Re: [bug fix] pg_ctl always uses the same event source - Mailing list pgsql-hackers

On Tue, Jul 15, 2014 at 4:01 PM, MauMau <maumau307@gmail.com> wrote:
> From: "Magnus Hagander" <magnus@hagander.net>
>
>> Is there a reason for there still being changes in guc.c, pgevent.c
>> etc? Shouldn't it all be confined to pg_ctl now? That's my
>> understanding from the thread that that's the only part we care about.
>
>
> Yes, strictly speaking, those are useful for the original proposal. However,
> they are still useful as refactoring, since the current code has the default
> event source "PostgreSQL" in many places.  Defining the default name only at
> one location would make it easier for vendors like EnterpriseDB to change
> the default name for their products.  So I hope this will also be included
> if you don't want to reject it by all means.

Well, it does in a couple of places. I'm nto sure it's that important
(as nobody has AFAIK ever requested that change from for example EDB),
but it's not a bad thing. However, with a hardcoded service name, I
think the changes to pg_event.c are probably both not necessary and
actually bad - as you'll now start getting errors in more harmless
scenarios.


>> More importantly, isn't it wrong to claim it will only be used for
>> register and unregister? If we get an early failure in start, for
>> example, there are numerous codepaths that will end up calling
>> write_stderr(), which will use the eventlog when running as a service.
>> Shouldn't the "-e" parameter be moved under "common options"?
>
>
> Yes, you are right.  -e is effective only if pg_ctl is invoked as a Windows
> service.  So it is written at register mode.  That is, -e specifies the
> event source used by the Windows service which is registered by "pg_ctl
> register".

Oh, right. I see what you mean now. That goes for all parameters
though, including -D, and we don't specify those as register mode
only, so I still think it's wrong to place it there. It is now grouped
with all other parameters that we specifically *don't* write to the
commandline of the service.


>> I also think we should have the documentation specifically note that
>> regular postgres output still goes to whatever the backend is
>> configured to do. (and of course, any failure within the backend
>> *before* we have parsed the config file for example will still go to
>> the default source, and not the pg_ctl source - so we need to make it
>> really clear that this is *only* for output from pg_ctl).
>
>
> I see.  I added this clarification to the description of -e.  I would
> appreciate it if you could correct my poor English when committing, if it
> needs improvement.

Sure, no problem.

-- Magnus HaganderMe: http://www.hagander.net/Work: http://www.redpill-linpro.com/