Re: Warm Standby question

On Feb 1, 2009, at 4:47 AM, Thomas Kellerer wrote:

> Hi,
>
> (Note: I have never used log shipping before, I'm just interested in
> the concepts, so I'm might be missing a very important aspect)
>
> I was reading the blog entry about HA and warm standby:
> http://scale-out-blog.blogspot.com/2009/02/simple-ha-with-postgresql-point-in-time.html
>
> The image that explained how log shipping works, strikes me as being
> a bit too complex.
> <http://1.bp.blogspot.com/_26KnjtB2MFo/SYVDrEr1HXI/AAAAAAAAAEY/ncq_AW-Vv-w/s1600-h/pg_warm_standby.png
> >
>
> According to the picture it basically works like this:
>
> Master -> Copy master archive directory -> Copy to standby archive
> dir -> copy to pg_xlogs.
>
> When I look at this chain I'm asking myself, why do I need the two
> archive directories?
>
> Why can't the master copy the files directly into the pg_xlogs
> directory of the standby server?

Well, on the master the implementation has to be resilient to issues
with the archive_command and Postgres re-uses files in the pg_xlog
directory.  If they weren't copied into the archive_status directory
then if a file couldn't be shipped for some reason you'd run the risk
of the file getting re-used before it was successfully pushed to the
standby.

As to where they go on the standby, remember that log shipping is not
just for warm standby implementation -- it's also a valid backup
method, i.e. archive the transaction logs along with a snapshot of the
data directory.  In addition, the copy of a file into the pg_xlog
directory needs to be atomic and there's no guarantee that any given
archive_command will use a tool that does atomic copies.

Erik Jones, Database Administrator
Engine Yard
Support, Scalability, Reliability
866.518.9273 x 260
Location: US/Pacific
IRC: mage2k