On Thu, 14 Jan 2021 at 15:32, Kyotaro Horiguchi wrote:
> At Thu, 14 Jan 2021 06:46:35 +0000, Li Japin <japi...@hotmail.com> wrote in 
>> 
>> On Jan 14, 2021, at 12:56 PM, Ashutosh Bapat 
>> <ashutosh.bapat....@gmail.com<mailto:ashutosh.bapat....@gmail.com>> wrote:
>> 
>> Hi Japin,
>> Thanks for the report.
>> 
>> I think that comment is correct. It refers to the following code
>> blocks of XLogSendPhysical()
>> 
>> 2744     /*
>> 2745      * OK to read and send the slice.
>> 2746      */
>> 2747     resetStringInfo(&output_message);
>> 2748     pq_sendbyte(&output_message, 'w');
>> 2749
>> 2750     pq_sendint64(&output_message, startptr);    /* dataStart */
>> 2751     pq_sendint64(&output_message, SendRqstPtr); /* walEnd */
>> 2752     pq_sendint64(&output_message, 0);   /* sendtime, filled in last */
>> 
>> 2803      * Fill the send timestamp last, so that it is taken as late
>> as possible.
>> 2804      */
>> 2805     resetStringInfo(&tmpbuf);
>> 2806     pq_sendint64(&tmpbuf, GetCurrentTimestamp());
>> 2807     memcpy(&output_message.data[1 + sizeof(int64) + sizeof(int64)],
>> 2808            tmpbuf.data, sizeof(int64));
>> 2809
>> 2810     pq_putmessage_noblock('d', output_message.data, output_message.len);
>> 
>> 
>> After a quick search, I found that WalSndPrepareWrite and WalSndWriteData 
>> are always pairs [1].
>> IIUC the space of sendtime leave by WalSndPrepareWrite, it always fill out 
>> by WalSndWriteData.
>> 
>> 
>> WalSndWriteData() also fills the timestamp there but it may not always
>> be used with WalSndPrepareWrite, at least theoretically. So it's the
>> XLogSendPhysical() that it's referring to.
>
> The two functions are the body of two logical-decoding API
> functions. They are assumed to be called in that order. See
> OutputPluginWrite() for the restriction. The sequence of the two
> logica-decoding funcitons and the code block in XLogSendPhysical are
> parallels in (theoretically) different protocols.
>

Is that mean the sendtime of WalSndPrepareWrite always fill out by 
WalSndWriteData?
If it is, I think we should modify the comment in WalSndPrepareWrite.

-- 
Regrads,
Japin Li.
ChengDu WenWu Information Technology Co.,Ltd.


Reply via email to