On Sun, Dec 16, 2012 at 1:16 PM, Tom Lane
<tgl@sss.pgh.pa.us> wrote:
>> I'm still not sure what
>> the OP actually wants to accomplish by moving the time of making the
>> choice.
> I guess he wants - and I have seen that before - to use the same
> postgresql.conf across a number of different systems. Where the
> individual servers are in different timezones...
> Doing initdb, extracting the few values that should be kept, and then
> putting those and an include for the common configuration file into the
> configuration is harder than just replacing it alltogether.
True, but if he's assuming that timezone is and always will be the only
thing that needs to vary across servers, he's in for pain someday
anyway. Probably sooner not later.
Another possibility for per-server overrides is to specify stuff on the
postmaster command line, assuming you're using a custom start script
that you can tweak to add parameter settings there.
regards, tom lane
With the exception of a few parameters (max_connections and the ssl related variables that we enable), the default configuration file (circa 9.0) has worked extremely well across 100+ machines so far over the last two years and counting. However, we are simply deploying these on commodity machines ($300-400 off the shelf). Spec wise such machines have not changed significantly (I suppose the shift away from higher clock speeds to more cores can be thanked for that).
So, really as a testament to Postgres and all the hard work you guys have done, using the same config file has really been quite pain free (truly, thanks!) =)
I'm not at all concerned about mainlining the regression; for windows machines we have to compile postgres ourselves anyways to enable SSL compression (the EnterpriseDB ssl library is not linked against zlib).
However, I figured I'd at least gain a little google notoriety with some specific keywords in case anyone else makes the leap from 9.0 forward and also found this runtime functionality useful.
I will take your custom script suggestion into consideration. However, since the regression seems to be working well, I think I'm likely to keep the simplicity of the system determining this for me.
Thanks,
T.J.
