On 28.03.2013 14:45, Simon Riggs wrote:
> On 28 March 2013 11:36, Robert Haas<robertmhaas@gmail.com> wrote:
>> On Thu, Mar 28, 2013 at 6:23 AM, Simon Riggs<simon@2ndquadrant.com> wrote:
>>> No, it *would* take effect. The parameter is set in a config file that
>>> is not part of the backup, so if you start the server from the backup
>>> then it doesn't know that the recovery_config_directory had been set
>>> and so it would read the recovery.conf written by pg_basebackup.
>>
>> Are you saying pg_basebackup doesn't back up postgresql.conf? I thought it did.
>
> postgresql.conf will be backed up if it is present in the data
> directory. If it is not present, it is not backed up.
Right.
> Therefore anybody using pg_basebackup and the config_file parameter
> does *not* have an executable backup when used with the -R option, as
> Heikki was suggesting was a requirement for this patch.
That's not related to the -R option, the situation with config_file is
the same with or without it. But yes, if you use config_file option to
point outside the data directory, the config file won't be backed up.
That feels intuitive to me, I wouldn't expect it to be. Same with
include or include_dir directives in the config file, as well as
hba_file and ident_file - I wouldn't expect any of the files that those
point to to be included in the backup.
The filesystem-level backup procedure documented in the user manual, not
using pg_basebackup, behaves the same.
> pg_basebackup's behaviour with respect to .conf files is undocumented
> so its a "feature" that it skips .conf files in the config_file case.
Ok. It's always beem self-evident to me, but I agree it should be
documented. How about the attached?
- Heikki
