[Pgpool-hackers] [hackers] PgPool II patches : regex and syslog

[Guillaume Lelarge]

Le 28/12/2010 01:53, Tatsuo Ishii a écrit :
>> Le 27/12/2010 15:59, Guillaume Lelarge a écrit :
>>> Hi,
>>>
>>> Le 27/12/2010 10:47, Gilles Darold a écrit :
>>>> [...]
>>>> Here is the rewritten patch that add syslog support to PgPool with all
>>>> your great suggestions applied.
>>>> This patch has been created from the current cvs HEAD branch.
>>> Thanks. There is a bug in it, that I fixed in the attached patch. Even
>>> if you declare log_destination and syslog_ident as reloadable, it was
>>> without effect because there was no code to handle opening or closing
>>> syslog if they changed.
>> Damn you're right, I just simply omit this part. Sorry for the
>> additional work.
>>> There is another thing that annoys me, but it isn't a bug. What is the
>>> purpose of pool_config->logsyslog? is it just to avoid a strcmp? which
>>> seems a poor optimization wrt having two variables with same meaning.
>> No, this is a flag to delay PgPool logging to syslog only after the end
>> of configuration file reading. The problem appears when PgPool is run in
>> debug mode, during the configuration file it detect that log_destination
>> is set to syslog so if you don't wait (or use a flag)  it will try to
>> write to syslog before the syslog connection is opened.
>>
>>>> This patch also modify the comment about logdir configuraion directive.
>>>>
>>>> "Logging directory" has been change into "PgPool status file logging
>>>> directory"
>>>>
>>>> Feel free to apply or remove it. It just seems to me that 'logdir' =>
>>>> "Logging directory" is very confusing and false as this directive is
>>>> only used to save the pgpool_status file. It will certainly be better to
>>>> rename this directive, but for the moment just changing the comment help
>>>> a lot for understanding and preserve backward compatibility.
>>> I completely agree with you on that change. But I would much prefer to
>>> see it applied as another patch rather than in this patch. They have two
>>> different purposes, so they shouldn't be mixed.
>> Ok, memorized: one purpose per patch. Please remove it and make an other
>> one if you think it should be applied.
>>
>>> BTW, Tatsuo-san, Gilles and I have absolutely no idea if we should
>>> change pool_config.c, pool_config.l, or both. We did both, but I would
>>> really like to know which of these three possibilities is the good one.
>>> Thanks.
>> Yes this is because in cvs pool_config.c has not been fully regenerated
>> from pool_config.l and some part of the regex patch have not been
>> applied to pool_config.c
> 
> I think you only need patches against pool_config.l. I will generate
> pool_config.c from pool_config.l.
> 

Do you need another patch from us, or is the current enough for you?


-- 
Guillaume
 http://www.postgresql.fr
 http://dalibo.com