fails to start after install if invalid locale is set

[Scott Moser]

For more information on locale in the cloud-images, see bug 859814.

If you install postgresql-9.1 with invalid LC_ALL, then postgres will not be running after install (which causes issues with things that depend on it).

Try:
$ LC_ALL=fr_FR.UTF-8 ssh ubuntu@$HOSTNAME
% locale
locale: Cannot set LC_CTYPE to default locale: No such file or directory
locale: Cannot set LC_MESSAGES to default locale: No such file or directory
locale: Cannot set LC_ALL to default locale: No such file or directory
LANG=en_US.UTF-8
LANGUAGE=
LC_CTYPE="fr_FR.UTF-8"
LC_NUMERIC="fr_FR.UTF-8"
LC_TIME="fr_FR.UTF-8"
LC_COLLATE="fr_FR.UTF-8"
LC_MONETARY="fr_FR.UTF-8"
LC_MESSAGES="fr_FR.UTF-8"
LC_PAPER="fr_FR.UTF-8"
LC_NAME="fr_FR.UTF-8"
LC_ADDRESS="fr_FR.UTF-8"
LC_TELEPHONE="fr_FR.UTF-8"
LC_MEASUREMENT="fr_FR.UTF-8"
LC_IDENTIFICATION="fr_FR.UTF-8"
LC_ALL=fr_FR.UTF-8
% sudo apt-get install postgresql-9.1
...
After this operation, 11.6 MB of additional disk space will be used.
perl: warning: Setting locale failed.
perl: warning: Please check that your locale settings:
 LANGUAGE = (unset),
 LC_ALL = "fr_FR.UTF-8",
 LANG = "en_US.UTF-8"
    are supported and installed on your system.
perl: warning: Falling back to the standard locale ("C").
locale: Cannot set LC_CTYPE to default locale: No such file or directory
locale: Cannot set LC_MESSAGES to default locale: No such file or directory
locale: Cannot set LC_ALL to default locale: No such file or directory
Selecting previously unselected package postgresql-9.1.
(Reading database ... 26213 files and directories currently installed.)
Unpacking postgresql-9.1 (from .../postgresql-9.1_9.1.3-2_i386.deb) ...
Setting up postgresql-9.1 (9.1.3-2) ...
Error: The locale requested by the environment is invalid.
Error: could not create default cluster. Please create it manually with

  pg_createcluster 9.1 main --start

or a similar command (see 'man pg_createcluster').
update-alternatives: using /usr/share/postgresql/9.1/man/man1/postmaster.1.gz to provide /usr/share/man/man1/postmaster.1.gz (postmaster.1.gz) in auto mode.

There is no postgres service running at this point.

Related bugs:
 * bug 859814: Locale issues with beta-1/2 cloud-images
 * bug 1134036: [maas] Failure when using ssh with a locale that is not configured on the server

ProblemType: Bug
DistroRelease: Ubuntu 12.04
Package: postgresql-9.1 9.1.3-2
ProcVersionSignature: User Name 3.2.0-20.33-virtual 3.2.12
Uname: Linux 3.2.0-20-virtual i686
ApportVersion: 1.95-0ubuntu1
Architecture: i386
Date: Fri Mar 30 18:16:50 2012
Ec2AMI: ami-bdea34d4
Ec2AMIManifest: (unknown)
Ec2AvailabilityZone: us-east-1c
Ec2InstanceType: t1.micro
Ec2Kernel: aki-805ea7e9
Ec2Ramdisk: unavailable
ProcEnviron:
 TERM=xterm
 LANG=en_US.UTF-8
 SHELL=/bin/bash
SourcePackage: postgresql-9.1
UpgradeStatus: No upgrade log present (probably fresh install)

[Launchpad Janitor (janitor)]

Status changed to 'Confirmed' because the bug affects multiple users.

[Robinpeder (robinpeder)]

This also affects the result of the installation, even when using a valid locale.

If postgresql is installed in a terminal with LC_ALL="C", it is not possible to create unicode databases:

$ createdb -E UNICODE test
createdb: database creation failed: ERROR: new encoding (UTF8) is incompatible with the encoding of the template database (SQL_ASCII)

Specifying --locale when running pg_createcluster in the postinstall script will make the installation independent of the terminal.

[Martin Pitt (pitti)]

$ LC_ALL=fr_FR.UTF-8 ssh ubuntu@$HOSTNAME
% locale
locale: Cannot set LC_CTYPE to default locale: No such file or directory

That means that fr_FR.UTF-8 doesn't actually exist on the target host.

[Julian Edwards (julian-edwards)]

I don't understand why this was marked invalid Martin. You're just confirming why the bug exists.

[Martin Pitt (pitti)]

If anything, this is a bug in ssh for insisting to apply your host's locale to the remote shell. I have always considered that EBW, but I don't want to work around this in PostgreSQL by picking some random locale if the one that's actually set is invalid.

[Robie Basak (racb)]

> picking some random locale

How about C or C.UTF-8? These should be suitable non-random fallbacks, right?

[Martin Pitt (pitti)]

Robie Basak [2013-06-04 8:37 -0000]:
> > picking some random locale
>
> How about C or C.UTF-8? These should be suitable non-random fallbacks,
> right?

That's still not what you would expect on the server, which is usually
to use the server's configured system locale. C or C.UTF-8 basically
means "broken for everybody", as that will have the wrong
collation/sorting for all actual languages.

I think it is better to fail with an obvious error than to silently
create a cluster with an unwanted locale.

[Robie Basak (racb)]

I've started a discussion thread at: https://lists.ubuntu.com/archives/ubuntu-devel/2013-June/037195.html

[Martin Pitt (pitti)]

In the discussion we found the following compromise: The postinst should ignore the locale from the environment (as sudo and ssh make that unreliable), and instead do the following fallback:

 * if /etc/default/locale defines a locale, use that
 * otherwise, if /etc/environment defines a locale, use that (that's a pam_env file, so source that in a subprocess and handle errors!)
 * otherwise, use C

That will break cases where the system locale is defined someplace else, but then people can still either drop and re-create the default cluster, or specify the locale on creating new DBs.

[Martin Pitt (pitti)]

Fixed in p-common packaging branch r1355.

[Launchpad Janitor (janitor)]

This bug was fixed in the package postgresql-common - 145

---------------
postgresql-common (145) unstable; urgency=low

  [ Christoph Berg ]
  * pg_ctlcluster: Use "install" instead of File::Path to create
    unix_socket_directory. (Introduced in 141, Closes: #710093)

  [ Martin Pitt ]
  * debian/maintscripts-functions, configure_cluster(): Do not trust the
    locale from the environment, as programs like ssh and sudo propagate
    remote and user locale by default. Instead, only use the locale settings
    from /etc/environment and /etc/default/locale, to prevent trying to
    configure the default cluster with a nonexisting or hard to predict
    locale. (LP: #969462, also see Debian #700271)

 -- Christoph Berg <email address hidden> Mon, 10 Jun 2013 17:01:01 +0200