On 27 March 2013 17:23, Tom Lane <tgl@sss.pgh.pa.us> wrote:
> Heikki Linnakangas <hlinnakangas@vmware.com> writes:
>> On 27.03.2013 18:10, Simon Riggs wrote:
>>> On 27 March 2013 15:35, Heikki Linnakangas<hlinnakangas@vmware.com> wrote:
>>>> Ok, cool. Can you please revert this commit so that we can move on, then?
>
>>> Please explain why you want this reverted, without mentioning the
>>> other task we agree is required.
>
>> If an admin can't trust that the file is placed in $PGDATA, it's harder
>> to determine if a server is a master or a standby. It makes tools that
>> try to promote / demote a server more complicated, because they need to
>> take this setting into account. Lastly, it breaks the new pg_basebackup
>> -R functionality; pg_basebackup will create the recovery.conf file, but
>> it won't take effect.
>
> FWIW, I agree that this is a bad idea and should be reverted.
>
> Simon is claiming that because he described this idea in one sentence
> (out of a larger post) three months ago, everyone agreed to the idea and
> there is no longer any room for discussion. In reality I suspect nobody
> really thought about the implications at the time. In any case, the
> arguments that have been made today seem to me to be sufficient reasons
> why we *don't* want to put recovery.conf in random places outside the
> data directory.
If anybody thought one sentence wasn't descriptive enough, they could
have said. They didn't because its a trivial patch with very little
room for alternative interpretations.
Arguments against? I have seen only one, Heikki's above, and its not a
good one, given related similar issues.
It's an option, you don't have to put recovery.conf anywhere else,
unless you wish to.
Anyway, as I said, I didn't do this because I want it. I did it
because it's been agreed. Without some reasonable objection, I see no
reason to revoke.
-- Simon Riggs http://www.2ndQuadrant.com/PostgreSQL Development, 24x7 Support, Training & Services
