Re: wal shipping/pg_standby docs patch

On Wed, Mar 25, 2009 at 9:04 PM, Tom Lane <tgl(at)sss(dot)pgh(dot)pa(dot)us> wrote:
> Frank Wiles <frank(at)revsys(dot)com> writes:
>> Attached is a patch that makes the suggestion be 'rsync -avr' in both
>> backup.sgml and pg_standby.sgml.
>
> Seems to me the first question to ask is exactly why you got "bit" by
> cp.  I doubt that rsync will provide a magic answer.
>
>                        regards, tom lane

Hi Tom,

Using the example in the docs in both places suggests using cp to copy
to your archive location, which isn't atomic. This leaves open the
possibility of transferring to your warm standby server part of a WAL
file. In my case I was using rsync in cron to pull from the "master"
onto the warm standby server. Ran fine for 3 days, then I transferred
a 14MB and a 8.2MB WAL file. Basically being unlucky and transferring
it at just the wrong time.

This obviously hosed the warm standby and I had to go through the
whole backup procedure again. Using rsync, as it is atomic, prevents
this.

--
Frank Wiles
Revolution Systems | http://www.revsys.com/
frank(at)revsys(dot)com | (800) 647-6298