Hi all.
(I tried raising this on -hackers but got "this is not supported"-answer, which is quite dissatisfactory..)
In short pg_upgrade fails with:
Linking user relation files
No match found in new cluster for old relation with OID 16388 in database "andreak": "pg_toast.pg_toast_2613" which is the TOAST table for "pg_catalog.pg_largeobject"
No match found in new cluster for old relation with OID 16390 in database "andreak": "pg_toast.pg_toast_2613_index" which is an index on "pg_toast.pg_toast_2613" which is the TOAST table for "pg_catalog.pg_largeobject"
Failed to match up old and new tables in database "andreak"
Failure, exiting
I issued the following command:
PG_NEW_VERSION=9.6
PG_OLD_VERSION=9.5
/usr/lib/postgresql/$PG_NEW_VERSION/bin/pg_upgrade \
--old-bindir=/usr/lib/postgresql/$PG_OLD_VERSION/bin/ \
--new-bindir=/usr/lib/postgresql/$PG_NEW_VERSION/bin/ \
--old-datadir=/var/lib/postgresql/$PG_OLD_VERSION/main \
--new-datadir=/var/lib/postgresql/$PG_NEW_VERSION/main \
-o " -c config_file=/etc/postgresql/$PG_OLD_VERSION/main/postgresql.conf" \
-O " -c config_file=/etc/postgresql/$PG_NEW_VERSION/main/postgresql.conf" \
--link
My database and tablespaces are created like this:
I have tablespaces configured outside PGDATA:
mkdir /var/lib/postgresql/9.5/tablespaces/andreak
mkdir /var/lib/postgresql/9.5/tablespaces_lo/andreak
psql -c "create tablespace andreak OWNER andreak location '/var/lib/postgresql/9.5/tablespaces/andreak'" postgres;
psql -c "create tablespace andreak_lo OWNER andreak location '/var/lib/postgresql/9.5/tablespaces_lo/andreak'" postgres;
createdb --tablespace=andreak -O andreak andreak
psql -U postgres -c "alter table pg_largeobject set tablespace andreak_lo" -d andreak
(I've set allow_system_table_mods=on in postgresql.conf)
These are symlinked:
ln -s /storage/wal/9.5/pg_xlog /var/lib/postgresql/9.5/main/pg_xlog
ln -s /storage/fast_ssd/9.5/tablespaces /var/lib/postgresql/9.5/tablespaces
ln -s /storage/archive_disk/9.5/tablespaces_lo /var/lib/postgresql/9.5/tablespaces_lo
I would assume that having pg_largeobject in a separate tablespace is more and more common these days, having real-cheap SAN vs. fast-SSD for normal tables/indexes/wal.
So - I'm wondering if we can fund development of pg_upgrade to cope with this configuration or somehow motivate to getting this issue fixed?
Would any of the PG-companies (2ndQ, EDB, PgPro) take a stab at this?
Any feedback welcome, thanks.
--
Andreas Joseph Krogh
CTO / Partner - Visena AS
Mobile: +47 909 56 963