On Mon, Jun 11, 2012 at 5:24 PM, Fujii Masao <masao.fujii@gmail.com> wrote:
> On Mon, Jun 11, 2012 at 10:04 PM, Magnus Hagander <magnus@hagander.net> wrote:
>> On Sun, Jun 10, 2012 at 4:02 PM, Fujii Masao <masao.fujii@gmail.com> wrote:
>>> On Sun, Jun 10, 2012 at 7:55 PM, Magnus Hagander <magnus@hagander.net> wrote:
>>>> How about this?
>>>
>>> + /*
>>> + * Set flushed position to the last byte in the previous
>>> + * file. Per above we know that xrecoff%XLOG_SEG_SIZE=0
>>> + */
>>> + flushedpos = blockpos;
>>> + if (flushedpos.xrecoff == 0)
>>> + {
>>> + flushedpos.xlogid--;
>>> + flushedpos.xrecoff = XLogFileSize-1;
>>> + }
>>> + else
>>> + flushedpos.xrecoff--;
>>>
>>> flushedpos.xrecoff doesn't need to be decremented by one.
>>> If xrecoff % XLOG_SEG_SIZE = 0, the position should be the last
>>> byte of previous (i.e., flushed) WAL file.
>>
>> Hmm. I thikn I confused myself with "last byte written" vs "current
>> position". And we're dealing with current position here...
>>
>> So it should just be flushedpos = blockpos and be done with it, right?
>
> Yep.
>
>> Though before I commit anything with this, we need to decide what to
>> wrt syncrep on that, per the other thread.
>
> Yep.
Per the other thread, we decided to postpone this until 9.3. And also
figure out a better set of switches for pg_receivexlog to control it
with.
-- Magnus HaganderMe: http://www.hagander.net/Work: http://www.redpill-linpro.com/