Table of Contents
PostgreSQL/PostGIS Upgrade from 9.6 to 11
The present notes are about upgrading PostgreSQL from Debian 9 Stretch to Debian 10 Buster, i.e. upgrading from PostgreSQL 9.6 to version 11 and consequently upgrading PostGIS 2.3 to PostGIS 2.5.
We did the Debian upgrade with the standard apt-get dist-upgrade procedure. Once the upgrade finished, we can verify that both versions of PostgreSQL are running:
~# pg_lsclusters Ver Cluster Port Status Owner Data directory Log file 9.6 main 5432 online postgres /var/lib/postgresql/9.6/main /var/log/postgresql/postgresql-9.6-main.log 11 main 5433 online postgres /var/lib/postgresql/11/main /var/log/postgresql/postgresql-11-main.log
Notice that the new 11/main cluster is running on the non-standard TCP port 5433.
We can also verify that all our databases are running into the old cluster 9.6/main:
~$ psql --cluster 9.6/main postgres=# \l Name | Owner | --------------------+--------------------+- first_database | ... | second_database | ... | third_database | ... | ...
Instead, into the new cluster, only the three system databases should exist: postgres, template0 and template1:
~$ psql --cluster 11/main postgres=# \l
The pg_upgradecluster can safely upgrade standard databases, but the relative manpage clearly states that: Some PostgreSQL extensions like PostGIS need metadata in auxiliary tables which must not be upgraded from the old version, but rather initialized for the new version before copying the table data.
We have several standard (non-GIS) databases and a few PostGIS-enabled ones. You can see if the PostGIS extension is enabled issuing the \dx once connected to the database. So our strategy will be:
- Dump the PostGIS-enabled databases into files.
- Drop the PostGIS databases.
- Do a pg_upgradecluster to migrate and upgrade all the non-GIS databases to the new cluster.
- Restore the PostGIS databases from the dump files into the new cluster.
Dump and DROP the PostGIS databases
First of all, we create a dump of all the PostGIS-enabled databases:
~# su - postgres ~$ pg_dump -Fc gis_database > gis_database.dump
Then we DROP them:
~$ psql postgres=# DROP DATABASE gis_database; DROP DATABASE
Upgrade cluster from 9.6 to 11
Drop the new 11/main cluster (which should be indeed empty):
pg_dropcluster --stop 11 main
Do the automatic upgrade of all the non-PostGIS databases:
pg_upgradecluster 9.6 main
This will print some information for each database:
Fixing hardcoded library paths for stored procedures... Upgrading database nongis_database... Analyzing database nongis_database...
Verify the new situation. Notice that the new 11/main cluster was moved to the default TCP port 5432 and it is the only one running:
~$ pg_lsclusters Ver Cluster Port Status Owner Data directory Log file 9.6 main 5433 down postgres /var/lib/postgresql/9.6/main /var/log/postgresql/postgresql-9.6-main.log 11 main 5432 online postgres /var/lib/postgresql/11/main /var/log/postgresql/postgresql-11-main.log
Finally we can drop the old 9.6/main cluster
~$ pg_dropcluster 9.6 main
Now the 11/main cluster is running, but started by the postgres user. We want it to be started by the regular systemd subsystem. To fix the situation run the following as the postgres user:
~$ pg_ctlcluster 11 main stop
and the following as the root user:
~# systemctl daemon-reload ~# systemctl stop email@example.com ~# systemctl stop postgresql@11-main ~# systemctl start postgresql@11-main
Restore the PostGIS databases
We need to CREATE the PostGIS databases as new, users should instead aready exist because they were automatically restored by pg_upgradecluster. As the postgres user:
~$ psql postgres=# CREATE DATABASE gis_database OWNER gis_owner ENCODING 'UTF8'; postgres=# \connect gis_database gis_database=# CREATE EXTENSION postgis;
Finally you can restore the dump, as a regular user (you will be prompted for the DB user password):
~$ pg_restore -U gis_user -W -h localhost -d gis_database -Fc gis_database.dump
NOTICE: during the restore some warnings will be printed, because some objects already exist into the database and they actually should be not restored: the public schema, the plpgsql language, the postgis extension and the spatial_ref_sys table.