On 27.03.2013 15:11, Simon Riggs wrote:
> On 27 March 2013 12:59, Michael Paquier<michael.paquier@gmail.com> wrote:
>
>> Also, based on Greg's spec (that Robert and I basically agreed on), if
>> recovery.conf is found at the root of data folder an error is returned to
>> user, recommending him to migrate correctly by referring to dedicated
>> documentation.
>
> I'm following what was agreed on 24 December.
Well, there wasn't much discussion about it back then. The way I read
the thread is that people agreed with the general approach, as now
implemented in Michael's patch, based on Fujii's earlier patch. This
might be a good idea or not, but it's a new and separate feature, not
related to whatever else we might do with recovery.conf.
If we are to discuss the merits of this patch now, a few thoughts:
1. This is going to make life more complicated for tools that want to
mess with recovery.conf, as it's no longer guaranteed to be in $PGDATA.
2. An admin can no longer tell if a server is in standby or PITR mode
just by checking for $PGDATA/recovery.conf
3. Would it make sense to make the option "recovery_config_file",
pointing to the file, instead of just the directory?
4. Could you achieve the same with a symlink in $PGDATA?
> We can have the whole debate again, if you wish. There is no reason to
> break backwards compatibility to get what we want.
AFAICS this is completely orthogonal to backwards-compatibility and
other aspects of the upcoming patch to merge recovery.conf and
postgresql.conf.
- Heikki
