From: Tom Lane <tgl@sss.pgh.pa.us>
To: Massimo Ortensi <mortensi@unimaticaspa.it>
cc: pgsql-admin@lists.postgresql.org
Subject: Re: Out of memory error during pg_upgrade in big DB with large
 objects
In-reply-to: <1ox9nq-005Lel-UF@unimaticaspa.it>
References: <1ox9nq-005Lel-UF@unimaticaspa.it>
Comments: In-reply-to Massimo Ortensi <mortensi@unimaticaspa.it>
	message dated "Mon, 21 Nov 2022 17:37:50 +0100"
MIME-Version: 1.0
Content-Type: text/plain; charset="us-ascii"
Content-ID: <3215515.1669051820.1@sss.pgh.pa.us>
Date: Mon, 21 Nov 2022 12:30:20 -0500
Message-ID: <3215516.1669051820@sss.pgh.pa.us>
Archived-At: 
 <https://www.postgresql.org/message-id/3215516.1669051820%40sss.pgh.pa.us>
Precedence: bulk

Massimo Ortensi <mortensi@unimaticaspa.it> writes:
> I'm trying to upgrade a huge DB from postgres 10 to 14

> This cluster is 70+ TB, with one database having more than 2 billion 
> records in pg_largeobject

> I'm trying pg_upgrade in hard link mode, but the dump of databas schema 
> phase always fails with

> pg_dump: error: query failed: out of memory for query result
> pg_dump: error: query was: SELECT l.oid, (SELECT rolname FROM 
> pg_catalog.pg_roles WHERE oid = l.lomowner) AS rolname, (SELECT 
> pg_catalog.array_agg(acl ORDER BY row_n) FROM (SELECT acl, row_n FROM 

FWIW, this query was rewritten pretty substantially in v15.
It's still going to produce a row per large object, but it
should be a lot narrower because most of the ACL-wrangling
now happens somewhere else.  I don't know if migrating to
v15 instead of v14 is an option for you, and I can't promise
that that'd be enough savings to fix it anyway.  But it's
something to think about.

			regards, tom lane