Scott Ribe wrote:
On Nov 18, 2024, at 12:10 PM, Achilleas Mantzios <a.mantzios@xxxxxxxxxxxxxxxxxxxx> wrote:
I'd like to add if someone can tolerate some seconds of downtime, with logical replication it is possible to swap the roles of publisher / subscriber so that one can keep the old database as a means of extra safety. If his/her app breaks beyond repair and there is no time for fixing, one can simply go back to the old DB with some minor adjustments (such as sequences).
In the same vein, if one's file system offers atomic snapshots:
1) shut down PG
2) run pg_upgrade with the hard links option
3) take snapshot
4) start up new PG
Now, if there's a failure, you can go back to the snapshot and either start up the old PG, or do some troubleshooting and try upgrading again.
Er... I would think that you'd want to do:
1) shut down PG
2) take snapshot
3) run pg_upgrade with the hard links option
4) start up new PG
as otherwise the old PG won't start properly, due to the changes made by
pg_upgrade.
-kgd