On Thu, Sep 5, 2013 at 1:55 AM, Andres Freund <andres@2ndquadrant.com> wrote:
> On 2013-09-04 09:23:20 -0700, Jeff Davis wrote:
>> On Wed, 2013-09-04 at 11:32 -0400, Tom Lane wrote:
>> > Jeff Davis <pgsql@j-davis.com> writes:
>> > > I think code complexity matters quite a lot. If we can eliminate some
>> > > complex code in a complex area, and all we give up is a feature with
>> > > essentially no use case, that sounds like we're moving in the right
>> > > direction to me.
>> >
>> > Isn't this whole discussion academic in view of Andres' point?
>>
>> Maybe "complex code" was an overstatement. We'd be able to eliminate the
>> XLOG_FPW_CHANGE, UpdateFullPageWrites(), and one of the members of
>> XLogCtlInsert; and make xlog.c slightly shorter in the process.
>
> That path is also executed during a normal restart and during
> promotion. Check the invocation of UpdateFullPageWrites() in
> StartupXLOG(). Note that a standby needs to be able to follow a
> primaries full_page_writes setting during a promotion.
Yes, this is required for the backup from the standby.
If we make the GUC contect to PGC_POSTMASTER, I think that
we can remove XLOG_FPW_CHANGE and treat full_page_writes
the same way as wal_level, max_connections, i.e., the parameter
which CheckRequiredParameterValues() handles.
Regards,
--
Fujii Masao