To: =?ISO-8859-1?Q?Veikko_M=E4kinen?= <veikko.makinen@ecom.fi>
Cc: pgsql-performance@postgresql.org
Subject: Re: How does the transaction buffer work? 
In-reply-to: <42B1D2DE.4020709@ecom.fi> 
References: <42B1D2DE.4020709@ecom.fi>
Comments: In-reply-to =?ISO-8859-1?Q?Veikko_M=E4kinen?=
	<veikko.makinen@ecom.fi>
	message dated "Thu, 16 Jun 2005 22:28:30 +0300"
Date: Fri, 17 Jun 2005 00:08:09 -0400
Message-ID: <15555.1118981289@sss.pgh.pa.us>
From: Tom Lane <tgl@sss.pgh.pa.us>

=?ISO-8859-1?Q?Veikko_M=E4kinen?= <veikko.makinen@ecom.fi> writes:
> How does Postgres (8.0.x) buffer changes to a database within a 
> transaction? I need to insert/update more than a thousand rows (mayde 
> even more than 10000 rows, ~100 bytes/row) in a table but the changes 
> must not be visible to other users/transactions before every row is 
> updated.

There were some other responses already, but I wanted to add this:
there isn't any "transaction buffer" in Postgres.  The above scenario
won't cause us any noticeable problem, because what we do is mark
each row with its originating transaction ID, and then readers compare
that to the set of transaction IDs that they think are "in the past".
The number of rows affected by a transaction is not really a factor
at all.

Now of course this isn't Nirvana, you must pay somewhere ;-) and our
weak spot is the need for VACUUM.  But you have no need to fear large
individual transactions.

			regards, tom lane