On Tue, Dec 14, 2010 at 7:17 PM, Fujii Masao <masao.fujii@gmail.com> wrote:
> On Wed, Dec 15, 2010 at 6:35 AM, Robert Haas <robertmhaas@gmail.com> wrote:
>>> At least the reference to section 31.14 is needless since password
>>> authentication doesn't affect the pg_ctl -w for now. But, on the second
>>> thought, it can still fail because of miss-configuration of connection
>>> variable, for example PGHOST.
>>
>> I thought PQping() was supposed to handle that correctly. There are
>> four return values: PQPING_OK, PQPING_REJECT, PQPING_NO_RESPONSE,
>> PQPING_NO_ATTEMPT. I believe the last is intended to cover blatant
>> misconfiguration.
>
> True. But even though the server is running normally, the misconfiguration
> might still make PQping return PQPING_NO_ATTEMPT and make pg_ctl -w
> fail to wait for complete startup. Currently misconfiguration of password
> authentication doesn't interfere with pg_ctl -w, so we can remove the
> reference to that. But other misconfigurations can be still harm, so I thought
> the sentence still needs to be in the document.
Hmm. I'm inclined to just remove that sentence. The fact that you
can make the operation fail by pointing it at someplace where there's
not a server listening doesn't really seem to merit a disclaimer. I
think waiting IS a well-defined operation now; it just won't succeed
if you configure it in a way that doesn't make sense.
--
Robert Haas
EnterpriseDB: http://www.enterprisedb.com
The Enterprise PostgreSQL Company