Le 15/11/2016 à 16:44, Tom Lane a écrit :
> You'd have to provide a lot more detail before anyone could tell if there
> was a fixable bug here, but I rather doubt it. There are at least two
> ways this scenario might lose:
>
> 1. There are additional objects in the target database that have
> dependencies on ones that are in the dump. In that case there is
> no ordering of dropping the objects in the dump that will succeed.
The target databased was restored from the dump in an empty database.
So what I do is :
dropdb -U postgres -h localhost db1
createdb -U postgres -h localhost db1
psql -U postgres -h localhost -c "CREATE EXTENSION postgis" db1
pg_dump -h pgsql1 -U postgres -b -Fc -E UTF8 -N public -N osm -N osm2 -f
"D:\temp.dump" db1
pg_restore -U postgres -h localhost -j 2 -d db1 "D:\temp.dump"
That works, no error, no warning.
Then
pg_restore -U postgres -h localhost --clean -d db1 "D:\temp.dump"
fails.
So can we exclude additional objects in the target database in this case ?
> 2. There are dependency chains passing through objects that weren't
> dumped (ie, if A depends on B which depends on C, and you omit B
> from the dump, it might still be the case that A can't be restored
> before C).
Can I trust what pgadmin says about objects dependent on a schema ?
It says that public schema's dependent objects are only it's own
operators, functions, etc. (i.e., what's in the postgis extension), and
the same for the other two schemas.
They don't show any dependent objects outside themselves.
> If you think neither of those cases apply, please provide a self-contained
> test case.
That's not going to be easy. I'll try to trim down a pg_dump -s to see
how I can reproduce this.
--
Arnaud
