To: Geoffrey <lists@serioustechnology.com>
cc: pgsql-admin@postgresql.org
Subject: Re: COPY errors when trying to convert from 7.4.19 to 8.3.1 
In-reply-to: <4819EDF9.2080000@serioustechnology.com> 
References: <4819EDF9.2080000@serioustechnology.com>
Comments: In-reply-to Geoffrey <lists@serioustechnology.com>
	message dated "Thu, 01 May 2008 12:21:13 -0400"
Date: Thu, 01 May 2008 12:44:51 -0400
Message-ID: <15880.1209660291@sss.pgh.pa.us>
From: Tom Lane <tgl@sss.pgh.pa.us>

Geoffrey <lists@serioustechnology.com> writes:
> We are trying to test our process for upgrading our database from 7.4.19 
> to 8.3.1.  We are following the instructions for dumping the 7.4.19 
> database from:
> http://www.postgresql.org/docs/8.3/static/install-upgrading.html
> ie:
> /usr/local/pg83/bin/pg_dumpall -h master -p 5434 |gzip > dump.gz
> Create a new 8.3.1 cluster as follows:
> /usr/local/pg83/bin/initdb -D /data/pgsql/master83
> Load the data into the new cluster:
> gunzip < dump.gz |/usr/local/pg83/bin/psql -d postgres

> We know we have some old data in the database that does not meet current 
> criteria, thus we are seeing errors such as:

> ERROR:  insert or update on table "aaccess" violates foreign key 
> constraint "$1"
> DETAIL:  Key (code)=(CFSAVAUS000) is not present in table "arates".

> What baffles us is, that although we get these errors, the data is still 
> being inserted in to the table, in this case, the aaccess table.

Well, the insert has already happened: the error is actually coming out
when the dump script tries to do ALTER TABLE ADD CONSTRAINT FOREIGN KEY.
(The message is perhaps a bit misleading if you don't notice which
command caused it.)

What this looks like to me is a misordering of the dump operations such
that we're trying to do the ALTER on aaccess after loading its data,
but before loading arates's data.  In theory 8.3's pg_dump should
avoid such mistakes, but it's possible you've got a case that fools it.

Anyway, the first thing you should do to investigate is to check exactly
what's the order of operations in the script and which command is
causing the failure.  If the data involved is too large to make it
convenient to eyeball the dump script with an editor, you might try
running the dump script with log_statement = all so you can see just the
SQL commands in the postmaster log.

			regards, tom lane