoo.com>
	message dated "Thu, 16 Sep 2004 12:17:43 -0400"
Date: Sat, 18 Sep 2004 12:06:24 -0400
Message-ID: <4715.1095523584@sss.pgh.pa.us>
From: Tom Lane <tgl@sss.pgh.pa.us>

"G. Thomas Yagel, Jr." <tyagel@yahoo.com> writes:
> After 12 hours I stopped inserting recods and started a "vacuum verbose".
> The vacuuming has taken almost 2 hours now and is not even finished.  What I
> don't understand is why there is so much vacuuming to be done on some tables
> that are only being inserted into.  As an example, I have pasted the output
> from the vacuuming of the pg_largeobject table.

You may think you've only inserted, but this output says differently:

> INFO:  "pg_largeobject": found 18317486 removable, 21326336 nonremovable row
> versions in 2119638 pages

Those 18 million dead rows didn't appear out of nowhere; they could only
have come from deleting or updating large objects (BLOBs).

The reason for the multiple index-cleaning cycles is that vacuum_mem was
only high enough to remember about 1.4 million dead rows at a time.
You could have sped up the process by setting vacuum_mem higher.

			regards, tom lane