Re: BUG #14009: pg_restore not restoring privilegs when one login/group role is missing

I tried the command as you suggested but still had the same issue.

[postgres@server4 ~]$ pg_restore -p 5432 -d stgsample04
/var/lib/pgsql/data/dumps/stgsample01_schema.dump | psql

pg_restore: [archiver (db)] Error while PROCESSING TOC:
pg_restore: [archiver (db)] Error from TOC entry 3067; 0 0 ACL country
stgsamplelogin01
pg_restore: [archiver (db)] could not execute query: ERROR:  role
"sample_read_role" does not exist
    Command was: REVOKE ALL ON TABLE country FROM PUBLIC;
REVOKE ALL ON TABLE country FROM stgsamplelogin01;
GRANT ALL ON TABLE country TO st...
pg_restore: [archiver (db)] Error from TOC entry 3068; 0 0 ACL
country_alias stgsamplelogin01
pg_restore: [archiver (db)] could not execute query: ERROR:  role
"sample_read_role" does not exist
    Command was: REVOKE ALL ON TABLE country_alias FROM PUBLIC;
REVOKE ALL ON TABLE country_alias FROM stgsamplelogin01;
GRANT ALL ON TABLE c...
pg_restore: [archiver (db)] Error from TOC entry 3069; 0 0 ACL log
stgsamplelogin01
pg_restore: [archiver (db)] could not execute query: ERROR:  role
"sample_read_role" does not exist
    Command was: REVOKE ALL ON TABLE log FROM PUBLIC;
REVOKE ALL ON TABLE log FROM stgsamplelogin01;
GRANT ALL ON TABLE log TO stgsamplelogin...
pg_restore: [archiver (db)] Error from TOC entry 3070; 0 0 ACL result
stgsamplelogin01
pg_restore: [archiver (db)] could not execute query: ERROR:  role
"sample_read_role" does not exist
    Command was: REVOKE ALL ON TABLE result FROM PUBLIC;
REVOKE ALL ON TABLE result FROM stgsamplelogin01;
GRANT ALL ON TABLE result TO stgsa...
pg_restore: [archiver (db)] Error from TOC entry 3071; 0 0 ACL tariff
stgsamplelogin01
pg_restore: [archiver (db)] could not execute query: ERROR:  role
"sample_read_role" does not exist
    Command was: REVOKE ALL ON TABLE tariff FROM PUBLIC;
REVOKE ALL ON TABLE tariff FROM stgsamplelogin01;
GRANT ALL ON TABLE tariff TO stgsa...
WARNING: errors ignored on restore: 5

[postgres@server4 ~]$ psql
psql (9.5.1)
Type "help" for help.

postgres=# \connect stgsample04 stgsamplelogin01
You are now connected to database "stgsample04" as user "stgsamplelogin01".
stgsample04=> \dp
                                   Access privileges
 Schema |       Name        | Type  | Access privileges | Column privileges
| Policies
--------+-------------------+-------+-------------------+-------------------+----------
 public | country           | table |                   |
|
 public | country_alias     | table |                   |
|
 public | log               | table |                   |
|
 public | result            | table |                   |
|
 public | tariff            | table |                   |
|
 public | tariff_bk20140630 | table |                   |
|
(6 rows)


Regards,
Alexander Spiteri

On 10 March 2016 at 17:24, Tom Lane <tgl@sss.pgh.pa.us> wrote:

> Alexander Spiteri <alexander@spiteri.org> writes:
> > Please find a test case attached. I tried to put as much detail as i
> could.
>
> Okay, I see what's going on.  I really doubt that there is a regression
> here from 8.4 though.  I will bet that the difference is you used plain
> text output from pg_dump back in 8.4, and now in 9.5 you are trying to
> use pg_restore direct-to-database mode.
>
> The proximate cause of the problem is that pg_dump bundles up GRANT/REVOKE
> statements for any object into a single "ACL" entry in the archive file.
> If you dump to SQL text this doesn't matter because it just looks like a
> stream of separate statements.  But if pg_restore is trying to restore
> direct to database then it issues the whole text of each archive file
> entry as a single PQexec() operation, which means that an error in one
> statement in that string causes the rest of 'em to be skipped.
>
> As a short-term workaround you could try restoring in this way:
>
>    pg_restore foo.dump | psql
>
> To really fix it, I think we'd need either to expand "ACL" archive entries
> into a separate entry for each target user, or to change pg_restore to
> parse the entry contents into separate statements.  The latter seems
> messy and bug-prone.  The former solution wouldn't fix the problem
> for dumps from existing pg_dump versions; but on the other hand, it might
> provide additional flexibility for selective restores, so I'm not sure
> that it's a bad idea going forward.
>
>                         regards, tom lane
>