Thread: Changing locale of an existing database
Hi
I am using PostgreSQL 17 and would like to take advantage of performance and stability across OS updates of builtin C.UTF-8 locale.
I have a cluster with a DB created with en_US.UTF-8 libc locale. I would like to migrate the DB to C.UTF-8. Ideally there'd be an "ALTER DATABASE ... SET LOCALE ..." command that would take care of it but it seems it doesn't exist. I was thinking that I could change the collation of all TEXT/CHAR/VARCHAR columns in all the tables to pg_c_utf8, REINDEX all those columns and change the default locale in the pg_database table.
Is it a sensible plan? Am I missing some steps? I can't find any reference to anybody doing that before or discouraging it.
Thanks,
Marcin
--
Marcin Gozdalik
On 6/17/25 09:20, Marcin Gozdalik wrote: > Hi > > I am using PostgreSQL 17 and would like to take advantage of performance > and stability across OS updates of builtin C.UTF-8 locale. > I have a cluster with a DB created with en_US.UTF-8 libc locale. I would > like to migrate the DB to C.UTF-8. Ideally there'd be an "ALTER DATABASE > ... SET LOCALE ..." command that would take care of it but it seems it > doesn't exist. I was thinking that I could change the collation of all > TEXT/CHAR/VARCHAR columns in all the tables to pg_c_utf8, REINDEX all > those columns and change the default locale in the pg_database table. > > Is it a sensible plan? Am I missing some steps? I can't find any > reference to anybody doing that before or discouraging it. How big a database are we talking about? To me it would seem easier to create a new database with new locale and do either a pg_dump/pg_restore or logical replication to the new instance. Of course this may depend on the answer to the question above. > > Thanks, > Marcin > > -- > Marcin Gozdalik -- Adrian Klaver adrian.klaver@aklaver.com
Around 100 DBs, ranging from 1TB to 50TB.
Initial testing showed that it takes 1.5h to dump the DB and 3h to restore it for a DB of around 3TB.
Just recreating the indexes takes around 30 minutes on the same DBs.
I understand that pg_dump/pg_restore is the safe route but it's too slow for me so I'd like to understand if I have any alternative.
wt., 17 cze 2025 o 16:55 Adrian Klaver <adrian.klaver@aklaver.com> napisał(a):
On 6/17/25 09:20, Marcin Gozdalik wrote:
> Hi
>
> I am using PostgreSQL 17 and would like to take advantage of performance
> and stability across OS updates of builtin C.UTF-8 locale.
> I have a cluster with a DB created with en_US.UTF-8 libc locale. I would
> like to migrate the DB to C.UTF-8. Ideally there'd be an "ALTER DATABASE
> ... SET LOCALE ..." command that would take care of it but it seems it
> doesn't exist. I was thinking that I could change the collation of all
> TEXT/CHAR/VARCHAR columns in all the tables to pg_c_utf8, REINDEX all
> those columns and change the default locale in the pg_database table.
>
> Is it a sensible plan? Am I missing some steps? I can't find any
> reference to anybody doing that before or discouraging it.
How big a database are we talking about?
To me it would seem easier to create a new database with new locale and
do either a pg_dump/pg_restore or logical replication to the new
instance. Of course this may depend on the answer to the question above.
>
> Thanks,
> Marcin
>
> --
> Marcin Gozdalik
--
Adrian Klaver
adrian.klaver@aklaver.com
--
Marcin Gozdalik
On 6/17/25 10:30, Marcin Gozdalik wrote: > Around 100 DBs, ranging from 1TB to 50TB. > Initial testing showed that it takes 1.5h to dump the DB and 3h to > restore it for a DB of around 3TB. > > Just recreating the indexes takes around 30 minutes on the same DBs. > > I understand that pg_dump/pg_restore is the safe route but it's too slow > for me so I'd like to understand if I have any alternative. Alright I see the issue. I don't know enough about this to say whether directly altering the system catalog would be a safe operation or not. An answer to that is going to need to come from someone with more knowledge on Postgres internals then I. -- Adrian Klaver adrian.klaver@aklaver.com